AI Overviews in France: staying visible when position 1 is no longer enough

Since 22 July 2026, AI Overviews and AI Mode have been live in France. On informational queries, position 1 now earns fewer clicks. This guide sorts it out: sourced figures, levers that genuinely help you get cited by Google’s AI answers, practices to ignore and a method for measuring the impact on your site.

What happened on 22 July 2026

On 22 July 2026, Google France announced the launch of AI Overviews and AI Mode on mobile, desktop and in the Android and iOS apps. France joins a worldwide rollout that began in May 2024 in the United States.

Two features, two logics

Key common point: according to Google, these features rely on Search’s ranking and quality systems. A page must be indexed and eligible for a snippet to be cited.

Why France waited two years

The delay comes down to the neighbouring rights case. France’s competition authority (Autorité de la concurrence) fined Google €500 million in 2021, then €250 million in March 2024; the commitments Google made in 2022 remain binding until 2027. On 11 August 2026, the Alliance de la presse d’information générale referred the matter to the authority, denouncing a rollout without dedicated remuneration. The outcome is not known at this time. For a business website, the practical consequence is simple: AI Overviews are here, and nothing suggests they will be withdrawn.

The opt-out setting in Search Console

Announced on 3 June 2026 and available to all sites since 31 August 2026, a Search Console setting lets you stop appearing in AI Overviews, AI Mode and Discover AI Overviews. Google specifies that it is not used as a ranking signal for classic results. We will see below why few sites are interested in it.

How many clicks are you really losing? What the studies say

No public study yet covers France. The data comes mostly from the United States: it gives an order of magnitude, not a forecast for your site.

What these figures do not say

• The drop is not due to AI Overviews alone. At Ahrefs, the CTR of informational queries without an AI Overview has also almost halved; Seer measures −46% over a year. Part of the questions is going to conversational assistants and social networks.
• Informational queries are the most exposed. These studies do not measure your transactional pages (quote, product, contact): check your own data before drawing conclusions.
• Being cited limits the damage. The gap Seer found (0.70% against 0.52%) is the most useful figure: the goal is no longer just to be first, but to be the source selected.

How Google chooses the sources for AI answers

• Query fan-out. The system launches several related searches and assembles the answer. A page can be cited for a query it does not target, because it answers a sub-question precisely.
• Same quality systems. A badly indexed, slow or unhelpful page has little chance of being cited.
• Snippet eligibility. The nosnippet, data-nosnippet or max-snippet directives also limit your presence in AI answers.

In other words, “GEO” (generative engine optimization) applied to Google is not a discipline separate from SEO. It is core SEO, with a stricter criterion: bringing what other sources do not.

What really counts for being cited

Google’s optimisation guide, updated on 10 July 2026, insists on “non-interchangeable” content: a distinctive point of view, real expertise, clear organisation.

1. Non-interchangeable content

An AI answer summarises the consensus. If your page says the same thing as ten others, nothing justifies citing it. What makes the difference: your own data, lessons learned from experience, a comparison built by you, an argued position.

We measured this in the iones lab EXP-03 report: 48 articles published over 10 weeks on a new domain, following four protocols of 12 articles each.

No manual penalty was observed. The report’s conclusion: “Google doesn’t need to detect AI. It only needs to detect the absence of value.” AI speeds things up (−65% of time between C and D), it does not replace expertise. This is the method we apply in content production and integration.

2. Visible proof and expertise

• An identified author, with a page presenting their experience.
• Primary sources linked rather than unlinked “according to a study” claims.
• Dated figures and a genuine update date.
• Concrete examples: screenshots, commands, configuration extracts.

3. A clear structure

One H2 per question addressed, a direct answer in the first sentences, then the detail, in semantic HTML. This is not “chunking for AI”, which Google considers unnecessary: it is clarity.

4. Structured data: useful, not magic

According to Google, it is not required for generative AI features and there is no special schema.org markup to appear in them. It remains useful for rich results and for removing ambiguity (organisation, author, product, breadcrumb), provided it matches the visible content exactly.

5. Performance, rendering and crawling

A page that Googlebot cannot crawl or render will not be cited. To check:

• main content present in the rendered HTML (watch out for JavaScript applications without server-side rendering);
• server response time and Core Web Vitals that are acceptable on mobile;
• no unintended block in robots.txt, and no noindex inherited from a pre-production environment;
• consistent canonicals.

These are topics our web development team handles every day: server-side rendering, caching, WordPress, Drupal or Next.js templates.

“Be citable” checklist

• [ ] Each section answers a precise question from its first lines.
• [ ] The page contains at least one element that cannot be found elsewhere.
• [ ] Sources linked and dated, author identified.
• [ ] Page indexed (URL inspection) with no unintended nosnippet.
• [ ] Content present in the rendered HTML.
• [ ] Structured data consistent with the visible content.

What is just folklore

Set against the official documentation, several “AI optimisation” recipes do not hold up for Google.

How to measure the impact on your site

Google includes clicks from AI features in the standard Performance report (“Web” type), without isolating them. The generative AI report provides impressions by page, country, device and date, but neither clicks nor queries. Here is the method we follow during a technical SEO audit.

Step 1: isolate your informational queries

In the Performance report, filter queries with a regular expression (RE2 syntax):

^(comment|pourquoi|quel|quelle|quels|quelles|qu.est|combien|quand|est-ce|que faire|définition|différence)\b

Also create the inverse filter and exclude your brand.

Step 2: compare at constant position

Compare two periods of the same length (for example 1 June – 21 July and 23 July – 12 September 2026), controlling for seasonality with 2025. Export and track CTR by position band (1, 2–3, 4–10). A CTR drop at stable position on informational queries, with no equivalent drop elsewhere, is the typical signature of AI Overviews.

Step 3: cross-check with the generative AI report

• Page cited, CTR falling: you are a source; give people a reason to click (tool, detailed comparison, practical case).
• Page not cited, CTR falling: rewrite priority.
• Page cited, CTR stable: leave it alone.

Step 4: track the other answer engines

Third-party assistants generate identifiable referral traffic. In GA4, create a channel group on the source:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

Server logs show which bots crawl your pages:

grep -E "OAI-SearchBot|ChatGPT-User|PerplexityBot|ClaudeBot|Claude-User|Googlebot" access.log \
  | awk '{print $7}' | sort | uniq -c | sort -rn | head -30

User agents can be spoofed: validate Googlebot by reverse DNS lookup.

The minimal dashboard

Google claims that clicks from pages with an AI Overview are “of higher quality”; check this against your conversions rather than assuming it.

Should you opt out of AI Overviews?

Opting out is all or nothing: no more impressions or visits from AI Overviews, AI Mode and Discover AI, while your classic results remain displayed below the AI Overview. It can be argued for a publisher living off advertising audience. For a B2B company, an agency or an e-commerce business, it most often amounts to leaving the citation to a competitor without winning the clicks back.

Frequently asked questions

AI Overviews: moving from position 1 to the cited source

AI Overviews do not make SEO obsolete; they make average content useless. What remains in your hands: producing pages that AI answers need to cite, keeping an impeccable technical foundation and measuring closely to tell the effect of AI Overviews from a real site problem.

Want to know which pages are really affected and which to rework? We run targeted audits and produce expert content, including white label for agencies. Talk to the iones team: reply within 48 to 72 hours.