The ten blue links haven't been the whole page for years. To see what a modern Google result actually contains, we score every result cloro collects for the features it carries. This edition covers 65,945 scored Google SERPs across 41 markets.
The headline this edition: on a fixed set of US queries we have tracked since April, AI Overview presence fell from 96.7% to 89.4%, with the break between June and July. It is the first decline we have measured on this surface, and because the queries and the market are both held constant, it is the one comparison here that isolates Google.
This edition also corrects the last one. The July edition reported a single global figure, and a single global figure is the wrong shape for this question: the markets underneath it range from 61% to 96%. Averaging them produces a number that describes our sample rather than Google, which is the one thing a reader will not assume it means.
Updated August 19, 2026 · first published July 5, 2026 · cloro monitoring corpus · 65,945 scored Google SERPs · 41 markets · 30-day window
US: 96.7% to 89.4%
On a fixed set of US queries tracked since April, AI Overview presence fell about seven points, with the break between June and July. The first decline we have measured on this surface.
A 35-point spread
AI Overview presence runs from 61.0% to 95.9% across the markets we monitor. Any single worldwide number is an average over parts that do not agree, so ask which market a figure came from. Which part of that spread is Google and which is query mix, this data cannot say.
PAA: 91.8%
People Also Ask remains near-universal and is the most stable feature on the page, holding above 90% in every market except Israel (56.1%) and the UK (85.7%).
Ads: 9.5% in the UK
Against 1.3% in the US. Sponsored results vary more between markets than any other feature we score, so an ad-presence rate borrowed from US data is worth very little elsewhere, whether the cause is Google or the queries.
Pooled across every market there is no trend at all, because the pool's composition moves month to month. Hold the market and the query set still and one appears. Below is the US only, restricted to prompts we have tracked continuously since April, so a growing query set cannot be mistaken for a change in Google.
Flat through the spring, then a step down between June and July that has not recovered. About seven points in total.
The unrestricted series looks worse and we are not publishing it as the finding. Across all US prompts the same period reads 96.7% down to 83.5%, thirteen points. But the US prompt set grew from 109 to 524 distinct queries over those months, and the queries added later trigger AI Overviews less often than the originals. Roughly half of the thirteen-point drop is that change in the sample rather than a change in Google. The seven-point figure is the one that survives holding the queries constant.
Share of 65,945 scored Google SERPs carrying each feature, pooled across all 41 markets. Read this as the shape of the page, not as a rate for any one country: the market table below is where the usable numbers are.
On most of these pages position #1 is not the first thing on the screen, so a tracker recording position alone reports a steady #1 and misses the AI Overview and PAA above it.
Markets with at least 500 scored results in the window. This is the cut the last edition did not have: a rate per market, instead of one global average that none of them actually matched. The caveat directly underneath is load-bearing, though, because the table cannot separate the market from the questions asked in it, so read it before comparing one row against another.
| Market | Scored SERPs | AI Overview | People Also Ask | Ads |
|---|---|---|---|---|
| Austria | 588 | 95.9% | 99% | 0% |
| Japan | 7,008 | 90.7% | 96.2% | 1.2% |
| UAE | 3,807 | 87.9% | 97.4% | 2.3% |
| United States | 11,644 | 87.2% | 90.6% | 1.3% |
| Canada | 1,437 | 86% | 93.4% | 4.8% |
| Israel | 3,793 | 85.4% | 56.1% | 1.3% |
| Brazil | 4,107 | 83.7% | 96.5% | 3.3% |
| Denmark | 2,737 | 83.1% | 92.9% | 1.8% |
| Mexico | 4,631 | 82.8% | 97.9% | 3.1% |
| Australia | 1,254 | 78.9% | 96.3% | 5.4% |
| United Kingdom | 1,569 | 75.8% | 85.7% | 9.5% |
| Poland | 3,287 | 71.3% | 95.8% | 2.1% |
| Germany | 8,730 | 67.9% | 94.7% | 1.9% |
| France | 7,833 | 61% | 96.6% | 1.1% |
The honest limit of this table: almost every prompt in this corpus runs in exactly one country. Of the 2,282 distinct prompts behind this edition, 2,278 run in a single market, and no two of the fourteen markets tabled above share one. So a market's row reflects both what Google does there and what the customers tracking that market happen to ask, and this data cannot separate the two.
Israel is the clearest illustration. Its People Also Ask rate reads 56.1% against 90% or better nearly everywhere, which looks like a striking market difference until you read the queries: the Israeli set is 130 consumer-recommendation prompts about things like beer and furniture, while the US set is 524 prompts spanning navigational and local intent. Those are different questions, and different questions carry different features. We are not claiming Israeli SERPs suppress People Also Ask.
The same caution applies to the France and Austria endpoints of the range, and we can now put a number on how much caution. Four travel prompts in this corpus do run identically across 25 other markets, which is a small controlled panel. On those, AI Overview presence spans 91.2% to 98.4%: a seven-point spread where the table above shows thirty-five.
That panel shares no market with the table above and covers a single intent, travel booking, so it does not settle France against Austria. What it does show is the direction and the rough size of the problem: hold the questions still and most of the apparent gap between markets closes.
That hypothesis has since been tested, and it holds. SERP Features by Country runs one fixed 150-query basket across six of these markets. Austria and Germany, 28 points apart in the table above, come out 0.7 points apart on the same questions in the same language, and the result replicates. The spread across those six markets falls from 34.9 points to 18.7, and the ordering inverts: Austria leads here and comes last there. So read this table as a description of what cloro's monitoring sees per market, and treat any market-versus-market comparison in it as roughly half artefact of the query mix, with the ranking unreliable.
The ads column is the most volatile: 9.5% in the UK, 5.4% in Australia, 4.8% in Canada, against 1.3% in the US. These rates stay low in absolute terms because cloro's corpus is weighted toward the conversational and commercial-research queries customers track for AI visibility, which rarely carry ads. Given the confound above, read this as a range worth investigating rather than a league table.
A refreshed number is only interesting if the measurement did not change underneath it. This edition draws on a scoring pipeline that now covers far more results than the July sample did, so before reporting any movement we re-ran the new pipeline over the old edition's exact window.
| Feature | July edition | Re-run, same window |
|---|---|---|
| Organic results | 98.9% | 99.1% |
| People Also Ask | 92.3% | 91.5% |
| AI Overview | 70.2% | 70.4% |
| Ads | 0.2% | 0.2% |
| Local pack | 0.0% | 0.0% |
The pipeline reproduces the July figures to within a point on every feature, so the movement reported above is the world changing rather than our instruments. That same re-run is what exposed the pooling problem: on the July window, the pooled figure was 70.4% while the US alone was 90.5%. The 70% that has been quoted since was never a US number, and was never a Google number either.
Every Google result cloro collects is scored for the features it carries, and presence is binary: the feature rendered at least once on the page. This edition covers the 65,945 scored results captured in the 30 days to 19 August 2026, across 41 markets. The market table reports only those with at least 500 results in the window. The trend section uses the US subset, further restricted to prompts present in both the first and last month, which is why its counts are smaller.
The corpus is customer AI-visibility monitoring, so it over-represents conversational and commercial-research intent relative to all Google traffic. That matters most for ads, shopping cards and the local pack, which are query-type dependent. AI Overview and People Also Ask are the surfaces this corpus is built to measure and the two figures worth carrying into your own planning.
Of the 2,282 distinct prompts behind this edition, 2,278 run in a single country, and no two of the fourteen markets tabled above share one. Market rows are therefore confounded with query mix and are not a controlled comparison. The exception is a four-prompt travel panel that runs across 25 other markets, reported above. The US trend is unaffected either way, because it holds both market and prompt set fixed and varies only time. Removing the confound from the market table needs one fixed basket run across the markets in it, which is the next edition.
One change worth naming: shopping cards read 0.0% in April and 8.5% from early May onward, which is a Google launch showing up in the data rather than anything we altered. The figures on this page are reproducible with mine_serp_features.py --as-of 2026-08-20 in the cloro repository. The date matters: without it the window rolls with the clock and the figures move by about a tenth of a point per day as new results land. They are also registered for automated re-measurement, so this page cannot drift unnoticed the way its first edition did.
cloro. (August 19, 2026). SERP Feature Census: What Actually Shows Up on Google in 2026. cloro Research. https://cloro.dev/research/serp-feature-census/
More studies from cloro's monitoring corpus are in the research index.
cloro returns AI Overviews, People Also Ask, ads, and the organic list as structured JSON. Start with 500 free credits a month.
