Google AI Overviews: you keep your ranking, you lose your visitors
Every month, Fast Growth Advisors puts the same forty B2B buying queries to Google on the UK market, in English. We record what Google AI Overviews write, and whom they cite. That block appears above the results and cites only two to five sources. Our query set is frozen, and every reading can be rerun in one click.
Latest reading: 1 October 2026, 40 queries, UK market.
Why do you lose 23.1% of clicks without your report showing it?
Ahrefs tracked 963 domains in Search Console, twenty-eight days before the French launch and nine days after. Across that panel, the most exposed sites lost 23.1% of their click-through rate. Their impressions did not move.
Stable rankings, stable visibility, falling traffic. And the cost per lead climbs without the report saying why.
A team steering by impressions then buys traffic to offset a loss it cannot see. At Fast Growth Advisors we therefore treat this as a management question more than an SEO one: the top of the funnel thins out while the ranking indicators stay green.
In France, an AI Overview appears on more than half of all searches, according to SE Ranking (data from 7 and 8 August 2026).
Google rolled the feature out in France on 22 July 2026. Two weeks later, half of all searches were already affected: a visibility measurement taken before that date describes a market without AI Overviews, and says nothing about the situation since.
Read the full analysis of the −23.1% →
Sources: Ahrefs, “1 mois d’AI Overviews en France et déjà −23,1 % de taux de clics”, 19 August 2026, ahrefs.com. SE Ranking, “AI Overviews Already Appear on Over 52% of French Searches”, data collected 7 and 8 August 2026, seranking.com.
What do AI Overviews answer to 40 B2B buying queries, month after month?
Four sectors, ten queries each: human resources, artificial intelligence, security, data and analytics. In each sector, five comparison queries and five needs queries, written by hand and measured before entering the set. Readings take place on the first of each month.
Latest reading: 1 October 2026, 40 queries, UK market.
Since September 2026: video −7.8 points, websites +5.0 points, forums +1.9 points, public sector +0.9 points.
Who Google cites, by type of source and by month
- September 2026
- October 2026
Websites
Video
Forums
Public sector
The ten domains Google cites most, October 2026
| Rank | Domain | Citations |
|---|---|---|
| 1 | youtube.com | 21 |
| 2 | reddit.com | 20 |
| 3 | gartner.com | 9 |
| 4 | linkedin.com new | 6 |
| 4 | zapier.com | 6 |
| 6 | applytosupply.digitalmarketplace.service.gov.uk new | 5 |
| 7 | microsoft.com new | 4 |
| 7 | quora.com new | 4 |
| 9 | capterra.co.uk new | 3 |
| 9 | facebook.com new | 3 |
New in the top ten: linkedin.com, applytosupply.digitalmarketplace.service.gov.uk, microsoft.com, quora.com, capterra.co.uk, facebook.com. Out of the top ten: smallbusiness.co.uk, salesforce.com, ibm.com, pcmag.com, uk.pcmag.com, accuknox.com.
AI Overviews by sector, October 2026
| Sector | Queries | With an AI Overview |
|---|---|---|
| Human resources | 10 | 10 |
| Artificial intelligence | 10 | 10 |
| Security | 10 | 10 |
| Data and analytics | 10 | 10 |
Every month measured
| Month | With an AI Overview | Sources cited | Distinct domains | Websites | Video | Forums | Public sector | Status |
|---|---|---|---|---|---|---|---|---|
| September 2026 | 40 of 40 | 296 | 190 | 79.1% | 14.5% | 5.7% | 0.7% | complete |
| October 2026 | 40 of 40 | 314 | 226 | 84.1% | 6.7% | 7.6% | 1.6% | complete |
Since its first edition, the UK series has kept the same forty queries. Full data of the September 2026 UK edition →
A French series runs on the same four sectors. French barometer, in French →
On the ranking side, in our French reading of 1 September 2026, 96% of answers cited at least one of the top three results. But 30% of the sources cited were not in the top. A good position no longer guarantees the visit, and is no longer enough to get cited: a page outside the top can take the source slot if it answers the question better.
Why keep the same queries from one month to the next? A barometer is only worth its comparison over time.
Change a query, and the gap becomes unreadable: nothing tells you any more whether it comes from the market or from the change of query.
Every week, the AI answers barometer reads Google next to four assistants, to compare them with one another. Here the reading is monthly and covers Google alone. Both series are measured separately. Their figures never mix.
Is the software Google recommends among its sources?
On the query comparatif logiciel sirh pme, Google writes a comparison and puts Lucca first. Its five sources are ringover.fr, culture-rh.com, apogea.fr, quel-sirh.fr and bakertilly.fr. None of them belongs to a vendor it recommends.
So Google describes your product in other people’s words. You cannot correct them from your own site.
This case comes up again elsewhere in the September 2026 edition. On comparatif outils sales enablement, Google ranks Highspot among the references and cites four French vendors as sources, none of them the one it names.
For a vendor, the consequence is concrete. Your buyer reads a comparison written from sources you do not control, and your arguments only appear in it if a third party picked them up. Your relationship with those third parties becomes a visibility asset.
Sometimes it goes the other way round.
With outil gestion des vulnerabilites saas, Google describes families of scanners without ranking them, and its sources are cyberwatch.fr, paloaltonetworks.fr, fortinet.com, xygeni.io and getapp.fr. Vendor pages become the source: there, editorial work pays off.
The engine still has to read your page.
At Fast Growth Advisors, our measurements on three B2B vendor sites show that, in the worst case, an extractor keeps only 23% of the text served.
ChatGPT visibility, what extractors keep →
Perplexity visibility, the engine that cites the most →
Read the September 2026 French edition, in French: three cases, the full data →
Which of the three situations should worry you?
For each of the 40 queries, the Fast Growth Advisors barometer records one of three states, never merged into one another.
| What Google does | What we write | Should you act? |
|---|---|---|
| It answers and cites your page | Google cites you | No. Protect it |
| It answers without citing your page | Google answers without citing you | Yes |
| It does not answer | Google does not answer this question | No. Nothing to measure |
Most monitoring tools confuse the last two states. They announce an invisibility that nothing has observed.
A question without an AI Overview measures nothing. Google alone chooses the queries it writes an answer for, and that choice says nothing about your site. Counting it as a failure would mean inventing a fault; so we apply no penalty to a question left unanswered.
That leaves the second state, the only one that calls for a budget.
You can run this reading yourself. Ask the questions of your category, in France and in French, record the state of each one, then start again a few days later, dating every pass and keeping the search link. A single reading is not enough, because the answer changes from one day to the next.
Where should the budget go, by type of question?
On a comparison query, the source slot mostly goes to third parties: media, comparison sites, integrators. Your own pages rarely make it. The useful budget is press relations and the sites that rank you.
With a needs query, Google cites vendor pages, and yours can become the source if its content answers the question and survives extraction intact.
Two families, so two budgets.
Yet most companies do the opposite: they produce their own content to win the comparisons, and buy press relations for the rest.
One example fits in two lines. Lucca, recommended on comparatif logiciel sirh pme, is not among its sources. On meilleur logiciel note de frais, however, lucca.fr is cited although it does not appear in the top eight results.
Same vendor, and two opposite outcomes depending only on the question asked.
Across the 40 queries, the Fast Growth Advisors barometer therefore never publishes a single score: adding the two families together would erase the information you need to decide.
A source slot once won is not necessarily kept. Replayed eleven times in a row, the query comparatif logiciel sirh pme returned sources that overlapped by 62% at the median, and only three domains were present in all eleven readings.
Measure the questions in your category, for free →
Scope: Google AI Overviews, French market, French language. Not ChatGPT, not Perplexity, and not the AI-driven browsers that browse the web on the user’s behalf.
Two Fast Growth Advisors white papers detail the method: what survives extraction, and how to count bots whose declared agent can be forged. The recoverable message → · Counting bots →
Frequently asked questions about Google AI Overviews
What is a Google AI Overview?
The answer block Google writes itself, above the results.
Instead of ten links, it gives the answer and cites two to five sources. In France, it appears on more than half of all searches, according to SE Ranking data collected on 7 and 8 August 2026.
Do Google AI Overviews reduce traffic?
On the queries concerned, yes.
Ahrefs tracked 963 domains in Search Console, 28 days before the French launch and 9 days after. The most exposed sites lost 23.1% of their click-through rate. Their impressions did not fall, which makes the loss invisible in a report that tracks rankings and visibility.
How do you get cited in Google AI Overviews?
First, by ranking well.
In the Fast Growth Advisors French reading of 1 September 2026, 96% of answers cited at least one of the top three organic results. Then by choosing your questions. On a comparison, Google mostly cites third parties, media and comparison sites; on a needs question, it cites vendor pages, and yours can become the source. One condition is rarely checked: the engine has to read the page. In the worst case we measured, an extractor keeps only 23% of the text served.
Google does not answer my question: am I invisible?
No.
A question without an AI Overview measures nothing and says nothing about your site. It is an absence of measurement, and not an absence of citation. We keep the two apart in every reading and apply no penalty to a question left unanswered.
Should you still invest in SEO with Google AI Overviews?
Yes, but not for the same reason.
Our French reading of 1 September 2026 found that 96% of answers cited at least one of the top three organic results: ranking still decides whom Google cites. It no longer decides who gets the visit, since the most exposed sites lost 23.1% of their clicks without losing a single position. And 30% of the sources cited were not in the top, so ranking well is not enough either.
The rest of this series:
Measure AI visibility, the method and the instruments →