Most writing about AI Overviews is either a vendor survey or somebody’s feelings about one client. I wanted to look at properties where I can open Search Console myself and read the number, so I pulled six months of data from seven of them on 20 September 2026.
Seven sites is a small sample and I am not going to pretend otherwise. What seven sites can do is show you the shape of the question, and point you at a report on your own property that most SEO reporting ignores completely.
What did the data show? Four of the seven sites average a position between 6.7 and 8.5, which is the same band of the first page. The one with almost no AI exposure, 1.7% of its impressions, earns 4.3% click-through. The three sitting at 21.8%, 26.4% and 30.2% AI exposure earn 1.0%, 1.1% and 0.7%. The best-ranking site of the four, at position 6.7, earns 1.1%, about a quarter of what the least exposed site earns from a lower position.
One site breaks the pattern in the useful direction, and it is in the chart and the table below along with everything else. The raw CSV is published, including the rows that do not support the story.
7 properties · 6 months to 18 September 2026
Click-through rate against share of impressions inside an AI feature
Each circle is one site. Circle size is total impressions. Tap or hover a circle for its numbers.
Source: Google Search Console, each property’s trailing six months ending 17 or 18 September 2026. AI feature impressions come from the generative AI features report and are a subset of total impressions. Download the raw CSV.
What I actually measured
Two numbers per site, from two reports that are one click apart in Search Console:
- Performance on Search results, set to six months, which gives total clicks, total impressions, average CTR and average position.
- The generative AI features report, linked from just above the query table in that same screen. It reports impressions and clicks for AI Overviews and AI Mode.
Because the AI number is a subset of the total, dividing one by the other gives a single figure I have started calling AI exposure: the share of a site’s search impressions that appeared inside a generated answer rather than a plain list of blue links. That is the horizontal axis in the chart.
Nobody I work with tracks this. Every monthly report I have inherited from another agency tracks clicks, impressions, position and conversions, which means a site can lose half its click-through to AI Overviews while the report keeps saying rankings are stable. They are stable. That is the problem.
The four sites in the same ranking band
Sites A, E, F and G all average a top-10 position over six months. A search engine would call all four of them page-one sites. Here is what they earn:
| Site | Avg position | AI exposure | CTR |
|---|---|---|---|
| A, results and news | 8.5 | 1.7% | 4.3% |
| E, registration help | 8.1 | 21.8% | 1.0% |
| F, registration help | 6.7 | 26.4% | 1.1% |
| G, travel affiliate | 7.5 | 30.2% | 0.7% |
Site F ranks the best of the four, at an average position of 6.7, and earns a quarter of the click-through that site A does at position 8.5. If average position were the thing driving clicks here, that row would be the other way round.
Site A is the useful control rather than the outlier. It publishes results and news that change every day, on queries where the answer people want is a number that Google’s generated answer is reluctant to state. Almost nothing it ranks for triggers an AI feature, 1.7% of impressions, and its click-through looks like search used to look.
The other three are exactly the kind of thing an AI Overview is built to eat. Two of them explain a government registration process. One recommends travel itineraries. Those are questions, and a generated paragraph answers a question without anyone needing to leave the results page. Site G, at 30.2% exposure, has 292,000 impressions and 2,110 clicks to show for them.
The site that does not fit
Site C is the one I would want somebody to point at if I were reading this sceptically, so here it is up front.
Site C averages position 18.2, which is page two. It has 9.8% AI exposure. It earns 2.3% click-through, the second-highest number in the whole set, ahead of three sites that rank ten positions better than it does. A site averaging page two out-clicks a site averaging position 6.7.
My reading, and it is a reading rather than a finding: site C ranks for a narrow set of queries where the searcher has to reach a specific government form to finish the job. The answer is not a paragraph, it is a destination. That is the same reason the second play in my AI Overviews recovery method is to move effort toward queries a generated answer cannot close. Site C is what that looks like when it is already true of your query set.
Site D, this website, is in the data for completeness and proves nothing either way. It has 181 clicks over six months from an average position of 20.2. At that volume the CTR figure is noise, and I have left it in because leaving it out would be choosing the sample after seeing it.
What this does not show
Being straight about the limits is most of what makes a small study worth reading.
- Seven sites is not a sample, it is a handful. These are every property I had direct Search Console access to with enough volume to read. That is convenience, not randomness.
- Six of the seven are India-heavy. AI Overview coverage differs by country and by language, so do not port these numbers to a US or UK site and call them a benchmark.
- This is a snapshot, not a before-and-after. I did not measure these sites before AI Overviews rolled out and compare. Comparing sites to each other cannot separate the effect of AI features from the effect of being in a different niche.
- Average position is an average. A site at 8.5 might rank second for its money queries and fortieth for a long tail, and the query mix inside each site is invisible at this level.
- Google’s AI features report covers AI Overviews and AI Mode together. They are not the same surface and I cannot split them.
- One client property was measured and withheld. It has 1.63M impressions and would have been the largest row in the set. It is not mine to publish, so it is not published, and the pattern in the chart does not depend on it.
If you want to call this seven data points that are consistent with a thing everyone already suspects, that is a fair description. It is still seven more first-party data points than most posts on this subject contain, and the raw file is there for you to disagree with.
The full dataset
All seven rows, sorted by AI exposure. Tap any column heading to sort it yourself.
| Site | Clicks | Impressions | CTR | Avg pos | AI impr. | AI exposure |
|---|---|---|---|---|---|---|
| A, results and news | 13,600 | 313,000 | 4.3% | 8.5 | 5,300 | 1.7% |
| B, SEO services | 1,130 | 77,500 | 1.5% | 12.4 | 3,140 | 4.1% |
| C, registration help | 6,860 | 292,000 | 2.3% | 18.2 | 28,600 | 9.8% |
| D, this site | 181 | 17,900 | 1.0% | 20.2 | 2,430 | 13.6% |
| E, registration help | 7,090 | 680,000 | 1.0% | 8.1 | 148,000 | 21.8% |
| F, registration help | 4,040 | 366,000 | 1.1% | 6.7 | 96,700 | 26.4% |
| G, travel affiliate | 2,110 | 292,000 | 0.7% | 7.5 | 88,100 | 30.2% |
Sites are anonymised by category because most of them belong to other people. Site D is this website. Download the CSV and check my arithmetic.
Three checks worth running on your own site this week
None of this is useful as a statistic. It is useful as a reason to open a report you have probably never opened.
- Work out your own AI exposure. Open Search Console, Performance, Search results, set the range to six months, and note total impressions. Then open the generative AI features report from the link above the query table and note its impressions. Divide. If you are north of 20%, the sites in this data at that level were earning around 1% click-through.
- Compare position against CTR over 16 months. If your average position is flat or improving while CTR falls, you already have the answer, and it is not a penalty. The diagnosis in this post separates that pattern from a ranking loss and from seasonality.
- Split your queries by whether an answer can close them. Questions get eaten. Anything that needs your price, your stock, your form, your location or your login cannot be. That split, not your keyword list, is what decides how exposed your traffic is next year.
If you want the AI side of this handled properly rather than measured once, that is what my AI SEO and AEO audit does: 50 manual searches across ChatGPT and Perplexity on prompts we agree in advance, plus your own first-party AI numbers from Search Console and Bing.
The honest summary
Across these seven properties, average position told me almost nothing about how many people clicked. The share of impressions sitting inside an AI answer told me a lot. A site at position 6.7 with a quarter of its exposure inside AI features earns 1.1%. A site at position 8.5 with almost none earns 4.3%.
That is not a law. It is a variable that belongs on your monthly report and is not on it. Add the column, watch it for two quarters, and you will know whether your niche is one Google is answering for you.
