Since the summer I have been getting the same question almost weekly. My traffic is down, what is happening.
The quick answer, the one you read everywhere right now, is AI Overviews. Sometimes that is true. Often it is more complicated. And in a fair number of cases the drop is real and costs you nothing at all.
Before you sign anything with anyone, run the diagnosis. It takes ten minutes.
1. What actually happened
On 22 July 2026, Google switched on AI Overviews and AI Mode in France, two years after the global launch. According to an analysis published two days later by a specialist agency, the rollout is progressive and is due to cover the full range of queries before 23 September 2026.
France was the last major market to get it. The feature had been working in French in Belgium and Switzerland for months, which means there is data available before it even reaches you.
That data says the following. In already-exposed markets, the same analysis reports drops from 15 to 60% depending on sector, with extreme cases at 79% for sites heavily dependent on informational queries. An Ahrefs 2026 study, relayed in August by a French analysis, measures a click fall of around 58% on average where an AI Overview appears, and around 44% on French-language queries.
That last figure deserves perspective. The same Ahrefs source measured a 34.5% drop across 300,000 keywords at the end of 2025. The gap between the two measurements in under a year suggests the effect is deepening rather than settling.
On frequency, Similarweb puts AI Overview presence at roughly 43% of Google searches in 2026. And the links shown inside the box generate under 1% of visits, according to the French analysis already cited. Being cited there brings almost nobody directly.
2. The diagnosis, in ten minutes
Open Google Search Console, Performance report, last 3 months compared against the previous period.
Look at impressions and clicks separately. That is the whole diagnosis.
Case 1: impressions stable or up, clicks down. That is the AI Overview signature. You still appear, fewer people click. Your rankings have not moved, the results page has changed.
Case 2: impressions down and clicks down, in similar proportion. This is not AI Overviews. You lost positions. The cause is elsewhere: an algorithm update, a technical problem, a competitor who did the work, content that aged. The fix is completely different, and buying GEO will not solve it.
Case 3: both down, but clicks much faster. Both effects are stacking. Deal with the ranking loss first, it is the older problem.
Then filter by query type. Almost nobody does this step, and it is the one that decides whether you have a real problem.
Split your queries into three buckets:
- Branded queries, the ones containing your name
- Transactional queries, the ones with buying intent: price, cost, quote, booking, your service plus your city
- Informational queries, the ones starting with how, why, what is
Several French and international analyses converge on the same finding in 2026: branded and transactional queries hold up markedly better than informational ones.
3. What the result means for your business
Here is the part the panic articles leave out.
If your drop is concentrated on informational queries and your commercial ones are intact, you probably have not lost anything vital. You lost blog readers. That is not pleasant, but those visitors were not buying. A restaurant losing traffic on “how to choose a wine” while holding its position on “restaurant + its city” is in a healthy position.
I say this because it happened on my own site. Technical guides pulling tens of thousands of impressions and zero clients. The day that traffic falls, revenue does not move.
If your commercial queries are affected, that is serious. Your pipeline is contracting and you need to act.
And if traffic falls while conversions hold, look closely before panicking. The Valiuz GEO barometer of July 2026 reports that visitors arriving after reading an AI summary show around 23% higher purchase intent. Fewer people, better qualified, is not a bad trade.
4. What works, and what does not
If the diagnosis shows a real problem, here is the current state of what we know.
What works. Being cited as a source. A 2026 international analysis notes that AI Overviews typically synthesise three to five sources per answer, which means you do not need to be the dominant authority in your sector to appear. You need to be useful on one specific point: a figure, a method, a piece of data nobody else publishes.
In practice that means pages answering directly under the heading rather than after three paragraphs of preamble, clean structured data, content served as HTML rather than generated by JavaScript, and specific facts a model can lift. I documented that work end to end in a case study.
What does not work, or no longer does. Planting posts on Reddit to get cited. According to Promptwatch data reported in September 2026, Reddit's share of ChatGPT citations fell from around 3.8% to around 0.5% on 14 August. The suggested explanation is that platforms are responding to manipulation attempts. The people selling that technique three months ago have stopped.
Producing more informational articles does not work either. That is precisely the content AI Overviews absorb.
5. Should you do anything, and when
My view, and it works against my own commercial interest.
If your commercial queries are holding, wait. The French rollout is not finished, the numbers are still moving, and committing a monthly budget to a problem you do not have is a bad decision. Run the diagnosis again in six weeks.
If your commercial queries are falling, do not wait. Every month counts here, because lost positions come back slower than they go.
Either way, run the manual test. Open ChatGPT in a private window and ask for a recommendation for a business like yours, in your city. See who comes up. That tells you where you stand on the other half of this, the half Search Console does not measure.
That is the work I do on a monthly basis, and I covered the market rates in a separate article on pricing. The detail of the service is on my SEO and GEO services page.
Month one is exactly this diagnosis. If your sector is not affected yet, I say so and we do not start.
Sources
Figures gathered on 17 September 2026 from published market analyses: a French agency analysis from July 2026 on the France rollout, an Ahrefs 2026 study relayed in August 2026, Ahrefs December 2025 data across 300,000 keywords, Pew Research 2025, Similarweb 2026, the Valiuz GEO barometer of July 2026, and Promptwatch data reported in September 2026. These measurements vary by sector, country, device and query type. Check your own data before deciding.
Want to know which case you are in? Send me your site address and a screenshot of your Search Console over three months. I will tell you which of the three applies and whether it is worth acting on. Write to me here, I reply within 24 hours. The detail of the service: SEO and GEO services.
