We Rank First and Get Zero Clicks: What Our Own Search Console Data Shows

Most writing about zero-click search quotes the same handful of industry studies. We wanted to know what it looks like inside a single site, so we are publishing our own numbers.

This is one small site over three months, which is a real limitation and we set it out in full below. What makes it worth reading is that the pattern is unusually clean.

Across 1000 queries, we rank on page one for 124 of them. 123 of those produced no clicks at all.

What does zero-click search look like in Search Console?

Short answer: It looks like healthy rankings attached to nothing. Impressions accumulate, average position improves, and the clicks column stays at zero. In our last 3 months, 123 of our 124 page one queries produced no clicks, and 26 of our 27 top three queries did the same.

Search Console reports impressions, clicks, click through rate and average position for each query. The failure mode most people watch for is a bad position, because that is the number that has always predicted traffic.

That relationship is what has broken. A query can sit at position 2 with a few dozen impressions and never send a visit, and nothing in the interface flags it as unusual.

Those 123 zero-click page one queries carried 2,481 impressions between them. The demand was real and it was ours, and it did not become a visit.

Why does ranking first produce no clicks?

Short answer: Because the answer is now rendered on the results page. When a summary, an AI Overview or a direct answer satisfies the query, the ranking that earned the citation has already done its work. The user reads and leaves, and your logs record an impression rather than a session.

Pew Research measured this directly across a panel of real users: people clicked a result on 8% of visits where an AI summary appeared, against 15% of visits without one. Pew Research Center, July 2025.

That is roughly half the click rate on the same results page. Our own numbers are what the thin end of that distribution looks like when a site is small enough that nothing averages it out.

It is worth being precise about the mechanism, because it is often described as a penalty. Nothing has been demoted.

The page ranks, the citation is granted, and the click is simply no longer the way the value arrives.

Which queries lose their clicks?

Short answer: At our volume, effectively all of them. 997 of our 1,000 queries produced zero clicks. We looked for a difference between question-shaped queries and the rest and could not find a meaningful one, because the zero-click rate is near total in both.

We expected to report that conversational queries lose clicks faster than head terms. The data does not support that, and it is worth saying so plainly rather than presenting a tidier finding than we have.

Among our page one queries, 49 are question-shaped and all 49 produced zero clicks. The other 75 produced one click between them.

Those rates are 100% and 98.7%, which is not a difference anyone should build a strategy on.

What does hold up is the absolute version. Across the full export, 189 queries are phrased as questions and they produced zero clicks between them, including four sitting at position 1.0:

• "best seo agencies for local businesses"
• "recommend a local seo agency"
• "which agencies specialize in reddit and quora brand placement"
• "which local seo agencies have the best results for service businesses"

Each is a question a person would ask an assistant out loud. Each has an answer that fits in a paragraph, which means the results page can deliver it whole.

The honest conclusion is narrower than the one we set out to draw. We cannot show that question-shaped queries lose clicks faster, only that ranking first for them produced nothing at all, which is the part that should change how you plan.

That still has a direct consequence.

The query class AEO work is most effective at winning is a class where a win may never appear in your click data, so the better your answer engine visibility gets, the less your traffic reporting will show for it.

Does this mean AEO work is not paying off?

Short answer: No, but it does mean clicks are the wrong first measure. Position and impressions move before clicks do, and in a zero-click query class they may be the only things that move. Judging a new programme on clicks alone will show failure while the work is succeeding.

Our position and impression data moved substantially over this window while clicks did not. Taken alone, the clicks column would read as a programme that achieved nothing.

The honest reading is that we cannot yet prove commercial return from this data, and neither can anyone else working at this scale. What we can show is that the leading indicators moved and the lagging one did not follow in the usual way.

The research on what improves citation rates is more useful here than click data.

Aggarwal and colleagues found that adding credible quotations, statistics and sources lifted visibility in generative engines by up to 40% in their test set, which is a lever you can actually pull. GEO, KDD 2024

What should you measure instead?

Short answer: Measure impressions and average position by query class, citations in AI answers, and branded search volume. Those three move before revenue does. Keep clicks in the report, but stop treating a flat click line as proof that nothing is working.

1. Impressions by query class. Split question-shaped queries from brand and navigational ones. They behave differently enough that a blended number hides both.
2. Average position on the queries you chose to target. Not the sitewide average, which moves whenever you add pages and tells you very little.
3. Citations. Ask the assistants the questions you want to win and record whether you are named. It is manual and it is the most direct evidence available.
4. Branded search volume. If people read about you in an answer and look you up later, this is where it surfaces.
5. Assisted conversions. A zero-click impression is not worthless if the visit arrives a week later through a branded search.

One trap worth naming. Sitewide average position gets worse every time you publish pages that start from a standing start, which can read as a decline while every individual page is improving.

We hit exactly that. Our own sitewide average position degraded sharply over this window while individual target keywords improved, purely because the number of ranked queries grew faster than their positions matured.

How should this change what you publish?

Short answer: Write answers that survive being quoted. Lead each section with a direct, self-contained response, attach the evidence to it, and accept that the citation may be the whole of the value. A page built to be excerpted earns mentions that a page built to tease a click does not.

The instinct when clicks dry up is to withhold the answer so people have to visit. That reliably fails now, because an engine that cannot extract a clean answer from your page will extract one from a page that offers it.

Make each answer self-contained

A paragraph that depends on the three paragraphs above it cannot be quoted. Lead with the answer, then support it, so any section can be lifted without losing its meaning.

Attach evidence to the claim

Statistics and named sources are what make a passage quotable, and they are what the GEO research found moved citation rates. A claim with a source attached travels; a claim without one gets paraphrased without attribution.

Say something only you can say

This article is an example of the argument it is making. We can publish our own Search Console numbers and no competitor can copy them, which is a more durable asset than another summary of someone else's study.

Google's own guidance on AI features is consistent with this: the features draw on core ranking systems, so there is no separate technical trick, only content that is genuinely useful and extractable. Google Search Central, AI features and your website.

What are the limits of this data?

Short answer: It is one small site, over three months, at low absolute volume. The percentages are stable and the pattern is consistent, but a sample this size cannot establish an industry benchmark. Treat it as a worked example, not a study.

We want to be explicit about this, because the alternative is to let a clean number travel further than it should.

• Single site. One domain in one set of niches. Nothing here generalises to ecommerce, publishers or anything at a different scale.
• Low absolute volume. Percentages computed over small counts are fragile. 123 of 124 is a striking ratio over a modest base.
• Three months. Long enough to see a pattern, not long enough to separate it from seasonality.
• Search Console position is an average. A reported 1.0 means it averaged first across the impressions recorded, not that it held first place continuously. Google, Performance report data
• Growth confounds it. We added a large number of pages during this window, which inflates impressions and drags sitewide averages independently of anything else.

What survives those caveats is the shape rather than the magnitude. Page one rankings existed, they were concentrated in question-shaped queries, and they did not convert into clicks.

What should you do if your own data looks like this?

Short answer: Export your queries, split them by position and by whether they are question-shaped, and look at the zero-click share in each group. If your page one queries are producing no clicks, the fix is in measurement and format, not in rankings you already have.

1. Export the last 3 months of queries from Search Console as a spreadsheet.
2. Filter to position 10 or better, then count how many have zero clicks. That single ratio is the diagnostic.
3. Split those into question-shaped and brand or navigational, and compare the two zero-click rates. If both are near total, as ours were, the problem is not a content gap in one group.
4. Check the pages behind those queries actually serve their content to something that does not run JavaScript. A page that renders only in a browser cannot be quoted by an engine that never runs one.
5. Rewrite the top ten offenders to lead with a direct answer, and re-measure in 90 days on impressions and citations rather than clicks.

Step four catches more sites than people expect. Check the structure of a page before rewriting it, and if you want the machine-readable version of the same idea, generate an llms.txt file so assistants have a curated map of what you publish.

The broader method is the subject of our guide to doing AEO step by step, and the measurement side is covered in what AEO services actually include.

Frequently asked questions

What is zero-click search?

A search where the user gets what they needed on the results page and never visits a site. It covers featured snippets and knowledge panels as well as AI Overviews and AI assistants, and it is measured as impressions without a corresponding click.

Is a zero-click impression worth anything?

Yes, but not as traffic. Being named in an answer builds familiarity and can produce a branded search later, which is why branded volume and assisted conversions are the places to look for its effect rather than the clicks column.

How can you rank first and get no clicks?

Because position measures where your result appears, not whether anyone needs to open it. If the answer is rendered on the results page, ranking first means your page was the source rather than the destination. Four of our queries sit at position 1.0 with zero clicks.

Which queries lose clicks first?

We could not establish that any query type loses them first, because in our data the zero-click rate was near total across every group. What we can report is that 189 question-shaped queries produced zero clicks between them, four of them from position 1.0.

Should I stop doing SEO and only do AEO?

No. The same ranking systems feed AI features, so technical health, content quality and authority still decide whether you are eligible to be cited. AEO changes what you optimise the content for, not whether the fundamentals matter.

How do I track whether AI engines cite me?

Ask the assistants the questions you want to win and record whether you are named, on a fixed schedule. There is no reliable automated equivalent yet, and GA4 referral data captures only the small share of AI sessions that produce a click.

Why did your average position get worse while your rankings improved?

Because sitewide average position is diluted by every new query you start ranking for. We added a large number of pages during this window, and each arrived at a weak position, which pulled the average down while individual target keywords improved.

Sources and further reading

• Pew Research Center: Google users are less likely to click on links when an AI summary appears (July 2025)
• Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024)
• Google Search Central: AI features and your website
• Google Search Central: Creating helpful, reliable, people-first content
• Google: Search Console Performance report, how position is calculated

Want to know what your own data says?

We will run this analysis on your Search Console export and tell you which of your rankings are already producing nothing, and which pages are not even readable by the engines doing the citing. Book a free audit.