Impressions But No Clicks: The Ten-Minute Check

I keep an eye on the search numbers for my own site the way I do for clients, and this month the headline looked like the best news I have had all year. Impressions up fourteen thousand on the previous month. A cluster of searches where I had suddenly arrived on the first page. If you handed me that as a report about your business I would have said, without much hesitation, that something good was happening.

Then I opened the click column.

What the impressions showed

The figures below are all from the twenty-eight days to 10 August 2026, on my own property, quoted exactly as they came out.

The site drew 35,059 impressions and 51 clicks. One page — a comparison piece I wrote about content management systems for Astro — accounted for 10,077 of those impressions and 44 of the clicks. Which is to say that one page out of everything I have published carried 86% of the clicks the entire site received.

Inside that page’s own numbers there was something stranger.

Forty-nine different searches sent people to it. Ten of them produced clicks: 530 impressions, 44 clicks, a click-through rate of 8.3%, which is a perfectly ordinary result for a page sitting near the top. The other thirty-nine produced 9,547 impressions and no clicks at all. Not a low number. Zero.

My first thought was the obvious one: the searches getting nothing must be the ones where I rank worse. That is true, and it is nowhere near enough. The thirty-nine average position 7.1 against the converting ten at 4.6, so they do sit lower down the page — but ranking lower thins a click rate. It does not take it to zero.

So I stopped comparing searches to each other and compared like with like: the same page, the same month, sorted into position bands, with each band split by how long the search was.

In the band where both kinds appear in volume, the long searches convert about eight times worse than the short ones sitting beside them. In the band below it, four thousand impressions produce nothing whatsoever, while the short searches down there are still clicking.

That is the part position cannot carry. Two searches at the same height on the same page, and one of them behaves like an audience while the other behaves like nothing at all.

Do AI Overviews cause impressions with no clicks?

Everybody’s first explanation in 2026 is the AI Overview — the answer box Google now writes at the top of a lot of result pages, which reads your page, summarises it, and gives the reader no particular reason to visit. It is a real phenomenon, it does take clicks, and it was sitting right there as an explanation.

So I checked. I had stored copies of the result pages for four of the zero-click searches, pulled a few days earlier, and every one of them had an AI Overview on it, with my page at organic position three to five underneath. Case closed, obviously.

Except a theory that only ever gets tested against the cases it was invented for is not a theory, it is a hunch wearing a lab coat. So I bought the result page for best cms for astro — the one that converts at 9.1% — for about a third of a penny, and looked at that.

It has an AI Overview too. My page sits at organic position three underneath it, in the same configuration as the search that gets nothing.

That kills it. An AI Overview above you is perfectly compatible with a 9.1% click-through rate, because I have one, on the same page, in the same month. If I had stopped at the four searches that agreed with me I would have written a confident piece blaming AI Overviews, and I would have been wrong, and nobody reading it would have had any way of knowing.

What separates the searches that click from the ones that don’t

With the easy answer gone I went back and looked at what separated the two groups, and four things did.

The searches that convert are short. The ten that produce clicks average four words: best cms for astro, best astro cms, best headless cms for astro. The thirty-nine that produce nothing average six and a half, and read like sentences rather than searches — what are the best git-based cmss for astro?, best astro cms for a marketing site. Several carry a question mark, which is a character that belongs to written questions rather than typed ones. One of them uses the word “cmss”, a plural I have never seen a human being write.

The volume curve runs backwards. In normal search, the short general term draws the most traffic and the specific variants trail behind it. Here, best hosted astro cms drew 1,359 impressions and the head term it is a variation of drew 252. A narrow variant out-drawing its own head term five to one is not how people search.

It arrived in a step, not a climb. I pulled the daily figures back through the summer. The long searches sat between five and sixteen impressions a day until 23 July. Then three days at 711, 644 and 557, then three days back down at ten, twenty, ten — and from 31 July they settle at six to nine hundred a day and stay there. A page climbing the rankings does not do that, and it would lift both kinds of search together. This lifted one and left the other alone.

And the device split is the one that stopped me. The zero-click searches ran 8,121 impressions on desktop against 24 on mobile. That is 338 to one. The searches that convert run thirteen to one, and my site overall runs under seven to one.

What Search Console can’t tell you

I do not know what those impressions are.

I can tell you what they are not, with reasonable confidence: they are not thirty-nine people a day reading a sentence-shaped question into their phone. The device ratio alone rules that out, and the step change and the backwards volume curve say the same thing from different directions.

But there is no field in Google’s reporting that says this impression was served to a machine. The dimension you would reach for — search appearance — returned one entry for my whole site, about product snippets, and nothing about AI or assistant surfaces at all. So anyone telling you with confidence which surface this is, including me, is telling you something they cannot check. I would rather write that sentence than the confident one.

What I can say is that a number I have reported on for years, and that you are probably being reported on by somebody, changed by a factor of twenty-nine in a single month without my audience changing at all.

This has happened before, and we know how it ended

In 1914 a group of American advertisers, agencies and publishers set up the Audit Bureau of Circulations. The reason was simple and slightly grubby: advertisers were buying space against circulation figures that publishers supplied about their own magazines, and the figures were not true. Not fraudulently, always — just optimistically, and with nobody in a position to check.

What the bureau changed was not the honesty of publishers. It was who counted. An independent body audited the figure, and the industry then did something sharper still: it stopped treating copies distributed and copies read as the same number, and measured them separately, because they had learned the hard way that one does not imply the other.

That is precisely the position search reporting is in now. The number is supplied by the platform, about its own distribution. It has never claimed to measure attention. And for most of the life of search it did not need to, because a link in front of a person and a person seeing a link were near enough the same event that the distinction was academic.

They have come apart, and nothing has yet arrived to separate them.

What impressions without clicks mean for your site

If you take one thing from this, take this: a rise in impressions is not evidence that more people saw you. It is evidence that Google put your link in front of something more often. Most months those are the same sentence. This month, on my site, two-thirds of the growth was not.

That matters commercially in a specific way. If you are paying somebody for visibility work and the report leads with impressions and average position, that report can improve, honestly and verifiably, in a month where not one additional human being encountered your business. Nobody has to be lying for that to happen. The number just stopped meaning what everyone agreed it meant.

It is the same shape as why a website gets visitors but not enquiries — the traffic is real, the failure is quiet, and the figure you are looking at is fine right up until you check the layer underneath it. It is the argument in the dashboard is a decoy arriving one level further up: every layer reports on itself, and reports honestly, and the honest report is not the answer to the question you were asking.

The Search Console check, and it takes about ten minutes

Open Search Console, go to Performance, set the date range to the last twenty-eight days, and turn on the Impressions, Clicks, CTR and Position columns if they are not already showing.

First, sort by impressions and read the top twenty searches. Look at the shape of them. Are they the way you would type something into Google, or are they written out like sentences with question marks on the end? If your top searches by volume are all long and none of them click, that is the thing I am describing.

Second, find your best-converting search and compare it to your biggest one. Not your average CTR — the average hides this completely. You want two rows side by side, the way I put them above. If your best position is your worst click rate, position is not your problem and you can stop looking there.

Third, check the device split. Add the Devices tab and compare desktop to mobile for the whole property, then filter to one of the big zero-click searches and look again. You are not looking for a specific number. You are looking for a search whose device mix is nothing like the rest of your site’s, because a genuine audience does not change shape from one search to the next.

None of that needs a specialist and none of it costs anything. What it needs is somebody to go and look at a month where the headline number went up, which is the hardest time to persuade anyone that something is worth checking.

What I’d do this week

Run the three checks. If everything looks ordinary, you have bought a clean answer for ten minutes and you can go back to work.

If it does not look ordinary, do not rush to fix it, because there may be nothing there to fix. The page I am describing is doing well: on the searches real people type, it sits at position two to four and converts at 8.3%, and the click count on those went from nine to forty-four over the same two months. That is the genuinely good news, and it was completely invisible underneath an impression number that was two-thirds something else.

Separate the two before you judge either. Then report them separately, and keep reporting them separately, because the moment one number can stand in for the other is the moment it can flatter you.

If you have run the check and the shape of it bothers you, send me what you’re seeing and I’ll tell you what I make of it.

Related: Why Your Website Gets Visitors But Not Enquiries · The Dashboard Is a Decoy · Best CMS for Astro