It can be a very manual process, so I built two AI-based tools (linked below) to do most of the legwork. These tools:

• Evaluate your URL and your competitors for a specific keyword phrase

• Generate and classify common fan-out topics

• Identify gaps in your current content

• Recommend specific content to cover common fan-out topics

• Determine if it’s better to create new pages or update existing pages

Last week, I used the output from these tools on one of our pages on Zyppy. The tools quickly identified common fan-out topics and gaps in our coverage, while prioritizing specific on-page recommendations to fill the intent gaps.

While it’s still early days (actual ranking changes can take weeks/months), it seems overnight we saw an explosion of AI citations in Google.

Most excitingly, Google results now recommend Zyppy’s title tag checker in the AI answer itself. So not only are we cited, but we became part of the answer. Excited to see where this process might take us.

This helps validate that our Fan-out Framework can produce results and that our AI-based tool recommendations helped.

The same increase in AI citations is visible in Bing.

In theory, these on-page improvements could also help us rank better in traditional search results.

One tool lives in ChatGPT, the other in Claude. Both require an account with one of these services. You can use the ChatGPT tool with a free account, though you’ll get better results with a paid account.

Here are the two tools below:

Read more

Subscribe now

Traditional SEO starts with keywords.

The idea was simple: you ranked for a keyword people searched for, and you received traffic from Google.

But AI answers are more complex.

AI engines generate answers from multiple sources, including LLM training data, top-ranking search results, and fan-out query results. Fan-out queries are specific sub-queries that AI engines create in the background. (iPullRank offers an excellent, detailed explainer.)

It then synthesizes all that information before formulating a final answer.

The more places you show up in the AI’s research, the better your chances are of being included in the final answer.

So how do you optimize for this?

Ranking in Google (and other search engines) for the primary keyword can help you get to the AI answer, but it is often not enough. Ranking—or being mentioned—within the additional fan-out queries is one way to greatly improve your chances.

As an SEO consultant, I see many companies and agencies using frameworks to optimize for fan-out queries to gain more visibility in SEO and AI, but I haven’t seen a good guide on exactly how to do it.

So I wrote this one. The process is super-simple.

1. Use a keyword you already rank for

2. Find common fan-out queries

3. Determine the most important fan-out topics

4. Optimize your page—or create new pages—for the fan-out queries

5. Measure the results

At this point, experienced SEO will recognize this looks a lot like “optimizing for related queries.” That’s because it is very similar, with a few critical tweaks.

1. Find a keyword topic you already rank for

Here’s something every marketer should know: the #1 predictor for appearing in AI answers is ranking highly in regular search results.

• A study from Ahrefs showed that 38% of Google AI Overview citations come from Google’s top 10 ranking pages

• AirOps found that ChatGPT cited page ranking #1 in Google 43.2% of the time

• Data from Semrush showed AI answers from Perplexity had an 82% overlap with Google’s top 10 results

This means that if you already rank in the top 10 of Google results, or even the top 20, you have a good shot of appearing in or being cited in the AI answer.

Additionally, if you rank for multiple fan-out queries as well, your chances of appearing in AI may be stronger still.

How do we find keyword topics we already rank for? The best place is typically your Google Search Console account.

Here’s a real example from one of my own websites. My page about title tags ranks on page 1 of Google for a bunch of “title tag length” queries, but it doesn’t get many clicks - likely because people are getting their answers from AI Overviews.

For this example, “seo title length” and related queries are what we’ll try to optimize for.

2. Find Fan-out Queries

Before we determine which fan-out queries to optimize for, we need to address an important point: the fan-out myth.

The fan-out myth is that for any given search, there is a stable, pre-determined list of fan-out queries for that search.

Nothing is further from the truth.

Fan-out queries, like AI, are probabilistic and personalized, and they vary greatly across AI models and even within user sessions.

This means you can perform the same search 10 different times using the same AI engine and get 10 different sets of fan-out queries. The differences become even greater when using different AI platforms.

But that’s okay. Fan-out queries will never be entirely stable, but they can be broadly similar.

The key is to identify the commonalities and optimize around those.

Our job now is to gather as many fan-out queries as possible. Keep in mind, there are dozens of tools and techniques for finding fan-outs, and they are all different. There is no “best” way to find fan-outs, but here’s a process I personally like to use.

Use QueryFan

In my opinion, QueryFan is one of the best all-around query-fan out generators.

Queryfan works best with a paid OpenAI or Gemini API key.

If you’ve never used an API before, it’s really easy to generate a key, and you don’t need any special skills. Here’s a video on generating an OpenAI API key

Here’s how to generate an API key with Google’s Gemini.

You can set up a Gemini API key for free (OpenAI requires a paid subscription), but in my experience, it’s better to add a billing method. Retrieving query fan-outs has never cost me more than a buck or two.

For richer results, you can also add an AlsoAsked API key, which requires a Pro plan. You can sign up for an AlsoAsked plan here.

QueryFan lets you define “personas” that simulate real users’ profiles. If you want broad, generic fan-outs, feel free to skip this step.

When the tool finishes, export all these results into a spreadsheet.

Use Qforia

For Google specifically, Qforia is another great tool for generating query fan-outs. (It also requires a Gemini API key)

Qforia lets you upload multiple search queries at once and switch between AI Overview and AI Mode.

You’ll also want to export these results to your spreadsheet.

Non-API Options: If you don’t want to deal with API keys, Dejan’s queryfanout.ai and Otterly’s Query Fan Out Analysis tool are great options.

Use Bing

Bing Webmaster Tools shows you all Grounding Queries in its AI Performance Report.

Technically, grounding queries are different from fan-out queries. In fact, they are opposite, but related. Fan-out queries are used by the AI to seek out new information, while grounding queries are used to verify and fact-check existing information.

Practically speaking, there’s a good amount of overlap between grounding and fan-out queries, so it’s likely a good idea to include these in our analysis.

Here are all the grounding queries for our title tag page.

We’ll add these to the spreadsheet.

Creating Synthetic Fan-out Queries With AI

You actually don’t need any special tool to get fan-out queries. You can simply ask AI to do it.

But you need to give it the right prompt.

Many fan-out tools work by creating synthetic fan-outs using AI to simulate what the fan-outs could be. When fine-tuned, this can work as well as recording actual fan-outs.

In fact, you can do this yourself using nearly any AI engine. Try the following simple prompt. (You can modify this as you wish.)

You are an expert SEO strategist and search intent analyst. Generate fan-out queries for this seed search phrase:

[INSERT SEARCH PHRASE]

A fan-out query is a realistic search query that helps satisfy the user’s underlying information need behind the seed phrase. It may restate, refine, define, compare, troubleshoot, or apply the topic to a specific platform, audience, tool, or use case.

Return 50 total queries grouped under these headings:
Primary intent queries
Supporting subtopic queries
Comparison queries
Problem/solution queries
Audience or use-case queries
Decision-stage queries

Rules:
- Stay tightly connected to the seed phrase, its intent, and the subtopics needed to answer it well.
- Include exact-intent variants, follow-up questions, definitions, attributes, examples, comparisons, problems, fixes, tools, platforms, and use cases when relevant.
- Use natural search-query wording.
- Include short-head, mid-tail, and long-tail queries.
- Avoid broad, vague, off-topic, redundant, awkward, or unlikely queries.
- Do not force commercial, tool, platform, or “best” queries unless they naturally fit.
- Put each query on its own line.
- Do not number queries.
- Do not include commentary.

Silently remove anything that is not useful for SEO content planning.

Run this, and we’ll add these to our spreadsheet.

3. Determine Your Most Important Fan-out Topics

At this point, you’ll likely have a lot of fan-out queries in your spreadsheet.

Too many, in fact.

Our “seo title length” spreadsheet now has nearly 400 possible fan-outs!

You’ll likely find many fan-outs that are slightly irrelevant, off-topic, repetitive, or so specific that no one would ever optimize for them.

We need to clean up this data.

You could filter your fan-out list by hand, but it’s much faster and easier (and typically more accurate) to use AI to do it for you.

Here’s a sample AI prompt you can use to clean up your list.

You are an expert SEO strategist and keyword researcher. I will give you a primary search phrase and a raw list of fan-out queries.Your job is to clean, filter, and consolidate the list into a focused keyword set for SEO content planning.

Primary search phrase:
[INSERT SEARCH PHRASE]

Raw fan-out queries:
[PASTE QUERY LIST]

Instructions:
1. Identify the core search intent behind the primary phrase.
2. Remove queries that are obviously not useful for ranking a single strong page for that intent. Remove:
- Different meanings of the main words
- Broad or generic topic drift
- Loosely related subtopics
- Queries better suited to a separate article
- Unrelated tools, platforms, trends, news, jobs, salaries, rules, or frameworks
- Duplicates, awkward phrasing, or unlikely searches
Do not remove adjacent queries if they directly support the primary topic’s specific angle.
3. Keep queries that naturally belong on the same page and help explain, expand, compare, refine, or satisfy the core intent, including:
- Core synonyms
- Beginner questions
- What/how/why/best/vs/for variations
- Problem and solution queries
- Comparison queries
- Attribute, feature, tool, process, or platform-specific queries, when relevant
4. Consolidate duplicates and near-duplicates.
Choose the clearest, most natural search phrase for each group.
Do not list multiple queries that mean essentially the same thing unless they represent meaningfully different search behavior.
5. Create:
- 20–25 primary keywords worth targeting directly
- 5–10 optional secondary queries to include naturally
Primary keywords should cover the strongest recurring intent clusters, not just close synonyms of the head term.
Optional secondary queries should be useful supporting concepts, not a dumping ground. Include them only when they add topical coverage but are too narrow, tool-specific, platform-specific, use-case-specific, or low-priority to be direct ranking targets.
Be strict. Do not include weak queries just to reach a count.

Output format:
Obvious off-topic buckets:
- [Bucket]: [brief examples]

Consolidated target query set:
1. [keyword]
2. [keyword]

Optional secondary queries:
1. [query]
2. [query]

Do not include a page structure or extra commentary.

Using this process, we’ve narrowed down our list of fan-out queries to a more reasonable list:

1. seo title length
2. title tag length
…
24. HTML title tag vs SERP displayed title
25. title tag length examples

We also have several related fan-out topics:

1. Google Search Central documentation on page titles
2. SEO title length for WordPress
3. SEO title length for Shopify product pages
… etc.

At this point, you could simply review your fan-out queries and identify topics that you don’t rank for or don’t cover well, and optimize around those.

But I’m going to add a couple of steps to make our jobs easier. These next steps are completely optional.

Get Search Volume for Query Fan-out Topics (optional)

You can use your preferred SEO tool to get search volume for your query fan-out topics.

Many, if not most, of your queries may have no search volume at all. That’s okay. If any do have existing keyword volume, this will help us prioritize those keywords.

Here, I’ve popped our query fan-out list into Ahrefs Keyword Explorer.

It looks like many of our fan-out topics do have significant search volume! We’ll want to pay extra attention to these.

Next, I use keyword clustering to better identify ranking opportunities and gaps in my content.

Keyword clustering is a technique that groups keywords together by search intent. It works by analyzing Google search results. If two different Google searches return roughly the same set of results, those keywords likely have the same intent and can be targeted by the same piece of content.

My favorite Keyword Clustering tool is Keyword Insights.

After giving Keyword Insights my fan-out queries and search data, it can perform various calculations.

Here, I can see how zyppy.com compares with competitors in terms of cluster visibility.

Not too shabby! But looks like there is room for improvement.

Looking at the cluster data itself, we can tell a couple of things about our project:

1. Our fan-out queries cover many distinct search intents

2. Despite this, our page already ranks really well for most terms (nice!)

3. We do have a number of gaps in our content coverage, so we can patch those up

In this case, we’re missing a major keyword gap around “title length checker” and “serp snippet preview tool.” (We also see some optimization opportunities around “Google search central title link documentation” and “title tag vs. H1 tag length,” though these don’t seem quite as important.)

These might be good targets to optimize for, and they could help us earn both AI citations and some additional organic search traffic.

👋 Subscribe free (it helps me produce research like this) and get tips + tactics on how to win more SEO and AI visibility.

Subscribe now

4. Optimize for the Fan-out Queries

Now we have a couple of specific query topics we’d like to optimize for: title length checker and serp snippet preview tool.

When optimizing for fan-out queries, we have a couple of options:

1. Create new pages covering the topic

2. Update our existing pages

Both approaches have their pros and cons. In this case, since the page is already ranking well and has strong authority, we’re going to update it with new information. This is also the easier route.

We’re not going to get into how to optimize a page to target a specific keyword (there’s lots of information out there already), but keep this in mind when optimizing specifically for query fan-outs:

Align your content to the fan-out query/answer.

This means using the query, or a close variant, in your headings (h1, h2, etc.) and following it immediately with the answer, however long that content is.

You really want to create content that’s worthy of ranking in the top 10-20 of Google’s search results. Otherwise, the AI engine may not see it.

Here, I tested several title tag length checkers and added a new write-up to our article.

Optimization Tips

1. Don’t blindly add new pages covering every fan-out topic. This is not the way you want to go. Remember that AI engines prefer content that ranks highly, and lots of low-quality content isn’t going to help. In fact, in a time when Google is more and more demoting scaled content, this could definitely hurt you.

2. Don’t need to add every fan-out topic to your page. You don’t want to creat Godzilla pages. Keep your topics tightly focused, while working to completely satisfy the user.

3. Don’t add generic, non-commodity content. The same rules of content apply to fan-out coverage as they do to any type of content. Make it original! Add first-party information or personal perscpectives when you can. Keep generic content out.

5. Measure The Results

The good news is that you can often influence AI answers much faster than it typically takes to influence traditional search results.

First of all, while it’s not the most reliable method, you can simply check Google to see if you appear or are cited in AI answers.

Be careful, though! Individual search results can be highly variable and personalized, so for more accurate tracking, you want a larger dataset.

The easiest and least costly way to check your AI visibility is through Google and Bing Webmaster Tools.

Bing Webmaster Tools AI Performance Report

https://www.bing.com/webmasters/aiperformance

Bing is ahead of Google in AI reporting. You can see how many citations each URL earned in AI answers, and also the grounding queries that led to those citations.

Here’s the actual report for our title tag page. On the right side of the chart, you can see a rise in citations after I made some improvements on the page.

Google Webmaster Generative AI Features

https://search.google.com/search-console/performance/search-analytics/ai

Google recently added an AI features report that shows how many times each URL appeared in Google’s AI Mode or AI Overviews.

Somewhat disappointingly, Google’s report does not tell you the grounding or fan-out queries that triggered these impressions.

Here’s what it looks like.

Premium Options

There are many professional AI trackers on the market, including Peek, Otterly, Profound, and Gumshoe.

Here’s what our fan-out queries look like in Ahrefs Brand Radar.

You can also estimate how many clicks you receive from LLMs using SEOTesting.

I’m not advocating for one AI tracking platform over another, or even whether you should have dedicated AI prompt tracking. That said, if you’re a large brand or AI is an important part of your search and discovery strategy, prompt tracking is often worth it.

Pro Bonus: AI Query Gap Analyzer Tools

The Fan-out Framework can be a very manual process, so I built two AI-based tools to do most of the legwork. These tools:

• Evaluate your URL and your competitors for a specific keyword phrase

• Generate and classify common fan-out topics

• Identify gaps in your current content

• Recommend specific content to cover common fan-out topics

• Determine if it’s better to create new pages or update existing pages

One tool lives in ChatGPT, the other in Claude. Both require an account with one of these services. You can use the ChatGPT tool with a free account, though you’ll get better results with a paid account.

These tools are available for paid subscribers in Pro Templates.

1. ChatGPT: Zyppy Signal Query Gap Analyzer

Give the Query Gap Analyzer a primary keyword and a URL, and it looks at:

• The core intent behind the keyword

• The most useful fan-out queries related to that intent

• Which query clusters does the page already cover well

• Which clusters are only partially covered

• Which important topics are missing entirely

• Whether those missing topics belong on the same page or should become separate content

The report includes a fan-out query set, a coverage scorecard, a list of missing or weak clusters, priority opportunities, example section ideas, and a roadmap for what to improve first.

It will also generate a complete PDF report - be sure to confirm this at the end!

2. Claude: Query Gap Analyzer

If you prefer Claude, this is very close in functionality to the ChatGPT tool, with a few differences.

Enter a keyword and URL. Claude doesn’t fetch pages as easily as ChatGPT does, so you may need to paste your page content manually.

• It finds the fan-out queries that search engines may expect the page to answer.

• It shows what the page covers, what’s weak, and what’s missing.

• It recommends when to add to the page vs. when a topic deserves its own page.

• It gives examples H2 heading and notes.

The goal: help your page better satisfy a broader set of fan-out queries around a keyword.

Frequently Asked Questions

Isn’t this just optimizing for related queries, with a different name?

Most things in AI optimization have echoes in traditional SEO. So the answer is always “kinda, sorta, with a twist.”

So while they are similar, they are not exactly the same. For example, AI answers tend to be far richer and include much more specific information, such as comparisons, problem/solution queries, and decision funnels. We’re also far less concerned with keyword volume compared to traditional SEO. Finally, we also use different surfaces for research and measurement.

They are certainly similar, but subtly different.

Is this process about appearing in the AI answer, or being cited by it?

Great question! Most of this process is about getting cited by the AI, but in reality, there’s a lot of overlap here with appearing in the AI answer as well. Different AI engines are constantly updating how they answer and format questions, too.

Getting recommended by AI is often more about what others say about you, but what you say about yourself counts too. In the end, the more places you appear in the surfaces AI uses as sources, the better your chance of appearing.

How can you influence AI answers faster than traditional search results?

We love this question! For one, they use different algorithms. Google uses FastSearch/RankEmbedBERT to populate AI Overviews, which use 70 days of user data rather than the 13 months used for traditional search results.

AI answers can also dig deeper into search results, so a recent update you made to a URL might influence an AI answer more quickly than trying to get that content to rank near the top of traditional results.

Is there a fourth frequently asked question about fan-out queries?

No. There are only three.

How do you optimize for AI fan-out queries? Join the conversation on LinkedIn.

👋 Subscribe free (it helps me produce research like this) and get tips + tactics on how to win more SEO and AI visibility.

Subscribe now

This represents the next step: when we understand which features are associated with AI citations, how do we put this into practice?

Earning AI citations is mostly good SEO, with a few extra steps. Based on our database of case studies, experiments, and other evidence, there are a number of steps you can take to increase your chances of winning AI citations. You want to:

1. Make your content accessible to AI engines

2. Rank for the main query

3. Find and target fan-out queries

4. Choose the right content format

5. Match the answer to the query

6. Put your primary answers at the top

7. Make the page easy to cite

Important: You should understand that AI citations are probabilistic and aren’t as stable as traditional search results. AI citations change often, even day to day. The goal is not to permanently “win” AI citations, but to increase your chances of appearing over time.

For review, here are the AI Citation Ranking Factors.

Step 1. Make your content accessible to AI engines

Making sure AI engines can access and read your content is more about what you don’t want to do:

A. Don’t block important crawlers in your robots.txt file. For AI citations, the most important user agents are:

• Googlebot

• bingbot

• OAI-SearchBot

• ChatGPT-User

• PerplexityBot

• Perplexity-User

If you use a platform like Cloudflare, you can review your AI Crawl Control to check which bots you may be blocking.

B. Don’t use tags that prevent search engines from displaying important content

The following meta tags can all prevent search engines such as Google and Bing from displaying and otherwise using your content in AI answers and citations:

• <meta name="robots" content="noindex">

• <meta name="robots" content="nosnippet">

• <meta name="robots" content="max-snippet:0">

At the text level, the data-nosnippet attribute can also prevent content from appearing in AI answers.

• <span data-nosnippet>Important text here</span>

C. Don’t rely on client-side JavaScript to render important text

LLMs like ChatGPT generally digest pure HTML, and even Google is known to encounter problems with JavaScript from time to time.

If your site relies heavily on JavaScript, use server-side rendering or another solution to deliver rendered HTML to the browser.

Use this LLM Content Visibility Scanner by Lily Ray to see if LLMs can access your content.

D. Don’t hide important text

We’ve known for a long time that text hidden behind tabs, carousels, modals, or other interactive elements tends not to rank as well as clearly visible text.

AI engines appear to work the same way.

If you have text hidden by tabs (CSS or JS), be sure to bring it into the open so humans, search engines, and LLMs can see it.

Step 2. Rank for the main query

To win AI citations, you want to aim for queries for which you already rank organically.

AI citations source their candidate URLs from top-ranking search results, so you want to be included in those results. You might even call it a requirement.

Do you need to rank #1? Not necessarily.

Although ranking higher helps, you probably need to be in the top 30 at a minimum to have a decent shot at consideration.

And even if you don’t rank at all, you can still appear as an AI citation if you rank for related searches, aka “fan-out” queries. (We’ll cover these next.)

Regardless, select the query that triggers a Google AI Overview or AI answer with citations. It could be something like:

• “Title Tag Length”

My website currently ranks around position 6 for this in Google search, with 20K impressions, so it should be a good query to target.

Step 3. Find and target fan-out queries

Here’s where optimizing for AI citations gets fun, in part because it can also raise your traditional SEO rankings and visibility.

When AI engines form an answer, they often perform fan-out or grounding queries to obtain additional information or to verify what they already know.

Ranking highly for these additional queries increases your chances of getting chosen for AI citations.

How do you find fan-out queries?

There’s no single “definitive” way to discover fan-out queries, as they change constantly among AI systems based on probability, personalization, and context.

That said, you can get pretty close by using a combination of three methods:

A. Observed Fan-Outs

For ChatGPT, several tools capture the exact fan-out queries used, including Mark Williams-Cook’s ChatGPT Fan Out Query Explorer. Note that these queries vary frequently, so use them only as a rough guide.

Other useful tools include:

• Qforia by iPullRank

• AI Visibility Fan Out by WordLift

B. Synthetic Fan-Outs

Many tools will “simulate” fan-out queries for you, like this one from Rankability.

C. Reverse-Engineer AI Answers

Oftentimes, simply noting what the AI engine returns for an answer will tell you exactly the type of content you want to rank for.

For our “title tag length” example, I probably want to make sure my content covers Optimal Length, Pixels, Placement, Truncation, and Purpose.

Should you try to rank for related queries with new pages or existing pages?

The answer to this question depends entirely on search volume and whether you think a separate page would be useful to users.

The goal is not to create a ton of pages simply for AI engines to digest. That’s probably a recipe for disaster.

If you think a topic/page can stand on its own, create it. Otherwise, it’s often fine—and sometimes easier—to update existing pages with updated section headings and content that directly addresses additional fan-out queries.

Important: You don’t need to try to rank for every fan-out query. In fact, it would be foolish to do so.

Instead, try to rank for the gaps in your existing content, for queries people are actually searching for. Not only will this likely help with AI citations, but it will broaden your ranking base across more real-world search terms.

To find gaps in your content, try IntentGaps.com.

4. Choose the right content format

If a user searches for “best places in Germany to visit,” the best content type for the answer will be very different for a query like “Apple watch specs vs Garmin Venu”.

The first is probably best served by a ranked list, the other by a comparison page.

It could be a side effect of AI engines citing highly-ranked pages, but many studies have found a correlation between page format and AI citations.

The easiest way to determine the right content format for your query is to simply observe what’s currently ranking for your query/fan-out queries.

5. Match the Answer to the Query

This is one of the stronger findings in the research. In an AirOps/Kevin Indig study, pages whose headings closely matched the original query were cited more often than pages with less strong matches.

Similar findings are supported by several other studies, patents, and research papers that show a strong semantic match between your page content and the AI’s query/fan-out query.

In practice, this means writing clear titles, headings, and answers that closely match the original queries.

To target multiple queries and fan-out queries, use subheads with accompanying text blocks where appropriate.

To target multiple queries and fan-out queries, use subheads with accompanying text blocks where appropriate.

6. Put your primary answers at the top

Don’t hide the main answer of the page after a long, fluffy intro.

Several studies have found that standalone sentences early in your content are more likely to be picked up by AI.

This matters. According to Dan Petrovic, only about 32% of a page’s content gets picked up for consideration in the grounding process by Google Gemini.

In practice, this may look like a short summary near the top of your content. Here’s how I wrote about “what’s the ideal title tag length?”

For a page about title tag length, you might write:

This is unambiguous and direct, and its positioning means it’s more likely to be picked up by AI.

7. Make the page easy to cite

This one is a bit controversial, because many in the SEO world believe that you shouldn’t have to “write for LLMs.”

It’s not so much about writing for LLMs, it’s more about being clear and focused.

Dan Petrovic again, from his research:

“A tight 800-word page can get 50%+ of its content grounded. A 4,000-word page gets ~13%. Focus on making every sentence count rather than adding volume.”

In our Ranking Factors study, we found that three sentence structures were more likely to support AI citations.

A. Factually Specific: pages and passages that showcase specific, verifiable facts.

B. Explicit Phrasing: use specific claims over vague statements.

C. Site Sources: back up facts with referenced sources

D. Self-Contained Passages: important statements can stand alone without looking elsewhere in the text

E. Entity Consistency: use of consistent names for products, brands, people, etc.

Here’s an example of all these in action.

The latter is more factually specific, explicit in its phrasing, and cites the source of the information. It also uses consistent entity naming, and all the ideas are self-contained within the passage.

You don’t need to cite sources for every sentence, but you should cite sources for facts that need evidence, even if the source of evidence is yourself.

Does Winning AI Citations Matter?

Yes, at least in Google AI Overviews.

A study by Seer found that “Being cited in the AIO delivers +120% more organic clicks per impression versus when you are not cited.”

To win more AI citations, make your content accessible to AI crawlers, find the right search queries to target, discover and optimize for fan-out queries, use the right content format, put your most important content at the top, and use a clear sentence structure.

Happy citation winning!

👋 Subscribe free (it helps me produce research like this) and get the 7-Step AI Citation Checklist with tips + tactics on how to win more AI visibility. Already subscribed? Check your inbox.

Subscribe now

Everyone is talking about AI citations.

To make sure we’re on the same page, what do we mean by AI citations? While they vary by platform, AI citations are clickable links to sources that AI engines use to support their answers. Like these below.

While there’s no debate that AI answers reduce clicks to the open web, citations can act as a safety valve for publishers. A recent study by Seer Interactive shows that getting cited in Google’s AI overviews results in 120% more organic clicks per impression and a 41% increase in paid clicks compared with when your brand is not cited.

It seems every week we see a solid new AI citation study, framework, or even debate over what works.

But we observe that most marketers aren’t implementing these tactics in their workflows, either because they aren’t sure what’s important or because of information overload.

To address this, I downloaded nearly every published AI citation experiment, study, explainer, and patent published over the past couple of years. These covered various AI engines, including ChatGPT, Gemini, and Perplexity. I then narrowed it down to the 54 most salient and helpful. I categorized the findings from each and cross-referenced them to identify the most similar and often cited observations.

The AI Citation Ranking Factors you’ll see below are based on actual data from the published experiments and studies. The strength of each “factor” is based on:

1. Repeatability: How many times has each similar finding appeared across multiple studies? Also, the consistency of outcomes—positive or negative—across studies.

2. Strength of Evidence: E.g., a 50-million-query study carries more weight than a 10-query case study.

3. Official Support: Whether or not the factor is supported by official docs, technical specs, or patents.

After all this evidence was tabulated, I manually assigned a score based on these criteria, using AI to help fine-tune the numbers.

Before we go any further, I want to acknowledge and thank many of the authors, researchers, and thought leaders who made this research possible. In a world of fake “Top 10” gurus, these are a few (but not all) of the authentic AI search folks whom I rely on to learn more about the AI universe, and they deserve a follow:

Dan Petrovic • Kevin Indig • Metehan Yesilyurt • Oshen Davidson• Lily Ray• David McSweeney • Aleyda Solís • Britney Muller • Dawn Anderson • Micheal King • Andrea Volpini • Rand Fishkin • Ann Smarty • Cyrus Shepard (kidding! But if you want to follow me, there I am ;)

Below are the most evidence-based characteristics of content that earn AI citations. Be sure to read the explanation of each factor beneath the graph, and get your free copy of the AI Citation Checklist, which shows you practical steps to put these factors into practice.

TL;DR - While the evidence suggests that specific practices can boost AI citations, most of the critical factors align with traditional SEO practices. This is important because there’s much debate in the AI / SEO world about whether optimizing for AI visibility requires a different toolset.

In this case: win SEO, win AI citations (most of the time, with extra steps.)

AI Citation Ranking Factor Descriptions

1. URL Accessibility

Score: 9.5 | Definition: Page is available and crawlable during training/grounding

This is basic SEO applied to AI engines. A URL typically needs to be accessible and crawlable, either during training or grounding, for an AI engine to cite it.

That said, this has grown increasingly complex as AI companies add more user agents to crawl the web (OAI-SearchBot, GPTBot, Google-Extended, etc.) and companies like Cloudflare offer AI-scraper-blocking protections. The end result is that it’s easier than ever to exclude pages from AI systems and lower the probability of citations.

2. Search Rank

Score: 9.4 | Definition: How the URL ranks for the exact query

Many, many studies have found a clear relationship between ranking highly in “traditional” search and AI citations. Ahrefs found that 38% of AI Overviews citations come from the top 10 Google results. Go beyond the top 10, and the overlap increases.

ChatGPT is more complex because it doesn’t fully disclose its search sources, but AirOps found a strong relationship between “retrieval rank” and citations from ChatGPT.

3. Fan-out Rank

Score: 9.3 | Definition: How the URL ranks for related fan-out queries

Aside from ranking for the main query, AI engines will perform many “fan-out” searches to supplement and ground their answers. Lots of evidence shows a clear relationship between ranking highly for fan-out queries and earning AI citations.

4. Preview Control

Score: 9.2 | Definition: Preview controls, such as "nosnippet," can influence visibility

More specifically for Google and Bing, preview controls are directives that allow publishers to control how much of a page search engines can show in snippets and on some AI surfaces. Examples include “nosnippet” and “data-nosnippet”. Limiting the visibility of specific text can lower AI visibility.

5. Query-Answer Match

Score: 9.2 | Definition: The page content closely matches the query (primary/fan-out)

Many studies have documented a “semantic closeness” between the AI answer and the cited content. This often means that page titles, subheads, and content closely match both the search query and the AI answer.

6. Intent-Format Match

Score: 9 | Definition: The page type matches the intent of the query, e.g., listicles for "best" queries

AI engines tend to cite articles whose content format is best suited to the query. For example, “best”-type questions (“best probiotics for men”) might prefer a listicle or comparison table, whereas “how-to” queries (“how to build a bird house”) are more likely to reward a step-by-step guide.

This may simply be an artifact of the Search Rank/Fan-out Rank

7. Topic Cluster Ranking

Score: 8.9 | Definition: The degree that the site ranks for multiple queries (primary + fan-out)

This is an interesting one, but easy to understand. The concept is that ranking for multiple related queries increases your chances of being cited at least once. The RRF Top-n Playbook, while technically challenging, was one of my favorite experimental explainers.

8. Answer Near the Top

Score: 8.8 | Definition: Important content placed near the top of the page is more likely to be cited

One thing that becomes immediately clear when reading all of these papers is that AI engines don’t treat all the text on the page equally. Dan Petrovic showed how Google’s Gemini uses a strict retrieval cap per URL, and content near the top of the page is more likely to make the cut. Several other studies support this as well.

9. AI-ready Structure

Score: 8.6 | Definition: Content is formatted so that AI can easily extract and understand information

Following the idea that AI engines typically don’t retrieve your entire page, consider that they break pages into sections before retrieval. If your content isn’t clearly organized, this can make the job more complex.

This doesn’t mean you need to “chunk” your content into bite-sized pieces; simply provide a clear structure with headings, sections, tables, etc. In fact, many studies found a clear relationship between these types of features and AI citations.

10. Factually Specific

Score: 8.3 | Definition: Pages and passages that showcase specific, verifiable facts

Because AI citations are included to support specific claims in the AI answer itself, it seems to help if you mirror that with specific facts that the AI can reference. Statements like “Adults need a lot of protein” aren’t as specifically citable as “Experts recommend 0.8 grams of protein per kilogram of body weight.”

11. Explicit Phrasing

Score: 8.1 | Definition: Specific claims over vague statements

The cousin to “Factually Specificity”, AI engines appear to prefer phrases that are more definitive, without hedging. For example, “Some people prefer magnesium glycinate, while others use citrate or threonate...” is a far weaker phrase than “Magnesium glycinate is the best choice for sleep.”

Subscribe now

12. Cites Sources

Score: 8 | Definition: Facts are backed up with referenced sources

Several studies found that facts with clearly cited sources are associated with more frequent AI citations. This makes sense as AI engines are working to generate answers and citations they can justify.

Practically speaking, this doesn’t mean you have to add citations to all of your content, but it would be prudent to “show your work” and demonstrate how you arrived at important conclusions.

13. Self-Contained Passages

Score: 8 | Definition: Important statements can stand alone without additional context

The idea of “Self-Contained Passages” simply means you're expressing key facts or points fully within sentences or blocks of text.

For example, if you say “This ingredient has better evidence,” the AI engine is forced to parse your meaning from other areas of text. What ingredient? What evidence? But if you say “Magnesium glycinate is supported by 137 scientific studies for heart health,” the information is unambiguous and self-contained.

14. Content Visibility

Score: 7.6 | Definition: Important information is in visible HTML text, not hidden

Modern web pages can contain a lot of text that’s not immediately visible, at least not without a lot of JavaScript or by requiring users to click on divs and tabs. We’ve known for a long time that even Google doesn’t seem to rank content as well when it’s not clearly visible on the page, and AI engines appear to share this bias.

15. Freshness

Score: 7 | Definition: How current the information is

Freshness is a known SEO ranking factor, and several studies have found a correlation between document freshness and AI citations. Like traditional search, freshness seems to vary by query. A question about a recent sports match will require more up-to-date information than a question about British history.

16. Brand / Entity Trust

Score: 6.8 | Definition: How much the AI engine knows about and trusts the brand/website

Increasingly, it appears AI engines seek out more credible sources of information. This means that what they already know about you may influence their level of trust or if they seek you out. For a health-related query, an AI engine is more likely to trust the Mayo Clinic than an anonymous health blog. Google works this way too, so there’s probably some overlap.

17. Length

Score: 6.7 | Definition: The length, in word count, of the content

Many, many studies looked at how content length correlated with AI citations. While the majority found that longer content tended to perform better, the evidence was inconsistent. Several researchers pointed out that longer content also reduced the likelihood that AI engines would retrieve all of your content.

18. Language

Score: 6.3 | Definition: The language of the content

The studies showed a clear bias towards the language—and sometimes location—of the question. A question asked in French by someone in France is more likely to use French citations.

19. Entity Consistency

Score: 5.8 | Definition: The use of consistent names for products, brands, people, etc.

Entity consistency simply refers to using consistent naming conventions for brands, people, products, etc.

For example, I could write “Zyppy makes SEO software which helps marketers rank.” That’s a much clearer sentence than “Zyppy makes SEO software. My software helps marketers rank.” The first version is clearer for both search engines and users.

20. Structured Data

Score: 5.6 | Definition: Page contains schema to identify entities and support content

Oh boy. There are few debates more heated in the SEO/AI space than the use of structured data for AI optimization. While it’s true that LLMs don’t typically ingest schema as part of their training, there’s limited evidence that they can at least see schema when searching (though they don’t use it in a schema-like way).

Regardless, we’re not here to debate that.

We’re here because practically every study that looks at schema and AI citations finds a positive relationship. The effect is typically small, but it’s amazingly consistent across studies.

21. Known Source

Score: 5.4 | Definition: The URL is already known to the AI engine via its training data

Sometimes (quite often, in fact) an AI will cite a URL simply because it knows about it from its existing training data. More typical of ChatGPT and Perplexity, this can bypass the usual grounding/search phase, leading to citations that no longer exist.

22. Domain Authority

Score: 5 | Definition: A link-based measure of website popularity

Several studies looked at the relationship between Domain Authority and AI citations. While many found a relationship, it was often weak.

23. LLMs.txt

Score: 2 | Definition: The website hosts an LLMs.txt file for AI engines

To be fair, I’m not certain many of these studies even considered the influence of LLMs.txt files. That said, we’re unable to find any credible evidence or experiments showing LLMs.txt files influence AI citations in any way.

The Steps To Earn More AI Citations

Ultimately, you don’t need an entirely new SEO playbook to earn AI citations, but you may want to tweak a few practices and execute a few other SEO strategies even better.

The overlap between traditional SEO and AI citation signals is significant. Additionally, many of these factors, when properly implemented, can improve the experience of human users on your page, which should be priority #1.

If we had to sum them up, they might be: Relevance, Trust, Topical Authority, & Extractability - all signals that should align with current SEO thinking. A few of the technical details shift, but we can still focus on creating great experiences.

Happy citation earning!

Methodology - How I Scored These Factors

I collected nearly every AI citation study, experiment, explainer, and patent from the past two years, covering ChatGPT, Gemini, and Perplexity, then narrowed the set to the 54 most useful sources.

I categorized the findings and cross-referenced them to surface the results that occurred most often across all studies.

Each factor was then scored on three criteria:

1. Repeatability: how often a similar finding appeared across different studies

2. Strength of evidence: the quality and strength of the data. For example, a 50-million-query study counts for more than a 10-query case study.

3. Official support: whether official documentation, technical specs, or patents support the evidence.

I assigned each score by hand against those criteria, using AI to help adjust the final numbers. The scores run from 9.5 to 2.

Each score reflects how consistently a factor correlates with AI citations across the source studies. These are not confirmed ranking factors. Correlation is not causation. Any errors are mine.

About Cyrus Shepard

Hi, I’m Cyrus. I started optimizing websites for Google traffic way back in 2009. Here at Zyppy, I run dozens of SEO and AI experiments per year, so I’m comfortable with our data. My experience allows me to craft successful SEO strategies for startups to Fortune 500 companies. Contact me or follow me on LinkedIn.

This represents the next step: when we understand which features are associated with AI citations, how do we put this into practice?

Earning AI citations is mostly good SEO, with a few extra steps. Based on our database of case studies, experiments, and other evidence, there are a number of steps you can take to increase your chances of winning AI citations. You want to:

1. Make your content accessible to AI engines

2. Rank for the main query

3. Find and target related queries

4. Choose the right content format

5. Match the answer to the query

6. Put your primary answers at the top

7. Make the page easy to cite

Important: You should understand that AI citations are probabilistic and aren’t as stable as traditional search results. AI citations change often, even day to day. The goal is not to permanently “win” AI citations, but to increase your chances of appearing over time.

For review, here are the AI Citation Ranking Factors.

Read more

Subscribe now

For many websites, Google Zero is already here.

For others, it’s coming soon.

Google Zero, a term coined by Nilay Patel, refers to a time when Google Search stops sending traffic to third-party websites.

Google’s Zero-Click era has been creeping up on us for a while now. The latest data from Datas/Sparktoro shows nearly ~60% of Google searches don't result in clicks to third-party websites. With Google’s shifting algorithms and AI Mode looming on the horizon, entire publishing industries are being decimated.

When AI can answer any question instantly, what does that mean for the rest of us? Is there anything web publishers can do?

Here’s the truth: traffic isn’t completely going away, and it likely never will.

Google still sends billions of visitors to websites each day: visitors looking for specific websites, visitors looking to buy things, visitors looking to connect with other humans, and visitors looking to solve a million unique problems.

At the same time, fewer and fewer of those visitors are looking for generic information.

The days when businesses simply “published a blog” trying to rank with only “content and links” are behind us.

We know for a fact that Google favors different types of sites now. Winning content tends to be proprietary, unique, and solves problems beyond providing information.

Looking at 100s of winning and losing websites, parsing Google guidelines, and reviewing Google statements about the future of AI search, we cataloged 17 content types that currently perform decently-to-excellent in the AI era when properly executed, and in our judgment, are most likely to survive the next few years and beyond.

• Many sites using these content types are positively thriving today.

• Others will most likely be stable for the next few years, given the moats they’ve built.

• Yet others will certainly see declines in Google traffic, but many may at least survive if they execute well.

This list of content types isn’t exhaustive. You’ll see other winning formats/strategies, but they generally share many of the same characteristics: winning content tends to be proprietary, experience-based, niche-focused, and allows task completion beyond merely providing information.

These 17 content types, in order to win, require those same characteristics.

These content types may look easy, but don’t be fooled into thinking that any of these are low-effort, at least when done well.

Before you hit publish—no, before you do any content strategy, consider where your content sits in one of these types.

Note: Please don’t consider these content types “get out of jail free” cards. When done with minimal or even above-average effort, you can still lose by producing a less-than-stellar output.

But they can help shape your content strategy in an increasingly zero-click world.

So instead, think of these content types as starting points for your most excellent, first-in-class efforts at content excellence.

Let’s explore each type. At the end, we've included a Pro Framework to adapt these content types to different business models.

1. Owned Audience

Very Strong | Highest Effort

“Owned Audience” is more of a content strategy than a format, but it’s also the gold standard. Owning your audience distribution channels means you don’t entirely rely on Google. Conversely, it often means more people actually search for you.

Most owned audiences these days operate via email, but they can also work via SMS, in-app notifications, or even snail mail catalogs.

The key, of course, is having content people want to subscribe to and pay attention to.

If you’re asking, “Why would anyone want to subscribe to our boring blog content?”, you should instead ask, “Why is my blog content boring in the first place? Who are we publishing it for?”

Examples: Morning Brew, Stratechery, Lenny's Newsletter

2. Transaction Pages

Very Strong | Highest Effort

Transaction pages are more than simple product pages (though these count, too). Transaction pages help users “complete” an action, such as booking a hotel reservation or subscribing to a service.

In the future, agentic commerce may see AI agents shop on our behalf, so users never need to visit a website. Even in that scenario, we still need transaction pages, and for now, they remain some of the strongest zero-click pages on the web.

Examples: Hipcamp, Chewy, Sweetwater Electric Guitars

3. Original Research

Strong | High Effort

Original research goes beyond publishing mere facts and data, as it must be unique and proprietary. This is research you originated, born of your own hard work.

It’s true, AI answers can cannibalize your original research, but they are much more likely to cite it, and users often want the original source.

Perhaps more importantly, original research is often cited by others—including journalists—and this can generate significant trust and authority signals for Google and LLMs alike.

Examples: ChartMogul Insights, Carta Data, Orbit Media Blogging Statistics

4. UGC Communities

Strong | High Effort

Forums and communities serve a dual benefit:

• Engaged members seek them out to connect with like-minded users, providing built-in Google Zero protection.

• They generate a lot of experience-based content, which Google currently favors.

Building a community is hard, hard work. But those that invest in their users have seen outsized rewards in the past few years.

Examples: Flyertalk Forum, Mumsnet Talk, Rick Steves Travel Forum

5. Creator Video & Podcast

Strong | High Effort

We live in a “Hot Ones” attention economy.

The point of Creator Content is not simply to pump out videos and podcasts on a topic. It’s about the host, brand, or format itself that makes the content unique.

My family used to listen to Car Talk on NPR, not because we cared about car repair, but to listen to the two cranky, funny hosts. When your creator content is so good that people search for you by name, you’ve made it.

Examples: Practical Engineering, @mkbhd, @Fireship

6. In-Depth Reviews & Tests

Strong | High Effort

We learned a lot from Google’s Helpful Content update, which saw many review websites lose significant traffic (though not so much for YouTubers). Google’s always been a little suspicious of commission-supported product reviews, but consumer demand is consistently high.

Winners in this category typically review products in great depth, with firsthand evidence, repeatable testing protocols, side-by-side comparisons, teardowns, and high-quality photos and videos.

AI answers can summarize these reviews, but they can’t replicate the firsthand evidence many users seek before buying

Examples: RTINGS: TVs on the Market, BabyGearLab Jogging Stroller, Tom’s Hardware CPU Benchmarks

7. Brand Pages

Strong | Low Effort

Most websites publish an “About” page, but in the age of AI, that’s hardly enough. While search engines, LLMs, and users can learn about you through third-party sources, what you say about yourself is often the most important.

Consider these different types of brand pages:

1. About: Company, Leadership/Author Bios

2. Trust: Editorial Policy, Legal, Values, Privacy

3. Social Proof: Press, Testimonials, Case Studies, Reviews

4. Contact and Support

These pages aren’t always sexy, and they don’t typically drive a ton of traffic, but they are absolutely foundational to establishing trust and defining your entity.

Examples: Buffer About, Bankrate Editorial Policy, Sleep Foundation Editorial Policy

9. Directories & Databases

Moderate | High Effort

We need to be careful here, because some people may hear “build a directory” and think they’ll rank easily.

Sadly, this is not the case.

As with everything else on our list, directories and datasets typically need to be supplemented with proprietary, first-party information. This is the AI moat. Freshness is another critical feature, as users typically want the latest information.

Examples: AlternativeTo, Clutch, Nomads

10. Expert Perspective

Moderate | Medium Effort

Moving from facts to opinion, audiences often seek out expert perspectives on important matters. Stripe’s Cheeky Pint interview with Google’s CEO, which we mentioned earlier, is a perfect example of this.

Expert perspectives can earn citations and boost your brand. They also work best when anchored with unique experience and insider information/data.

Examples: a16z News Content, Not Boring, Tom Tunguz

11. Templates

Moderate | Medium Effort

Reusable templates are frameworks that users can download or reproduce in order to get started, speed up work, or accomplish a specific use case. Consider Canva templates, which helped explode growth for the company.

Obviously, the harder the templates are to replicate, the more valuable they become. And templates tied to a particular software or physical product help with both lead-gen and retention.

Examples: Miro Miroverse, Airtable Templates, SlidesCarnival

12. Case Studies

Moderate | Medium Effort

We love case studies. Case studies typically don’t drive a ton of traffic, but they support conversions and boost the brand. A good case study shows how a product, service, or process got results.

One thing: most case studies we’ve looked at underdeliver. They tend to be more promotional than instructional.

To win, you want to display a clear before-and-after result, expose as many metrics as you can, and clearly explain the process.

Examples: Vanta Customers, Webflow Customers, Shopify Case Studies

13. Original Reporting

Moderate | High Effort

Reporting is a tough gig. News publishers have seen a steady decline in Google search traffic year after year. AI answers that summarize news events make things even worse.

That said, there’s still a place for original reporting in an AI world. AI can summarize commodity news, but original reporting that is fast, in-depth, or exclusive can still earn citations and visits.

To keep beating the same drum, like every other content type on this list, Original Reporting needs to add deep value beyond the same information that everyone else reports.

Examples: The Information, The Verge Tech, Reuters Technology

14. Support and Document Pages

Moderate | Low Effort

Support and documentation pages aren’t always sexy and don’t always drive a ton of traffic, but like brand pages, online product companies absolutely need them as a foundational layer for search and AI.

AI may summarize your API endpoints, but you want users to reference and seek out your canonical documentation. Well-documented help and support pages help LLMs learn about your product.

Finally, public support forums, like Adobe’s here, can be a great way to scale documentation

Examples: Plausible Docs, Mux Docs, Supabase Docs

15. Guides & Explainers

Weak | Medium Effort

On one hand, guides and explainer content are at high risk of being supplanted by AI summaries. ChatGPT can easily explain to me what a CRM is. On the other hand, niche explainers and in-depth guides often work when backed by brand expertise and proprietary data.

The key is to leverage differentiation to help users understand a topic. This differentiation can take the form of unique data (e.g., Examine’s Evidence Grades) workflows, comparisons, or expert content hubs.

Elevate your guides and explainer content in a way AI can’t replicate.

Examples: Nerd Fitness Start Here, Examine Creatine, HubSpot What Is CRM

16. FAQs & Glossaries

Weak | Low Effort

Definition, FAQ, and glossary pages are among the easiest to produce and to replicate with AI. That said, they still work well when you can establish them as the canonical definitions for an industry.

For example, if I’m learning DNS, I’m more likely to seek out and trust the latest from Cloudflare over an AI Overview.

FAQs and glossary pages also provide a surface for building reference hubs, adding internal links to important product pages, and paths to customer support.

Examples: Moz Glossary, Cloudflare What Is DNS, Agile Alliance Agile Glossary

17. Lists & Roundups

Weak | Low Effort

Curated “Best of” and “Top 10” lists have gotten a lot of flak recently, and deservedly so. Marketers are using fake lists, with themselves at the top, to manipulate AI-generated answers and recommendations.

But Google and AI answers generate a lot of lists and recommendations for a reason: people really want them.

Like every other content type on this list, the standards for winning with lists and roundups in Google search are incredibly high - especially for competitive terms. Google has shifted to favoring original testing, published scoring criteria, and hyper-niche focus.

Examples: ICAgile Project Management Software, Website Planet Builders, IRunFar Best Jogging Strollers

Pro Bonus: Content Frameworks for Specific Business Models

By this point, I hope we’ve convinced you to stop publishing commodity blog posts, but what should your specific business actually tackle?

To help, we created a publishing rubric for Pro Subscribers covering these content types with examples for many different business models/websites:

• Saas

• Affiliates

• E-commerce

• Local Services

• Media/Editorial

• Marketplace/Directory

These frameworks available via the link below can help marketing teams rethink their content strategies and plan more resilient, more purposeful content roadmaps.

Read more

17 Content Types to Survive Google’s Zero-Click Future

To help, we created a publishing rubric for Pro Subscribers covering these content types with examples for many different business models/websites:

• Saas

• Affiliates

• E-commerce

• Local Services

• Media/Editorial

• Marketplace/Directory

Read more

Use this scorecard to evaluate your site on quality.

The Importance Of Page Quality Scoring

As you may know, Google improves its search algorithms by working with over 10,000 human Quality Raters around the world. The raters manually evaluate both search results and individual web pages for quality.

Google regularly publishes its 100+-page Rater Guidelines, which instruct Quality Raters on how to evaluate web pages. According to the guidelines, raters ultimately assign two scores to every page they are assigned:

1. Page Quality (PQ): Raters assign a PQ rating ranging from Lowest to Highest, with many possible scores in between

2. Needs Met: Raters assign a Needs Met rating ranging from Fails to Meet (FailsM) to Fully Meets (FullyM)

While Google stresses that no single rating directly impacts a site’s ranking, Google uses the ratings to gauge the quality of its results and, perhaps more importantly, as valuable data points to train Google’s machine learning algorithms at scale.

Google says that understanding how Quality Raters work can help you to improve your content and do better in search.

So, to many SEOs and publishers, Google’s Raters Guidelines is a valuable document.

It’s also extremely difficult to use if you want to score your own site.

Because of my own part-time work as a Quality Rater, SEO clients occasionally ask for private evaluations of their websites, which I oblige.

The challenge is that because of strict NDAs, I can’t simply use the same forms and documents provided by Google to do the job. Years ago, internal Google scoring cards leaked online, but those are now far outdated.

If we only had something like this today, site owners could more easily evaluate their own pages against Google’s quality guidelines.

That’s why, along with Graphic Designer Dawn Shepard, we created a new Google Page Quality Rating Scorecard.

How The Quality Scorecard Works

This new scorecard is 100% based on Google’s own publicly available Rater Guidelines. It’s also organized into a format that makes it far easier to determine a Quality score for your web pages.

Read more

Subscribe now

If you sell anything online, chances are AI is changing how your customers find you, how you market to them, and ultimately how they make purchases.

Google, OpenAI, Amazon, and other platforms are releasing a flurry of new protocols to enhance AI product search, visibility, and may ultimately allow AI agents to buy your products directly.

Things are moving very, very quickly.

To get a grasp of these changes, I spoke with one of the world’s leading experts on “Agentic Commerce”. Tim Resnik led Product and SEO teams at Walmart and Wayfair, and in his role at Botify, advised some of the top e-commerce retailers in the world.

Tim has tips for everyone on AI shopping. Watch the interview here.

Here are the top takeaways from Tim’s interview.

1. Product Feeds and Catalog Enrichments Are Job #1

Traditionally, product feeds were left to the teams doing paid ads or pushed off to an agency. Tim argues that because feeds and new protocols are becoming a primary way for AI to understand your content, then optimizing those feeds and improving your catalog data becomes much, much more important.

Google is developing new protocols, such as Universal Commerce Protocol, along with feed attributes, such as Conversation Commerce (currently in limited beta) to expand AI capabilities, but these still rely on your core product feed as your base.

Here’s an illustration of how some of these updated capabilities work in Google Merchant Center (and will in the near future) to enable AI integration.

Action: Treat your product feeds like valuable, canonical assets. As AI companies develop new protocols, feeds become a source of truth that they can access easily and reliably. AI needs real-time information, and your feed can provide that faster than waiting for a bot to crawl your site.

Additionally, cleaning up your feed and making it as error-free as possible becomes a bigger priority. Most feeds in Google Merchant Center are riddled with errors, and this can lead to inefficiencies in surfacing your product.

2. SEO Basics Are Even More Important in an AI World

AI bots don’t all read your content the same as Google does, and can struggle with things like JavaScript-heavy sites. Slow load times, poor site architecture, and other basic SEO issues can prevent AI bots from even seeing your products in the first place.

Action: Focus on SEO basics before any new AI tactics. “Basics” may be the wrong term to use here, but making sure you have fast load times, good internal linking, and your site doesn’t rely too heavily on hard-to-parse JavaScript will go a long way in helping AI bots understand your product catalog. Pay those SEO folks!

3. Brand & Loyalty Play Outsize Roles

In an AI Shopping world, AI platforms have access to millions of data points, including information on your products and your competitors. This makes competing on things like price even harder. And if customers purchase on AI platforms instead of on your sites, your advantages erode even further.

Tim makes the point that Brand and Loyalty become increasingly important in these perfect information systems. To stand out, companies need good PR and a brand that stands out.

Action: Focus a little more on your Brand, and less on your keywords. We know that brand is important in SEO, but many companies sadly neglect it. Instead of pouring all your budget into ranking for “best mattress online”, try to get a little more attention for your company, “Mattress Kindom.” Invest in digital PR that showcases both your brand and your product.

4. True “Agentic Commerce” Isn’t Here Yet

Tim argues that while companies are rolling out new protocols, it’s far too early to call this the age of Agentic Commerce. While AI is changing how people discover and shop for products, almost all purchases are still via traditional routes.

We may soon have a future where our agents buy our batteries from Amazon without us thinking about it. That future is still far enough off that we’re not sure what it looks like yet.

According to Tim, the biggest winners in AI shopping are most likely going to be the companies that:

1. Possess operational excellence

2. Nails the SEO basics better and tighter than everyone else

3. Have a strong brand presence

The smart companies aren’t chasing shiny objects yet. But everyone is getting ready to play with new tools.

Important Links

• Tim Resnik on LinkedIn

• Flection i/o

• AI Brand Authority Index

Want to discuss AI Shopping or have questions? Join the subscriber chat.

Subscribe now

Is your Google traffic rising or falling?

Business owners reach out to us almost weekly seeking “traffic recovery.” Typically, they’re hoping for technical SEO fixes or quick wins to turn their traffic around in a couple of months.

In reality, it’s their business model, user experience, or content production that’s killing them.

Google simply wants to give that traffic to a different type of site now.

The data support this. While it’s true that Google has decimated entire categories of publishers, other studies show that overall traffic declines are probably modest, and for every website losing traffic, there is often a winner.

What do these winning sites do differently from everyone else? Importantly, can we learn any strategies from them to protect and grow our own traffic?

To answer this question, I analyzed over 400 winning and losing websites, revisiting many of the same sites covered by Lily Ray’s excellent analysis of Google’s December update.

You can see the full list with updated traffic stats.

These include some very big movers over the past year, with a mix of recognizable brands and smaller players. We analyzed traffic trends over the past 12 months and then classified them by business model, content types, creator profiles, and other definable characteristics.

When we analyzed these features against winners and losers, very clear patterns emerged.

In short, Google has moved beyond simply ranking “good content” to proactively rewarding what AI can’t replicate.

5 Website Characteristics of Winning Websites

After crunching the numbers, we identified 5 features that very strongly predicted whether a site was winning or losing Google over the past 12 months. Here they are with their Spearman correlation.

1. Offers a Product or Service - 0.391

2. Allows Task Completion - 0.381

3. Proprietary Assets - 0.357

4. Tight Topical Focus - 0.250

5. Strong Brand - 0.206

Let’s dive into each to understand what they mean.

1. Offers a Product or Service

Winning Sites: 70.2%
Losing Sites: 34.6%

Offering a product or service was the highest correlated differentiator in our dataset. Sites that sold their own products performed especially well (as opposed to third-party platforms).

Losers tended to be news, informational, and affiliate sites.

Most interestingly, winners didn’t always sell a physical product. Service-based offerings did well, too, such as subscriptions and digital goods.

Winning Examples:

• budgetbytes.com resembles other recipe sites but offers a subscription meal plan service.

• mathnasium.com is a math tutoring site that offers online and in-person tutoring.

Losing Examples:

• byrdie.com operates primarily as a fashion publisher with no real product.

• medicalnewstoday.com is a large informational publisher, not a service provider.

Google says your content should have an “existing or intended audience” that doesn’t rely on search traffic.

Make it actionable: If you can position (or shift) your business model to make a proprietary product or service a core offering, that seems like a smart decision. It also helps build your brand, which we’ll discuss below.

Many website owners ask me if they can simply set up a webstore on their site to sell easy digital downloads. The answer is probably not. Sites that we see winning offer products and services as a core part of their business, not a hasty add-on.

Yes, it’s a big ask. Huge, in fact, with no guarantees. Travel blogs might become tour guides and offer vacation planning. But what else will they do when the Google traffic dries up, if it hasn’t already?

Here’s the good news: even if you don’t offer a product or service, there are other avenues to winning on Google.

2. Allows Task Completion

Winning Sites: 83.7%
Losing Sites: 50.2%

Task Completion means the website allows users to ultimately complete the task they set out to accomplish by searching.

Most losing sites may offer great information on topics, but the user needs to go elsewhere to actually do anything with it. WalletHub, for example, produces great credit card comparison pages, but the application process for those cards happens off-site.

You might think task completion has a lot of overlap with “Offers a Product or Service,” and you’d be right. If someone buys a pair of jeans at Levy’s, then they’ve completed their task.

But importantly, sites don’t need to sell anything to allow task completion. Consider these examples below.

Winning Examples:

• mathisfun.com is a math tutorial site that offers interactive tools, quizzes, and workbooks, providing users a way to practice math.

• stockanalysis.com acts as a complete research platform for stock analysis.

• powerball.com allows users to check their lottery tickets from an authoritative source.

Losing Examples:

• fortune.com explains business topics, but it is typically not where business takes place.

• byrdie.com covers fashion products, but doesn’t offer a direct way to obtain those products.

Make it actionable: If you don’t sell products or services, offering useful tools and/or data is typically the next most effective step to help users complete their task. Examples include calculators, quizzes, searchable databases, and booking tools. The goal is to own the next step of the user journey.

3. Proprietary Assets

Winning Sites: 92.9%
Losing Sites: 57.1%

Proprietary Assets means owning something that other sites can’t easily replicate. This could be a unique product, special databases, user-generated content, software, reviews, or anything else of value that’s “owned” by the website.

Winning Examples:

• letterboxd.com - Popular, fast-rising site that uses data from its large user base to graph movie popularity.

• todaytix.com - Maintains an inventory of up-to-date theater ticket availability.

Losing Examples:

• lifewire.com - Offers mostly tutorials and explainer-style content with a few “owned” first-party assets.

• thespruce.com - Popular home and lifestyle blog, but with few first-party assets.

Make it actionable: For many sites, a good place to start is with the data and unique information you already have and turning it into an asset. This can mean original research, pricing surveys, aggregating reviews, original photo collections, or organizing your audience into a source of user-generated content.

4. Tight Topical Focus

Winning Sites: 75.9%
Losing Sites: 61.3%

When we originally analyzed “Topical Focus,” we found no difference between winners and losers.

But when we switched to “Tight Topical Focus,” the pattern became clear, because it seems Google is getting tighter with its selections, too.

Winning sites tended to be more niche, covering a single narrow topic more deeply and with more expertise. Losing sites, on the other hand, tended to cover topics broadly or only loosely related.

Winning Examples:

• minecraft.wiki - Like Wikipedia, except only for Minecraft. Very tightly focused.

• happiestbaby.com - Laser-focused on babies. If you are a baby, this is your site.

Losing Examples:

• businessinsider.com - Covers business, but also entertainment, culture, and parenting.

• newsweek.com - Perfect example of a broad publisher covering many verticals.

Make it actionable: Obviously, the play here is to become known as a go-to authority for a small handful of topics. It’s not uncommon to see hyper-specific travel blogs (e.g., “Vern Does Venice”) out-rank global travel brands. Specificity and deep topical expertise seem to be the way to win.

5. Strong Brand

Winning Sites: 32.6%
Losing Sites: 16.1%

First of all, what do we mean by “Strong Brand”?

Using Ahrefs data, we examined each site’s top 20 keywords and identified branded navigational terms. These are terms people use when looking for a particular site. For example, “mental floss” is an obvious branded search term for folks looking for mentalfloss.com.

We then scored sites as “Strong Brands” based on both the volume of branded navigational terms and the percentage of overall traffic they represented.

Sometimes a site can be highly recognizable, but almost none of its traffic is branded. These aren’t considered strong brands.

Winning Examples:

• zoom.com - Extremely high brand visibility and lots of traffic from navigational queries

• skims.com - A popular shopping destination that users seek out directly

Losing Examples:

• lifewire.com - Highly recognized, but not a destination people actively seek out.

• techtarget.com - A niche site known to many, but mostly attracts traffic from longer-tail keywords rather than people seeking it out.

Make it actionable: I could write an entire guide on building your brand in search (and I will - subscribe for updates!) Suffice to say, this often means shifting your marketing from keywords to your core assets: products, founders, company name, etc. Apple doesn’t advertise “computers”, but they know how to showcase a Mac. This is a challenging but essential shift, even for sites in boring industries.

The Additive Effect: Winners Possess More Core Features

Perhaps most interestingly, when we examined the data, we found that these features are additive.

Displaying only one feature wasn’t enough for a website to see the winner’s column; a site usually needed multiple features to win.

Offering task completion was good, but 1) offering task completion + 2) a tight topical focus +3) a strong brand was even better.

Here’s the win rate for various feature counts:

• 0 Features: 13.5%

• 1 Features: 15.4%

• 2 Features: 22.0%

• 3 Features: 30.7%

• 4 Features 68.1%

• 5 Features 69.7%

Other Features That Likely Help, But Didn’t Correlate

For this study, we also classified and studied a number of features we thought would be significant, but had no bearing on the data. These included:

• Demonstrating first-hand experience or personal perspectives

• Hosting User-generated Content (UGC) or community platforms

• Uniqueness of information

I hesitate to include this information for fear that people will interpret it the wrong way. I don’t think these things didn’t correlate with success because Google doesn’t reward them, but more likely, they were already baked into the algorithm long ago. I suspect a larger data study would reveal interesting patterns here.

What Should Publishers Do?

Commodity content is dying, but not all commodity content is dying equally.

A winning site going forward has these characteristics:

1. It often offers a unique product or service.

2. It helps the user do something: buy, download, search, or book.

3. It stays within a niche topic of expertise, and it does it well.

4. It owns proprietary assets that are difficult to replicate: datasets, products, images, and UGC.

5. It becomes a destination site by positioning and promoting its own brand.

If you have an established business model, changing your focus and approach can be difficult.

We don’t know what search will look like in five years, but if you can offer something unique and help solve people’s problems in ways that aren’t easily reproduced, you have a good chance of finding yourself in the winner’s column.

Methodology - How We Gathered And Analyzed The Data

The dataset started with sites covered in Lily Ray’s analysis of Google’s December 2025 core update, which drew on Sistrix visibility data. Year-over-year traffic change for each site came from the Ahrefs API, measured over a 12-month window.

We then classified every site by hand against a fixed set of features, including business model, content type, creator profile, task completion, and proprietary assets. Each feature was recorded as a binary flag.

Brand strength also used Ahrefs keyword data. For each site we took the top 20 keywords, isolated the branded navigational terms, and scored sites on both the count of those terms and their share of total traffic.

To relate each feature to performance, we calculated Spearman rank correlations between the feature flags and 12-month traffic change. We used Spearman because it operates on ranks and tolerates the outliers common in traffic data.

These findings are correlational and rely on manual classification, so they indicate association and cannot establish causation.

Subscribe now

It's a great irony that while AI answers can reduce clicks to our websites, AI output often relies on user-click data to produce more accurate results.

That’s what I learned after diving deep into user click signals from Google’s API leak, antitrust trial, and patent filings.

If algorithms like “RankEmbedBERT” sound boring, it was digging into how Google defines goodClicks and lastLongClicks that really deserve attention. These are seemingly fundamental to Google’s ranking algorithms, and, more importantly, marketers can influence these signals in the right direction.

During the recent US antitrust trial, it was argued that user data—what people click on and interact with—made up the majority of “Google’s scale advantage“ and that simply sharing these user interactions would allow competitors to “mimic Google’s ranking.”

In fact, Google records almost every action we take: what we search for, what we click, and how long we spend on a website.

We are all a giant Mechanical Turk, telling Google which results are good.

Importantly, it turns out these signals also play a big part in AI answers and in any AI output that relies heavily on Google (looking at you, ChatGPT).

So exactly what click signals does Google measure? Can we influence them?

First, some caveats. What we know about Google’s use of click signals mostly comes from the U.S. v Google antitrust trial, Google patents, and Google’s big API leak. Importantly, simply because something is patented or has an API endpoint doesn’t mean we know 100% if and how Google uses it.

That said, as these exact concepts are referenced across all three sources, including trial testimony, it’s a fairly strong indication of their importance.

So treat this post as a conceptual framework about how Google thinks about, collects, and uses click signals for ranking and AI citations. While we don’t have every nitty-gritty detail, we have overwhelming evidence to point us in the right direction.

Known Click Signals Tracked by Google

Google’s API leak exposed many of the specific data points it collects and uses for its algorithms.

While there are 1000s of attributes in the leaked data, many are used across multiple ranking systems. Importantly, several attributes closely match Google’s patent filings and metrics referenced by Googlers in the antitrust trial. Here are 5 of the most salient click signals:

1. impressions - When a URL appears in a search result. Importantly, Google seems to “test” different URLs in different positions in search results to determine which ones perform best.
   

2. clicks - When a user clicks on a search result. If users frequently click on a particular result relative to its impressions, Google may treat that as a positive signal.
   

3. badClicks - “...a short click can be considered indicative of a poor page”. E.g., if users click a result and immediately click back to the search results, it may be a clue that the page didn’t satisfy their question.
   

4. goodClicks - “...a long click can be considered indicative of a good page” and “historically, how long a user stayed at a particular linked page before bouncing back to the SERP.”
   

5. lastLongestClicks - “The number of clicks that were last and longest in related user queries.” Long clicks, where “the user doesn’t return to the main (search) page,” can indicate that the page satisfied their question. (Sources: 1, 2, 3)

A picture is worth a thousand words, so here’s how these concepts play out when a user searches Google. Credit to Dawn Shepard for the graphics.

How Click Data Informs AI Answers (Nerd Section)

We’ve known for a long time that Google uses click data to build and re-rank search results via algorithms like NavBoost and Glue.

In fact, Google calls clicks one of its “three fundamental signals” or “ABC” - Anchors (links), Body (text), and Clicks.

Now you might assume that answers from AI models (LLMs) wouldn’t need click signals, but this is increasingly not the case due to several factors:

1. Google grounds its AI answer with algorithms (FastSearch / RankEmbedBERT) that use “70 days of search logs plus scores generated by human raters.” (source) These search logs contain click data and help generate the reference citation links Google includes in AI answers.

2. Answers from Google AI Overviews and AI Mode often summarize top-ranking results (and top-ranking results of fan-out queries). Those top results are surfaced, in part, using click data.

3. Other AI platforms, like ChatGPT, rely on Google search results to generate responses. Even though OpenAI has access to Microsoft Bing’s search results, “it appears those aren’t enough to replicate Google’s accuracy at scale, especially for obscure or long-tail queries.” - Tom’s Guide

While Google files lawsuits to limit other AI companies’ access to Google results, the influence of Google click signals on ranking and visibility is everywhere.

Note: There’s a fair amount of debate about whether Google uses clicks as a direct ranking signal or more for machine learning and evaluation. There’s also evidence that Google can simply predict clicks without actual data. Regardless, you almost certainly want your webpage to “look” like the kind of result Google believes will satisfy visitors.

The Playbook: Optimize for Relevance, Engagement, and Satisfaction

From the evidence, Google is trying to answer three broad questions from user click data:

1. Is this result relevant?

2. Is the content useful to answer the question?

3. Does this completely satisfy the user search?

Let’s examine each in turn.

1. Prove Relevance - Earn The Click

The idea seems simple: Google displays a list of URLs it thinks are relevant to a search, then measures which results users actually click.

Get more clicks, rank higher. Right?

Not exactly. This is not a simple Click-through Rate (CTR) calculation.

Google itself says that clicks are very “noisy,” and factors like position and appearance can skew the data. They even have metrics that appear to punish “clickbait” titles.

So the goal is not “high CTR.” The goal is to earn good clicks by showing relevance to the user.

Your SERP snippet is your playground to accomplish this, and as marketers, we have a lot of ways to influence this appearance:

1. Intent - What exactly did they search for?

2. Promise - How does our content meet that need?

3. Brand - Who delivers that promise?

Again, and I cannot emphasize this enough, the goal is to improve CTR but not at the cost of relevance. Exaggerated or “clickbait” features may drive clicks, but they are not as relevant. Getting a high CTR without demonstrating usefulness and satisfaction (steps 2 and 3 below) can lead Google to associate your site with very bad signals.

So a high CTR is good, but it needs to be paired with other signals…

2. Prove Usefulness - goodClicks vs badClicks

The evidence is consistent: Google appears to measure long/good clicks by observing how long a user visits a result before returning to the search results.

Except, of course, it’s not that simple.

This isn’t exactly “dwell time” or “time on page.”

That’s because Google weighs each click time by query, user type, language, and country.

Then they can compare these scores with those from all other possible query-document combinations. One Google patent describes measuring the ratio of long clicks to overall clicks, and this click fraction “can be used to re-rank future search results.”

By using weighted clicks, Google understands that not every visitor needs the same amount of time for every search. For example, a user checking “how tall is Mount Everest” may only need a few seconds, while a question like “optimize investment portfolio for inflation” might take several minutes.

Context is everything.

The goal isn’t simply to improve time on site, but to demonstrate usefulness through engagement.

We can accomplish this with basic User Experience tools, but these practices often get lost in the realities of building web pages. Here’s what it can look like in practice.

You may think that answering the user query “quickly and obviously” isn’t the best way to earn “long clicks,” but there’s plenty of anecdotal evidence across the industry that it often leads to better traffic. (This may be related to step 3 - lastLongestClicks, but for now, we’ll focus on engagement.)

Note: Google has repeatedly said “time on site” isn’t a ranking factor, which makes sense given the evidence. They obviously slice the data in much more complex ways than that.

That said, many, many studies have found a positive correlation between time on site and Google rankings. In one study, our friend Backlinko found that “increasing time on site by 3 seconds correlates to ranking a single position higher in the search results.”

And, my marketing friends, time on site through better engagement is something we can influence.

3. Prove Satisfaction - lastLongestClicks

Perhaps the holy grail of click signals: lastLongestClicks. Consider this user journey:

1. User asks Google a question

2. Clicks into your website

3. Engages with content

4. Gets the question 100% answered

5. Doesn’t go back to click another result

Congrats, you earned the last, long click!

Each step builds on the others. In addition to earning the click and engaging the user, you also want to be the last place they search.

How do you do this? You can start by:

• Answering related questions

• Show your evidence

• Providing the user with the next steps

The goal here is to anticipate the next user need to complete their task. I love this example from AJ Kohn:

“If you are a local directory site, you usually have reviews from your own users prominently on the page. But it’s natural for people to wonder how other sites rate that doctor or coffee shop or assisted living facility. My recommendation is to display the ratings from other sites there on your own page. Why? Users want those other ratings. By not having them there, you chase users back to the search results…” (source)

If the user requires additional information to complete their search journey, you are better off providing it when they need it rather than having them leave to search elsewhere.

Things to Know: Manipulation is Hard By Design

One thing we should note is that these click metrics are, by design, incredibly hard to “game.” Google employs multiple techniques to ensure click signals reflect genuine user reactions:

1. Navboost, one of Google’s most important algorithms, memorizes 13 months of user interaction data. This long timeframe adds stability and means changes happen slowly. (As noted above, FastSearch uses 70 days of data.)

2. Google calculates a “squashed“ value for many of its metrics, which presumably filters out spam, manipulated, and/or noisy signals.

3. Not all user interactions are equal. “Slicing“ weighs each user click against the ranking position, the query, the device, the user’s location, and a host of other factors.

That said, there is anecdotal evidence that click spoofing (employing bots or humans to click on your search results) can sometimes temporarily boost rankings, but the effect is generally short-lived and potentially risky.

Measuring Clicks By Proxy

You can’t improve what you can’t measure. Unfortunately, Google doesn’t share click data, and it’s nearly impossible to measure individual clicks with any granular detail.

That said, there are many metrics we can measure that correlate with click data and deliver independent value. The important part to remember is that, since you can’t typically see competitors’ data, you are essentially competing against yourself.

1. Click-through Rate

You can measure CTR in Google Search Console. It’s tricky because CTR changes drastically by position, and impressions from AI answers muddy the waters. It’s typically easiest to measure CTR across broad sections of your site using URL patterns. Tools like SEOTesting and SEOGets can help track CTR.

2. Engagement

My absolute favorite metrics to measure are Engagement and Time on Site. Google Analytics 4 (GA4) may be everyone’s least favorite analytics tool, but I believe GA4 did us a big favor with the Engagement report, which shows us “Average engagement time per active user” and other metrics. (Remember to filter by organic search traffic.)

For those with a budget, competitor engagement metrics can be determined using Semrush Traffic and Market or SimilarWeb.

Remember, “Time on Site” isn’t itself a ranking factor, but it seems to play an outsized role in how Google calculates other metrics.

3. Satisfaction

There isn’t “one” metric to track visitor satisfaction. We simply don’t know if a user returned to search results. But there are a number of metrics you should track, regardless, as measures of brand strength.

1. Returning Users: Available in GA4. An increase in users returning to you via organic channels or otherwise is typically a good sign.

2. Conversions: Especially relevant to e-commerce, but applicable to other verticals as well. If a user makes a purchase, it’s a pretty good sign that you satisfied their journey.

3. Brand Search: The Holy Grail of search terms. When a user specifically searches for your brand, their journey can both start and end with you. In fact, all marketers should probably report branded search. And with branded search filters now available in Search Console, we have no excuse not to.

And here, we’ve reached the end of our own search journey. How do you think you can improve your own user click signals?

Want to discuss or have questions about Google Click Signals? Join the subscriber chat.