Across the Search Console accounts we watch, one pattern keeps repeating as AI Overviews and AI Mode take over more of the results page. Impressions climb while clicks stay flat or drift down. The query fires, your page qualifies, and the answer sits above it, assembled from your page and a few others. The searcher reads the answer and never clicks through. The impression is real. The visit is gone.

This site shows the shape in its rawest form. In our own Search Console, seo.concept211.com carries a stack of high-impression queries with click counts at or near zero. Some of that is us: we are a young domain writing about a crowded topic, and a fair number of those impressions sit too low on the page to earn a click regardless of AI. But part of it is the newer thing, where the impression is genuine and the click never had a chance, because the answer a searcher wanted was already rendered above the links. We would rather tell you that plainly than pretend our own numbers are clean.

That is the honest frame for reading the click figure Google recently put into circulation.

What number is this, and why does it not settle anything?

On July 18, 2026, Search Engine Journal published "Google Puts A Number On AI Search Clicks, Without The Data." The gist: Google has said that clicks originating from its AI experiences tend to be higher quality, and has attached a figure to the volume of those clicks, without releasing the underlying data anyone would need to verify it. So the number exists, and the data behind it does not, at least not anywhere a site owner can get to it.

That gap matters more than the figure itself. There is no query-level breakdown, no export in Search Console that isolates AI-answer clicks from ordinary ones, and no way to reproduce the calculation on your own traffic. You are asked to accept an aggregate about the whole web and apply it to a site you can actually see. Those are not the same object.

What the number is and isn't
Google's AI-search click figure is a web-scale aggregate reported without its supporting data. It is not something you can find, filter, or reproduce in your own Search Console. Treat it as a claim about the average site you will never be, not a forecast for the ten pages you own. A number you cannot break down or check is context, not a target.

What the figure actually implies for a small site

For a small site, the number describes something Google sees and you do not. Even if the figure is exactly right at web scale, it says nothing about your handful of pages, because AI clicks concentrate rather than spread evenly. The searcher who does click through from an AI answer tends to click the source that answered the question, and that source usually already had authority on the topic. If your page is one of the named sources, an AI answer can hand you a motivated visitor. If it is not, the same answer quietly removes a click you used to get, and no site-wide average will show you that happening.

This is why a rosy aggregate and a falling click count can both be true at once. The mean is being lifted by the sites that win citations, while the long tail of sites that used to catch a stray click from a mid-page ranking loses it to the answer box. If you run a small site, assume you are in the second group until your own data proves otherwise. Plan for your distribution, not Google's mean.

Why the average misleads
AI-search clicks concentrate on the sources an answer actually cites, which are usually pages that already ranked well. A web-scale average pools those winners with everyone who lost a mid-page click to the answer box, so a healthy-looking mean can sit on top of real losses at the small-site end. Your job is to find out which side of that split your pages are on, not to trust the pooled figure.

What to measure instead

Because the headline figure is unbacked and cannot be exported, chase the signals you can actually see and that actually move. Four are worth watching.

Impressions trend, by query and page

Rising impressions with flat or falling clicks is the signature of an AI answer eating your click. Read the direction over several weeks rather than a single ratio, and read it per query. A page whose impressions are climbing is still being surfaced; the question is whether it is being surfaced as a citation or as raw material for an answer that names someone else.

Citation presence in AI answers

Being named as a source in the AI answer for your core queries is the new equivalent of a ranking, and unlike Google's figure it is measurable. Pick a fixed set of real queries, check whether your domain is cited across the engines you care about, and record which page earned it. Watching that set move week over week tells you far more than any aggregate. We cover the mechanics in GEO citation tracking.

Branded-query lift

An AI answer that mentions you without a click still plants the brand. People who saw your name in an answer come back later and search for you directly. A rising trend in branded searches over the following weeks is how an uncredited mention shows up in data you can see, so watch branded impressions and clicks as a lagging read on AI visibility.

Assisted conversions

The visit that eventually converts may have started with an AI answer that sent no click and got no credit. Judging AI-era performance on last-click alone will understate it. Look at assisted paths and the branded or direct sessions that tend to follow an answer-box appearance, so you are valuing the whole journey rather than only its final step.

The measurement swap
Stop grading AI search on the raw click count and Google's unverifiable figure. Track four things you can see: the impressions trend per query and page, whether you are cited in the AI answers for your core queries, branded-search lift in the weeks after, and assisted conversions rather than last-click alone. Together those tell you whether AI search is finding you, naming you, and eventually paying you.

What to do about it

The takeaway is not to fight the answer box. It is to become the source it cites, and to measure yourself against that instead of a number you cannot audit. A short list of concrete moves:

  • Confirm the AI crawlers can reach your pages. Check robots.txt and your edge rules so GPTBot, PerplexityBot, ClaudeBot, and Google-Extended get a real 200 and not a bot challenge. A page that cannot be fetched cannot be cited.
  • Front-load the answer on pages that matter. Open each key section with a direct, quotable answer to the question it targets, then expand. The passage that states the answer up front is the one an engine lifts and attributes. Our guide to earning citations in AI search walks through the structural work.
  • Set up citation tracking on a fixed query set. Decide the queries that matter, check them on a schedule, and log whether you are named. This is the ranking report for the answer era.
  • Rebuild your dashboard around the four signals above. Retire the click count as your headline metric for AI-influenced queries and replace it with impressions direction, citation presence, branded lift, and assisted conversions.

None of this is exotic. Most of it is a few hours of technical and editorial work per important page, and it compounds once done. The reason it usually does not get done is the same reason a lot of good SEO never ships: it lives in a report instead of going to production.

Where this leaves you

Google's click figure is a headline, not a forecast for your site. The useful response is to stop reading averages you cannot check and start reading your own impressions, citations, branded lift, and assisted paths, then do the work that turns your pages into the sources AI answers name. If you want to see where your site stands on the crawler access and citable-passage basics right now, drop your domain below and we will send a free audit that checks exactly that.