The whole dataset, first
Over 28 days our site recorded 121,388 impressions and 467 clicks. That is a 1.20% click-through rate at an average position of 12.2.
Most published commentary on AI Overviews cites aggregate industry studies. This is one site's raw Search Console export, which has the opposite trade-off: no sample size worth boasting about, but nothing hidden either.
Two pages ranked on page one and got nothing
These are the two that stopped us:
best-marketing-dashboard-software-2026— 2,298 impressions, 0 clicks, average position 5.4best-marketing-analytics-software-2026— 3,149 impressions, 0 clicks, average position 6.7
Not one click each, across four weeks, from mid-page-one. A page cannot take zero from 3,149 impressions at position 6.7 by chance. Something is consuming the click before it happens.
The pattern across twenty pages
Twenty pages on our site hold 93,625 impressions between them and returned 215 clicks. That is 0.23%. At the positions these pages occupy, an ordinary CTR would be several percent. We are getting roughly a tenth of it.
The worst offender by volume is a Meta Advantage+ guide: 17,072 impressions, 20 clicks, 0.12% CTR, average position 7.5. It is one of the better-performing pages on the site by ranking, and it delivers about five visits a week.
What still gets clicked — and it is not subtle
Two pages on the same site, in the same period:
- A page of Google Ads audit prompts — 14.55% CTR at position 5.7
- A white-label reporting walkthrough — 20% CTR at position 4.4
Same domain, same authority, similar positions, a hundred times the click-through rate. The difference is not ranking. It is whether the answer can be summarised.
The clearest illustration sits inside a single topic. For the query meta safe zone template we rank 13.8 and take a 7.14% CTR. For 1080x1920 safe zone we rank 2.2 and take 0%. We rank six times better on the second query and get nothing from it, because the answer is a number and the number is already on the results page. A template is a file. You cannot put a file in a summary.
How much of this is actually AI
Search Console reports our appearances in generative AI features separately: 50,700 impressions over three months. Bing's equivalent report shows 8,400 AI citations in the same window. Against roughly 59,000 AI appearances, our AI-referred sessions numbered 1,027 — a 1.7% click-through from being cited.
So being quoted by an AI system is not the same as being visited by its user, and the gap is roughly fifty to one. That is worth knowing before anyone builds a strategy on citation counts.
What this data cannot tell you
It is one site, one 28-day window, in one niche. Treat it as a case study, not a study. We cannot separate AI Overviews from ordinary SERP features, ads pushing results down, or our own titles being weak — Search Console does not expose that breakdown.
What survives those caveats is the internal comparison. Every page here shares one domain, one authority profile and one time window. When a prompt page converts at 14.55% and a specification page at the same site converts at 0% from a better position, the variable that changed is what the page offers, not who published it.
What we changed
We rewrote titles across those twenty pages to lead with the thing a summary cannot deliver, and we built the tool the specification pages were only describing. Our safe zone checker now does what the pixel-spec query cannot be answered with: you upload your own creative and see what gets cut.
Whether that recovers the clicks is an open question and we will publish the follow-up either way. The measurement window closes in mid-September.
The takeaway for anyone publishing
Ranking is no longer the finish line for informational queries; it is the price of entry to a results page that may answer the question without you. The pages that still earn a visit offer something the answer box cannot reproduce — a tool, a template, a calculator, a dataset, or numbers only you have.
The practical test before publishing: if an assistant read this page, could it give the reader everything of value in three sentences? If yes, expect impressions rather than visits.
If you want to see this pattern in your own Search Console data, 1ClickReport reads it alongside Google Ads, Meta and GA4 in one conversation — so “which pages get impressions but no clicks” is a question you ask once rather than a spreadsheet you build. You can try it free.
Frequently asked questions
Do AI Overviews reduce click-through rate?
In our 28-day data, pages ranking at positions 5.4 and 6.7 for informational commercial queries returned zero clicks from 2,298 and 3,149 impressions. Across twenty such pages, 93,625 impressions produced 215 clicks — a 0.23% CTR against several percent expected at those positions.
Which pages still get clicked?
Pages offering something an answer cannot contain. A prompt pack returned 14.55% CTR at position 5.7 and a template page 20% at position 4.4, while specification pages ranking first to third in the same cluster returned zero.
Is this just a small-sample problem?
It is one site over 28 days, so treat it as a case study. The individual results are large enough to stand alone: a page cannot take zero clicks from 3,149 impressions at position 6.7 by chance.
What should publishers do about it?
Stop measuring by ranking alone. Measure whether the page offers something the summary cannot reproduce — a tool, a template, a calculator, a dataset, or first-hand numbers.
Related reading: why Search Console impressions have been unreliable, what the llms.txt evidence actually shows, and how to track AI traffic in GA4.