AI Overviews in France: how it works, visibility and measurement in 2026

This article explains how AI Overviews works, what it changes for your traffic and how to track your presence in these answers. Data, sources and a checklist. Written by Louis Choquet, founder of Pharelia, based on the work we carry out for our clients.

AI Overviews is a Google Search feature that displays a summary generated by the Gemini model at the top of the results page. The summary answers the query directly and cites several source pages as clickable links. It is the equivalent of a featured snippet, but written by an LLM using several sources rather than extracted from a single page.

Google selects sources by combining several signals: relevance to the query, content quality measured against E-E-A-T criteria (experience, expertise, authoritativeness and trustworthiness), an extractable page structure, freshness and the authority of the publishing entity. Google's index remains the foundation: a page that is not properly indexed cannot be cited.

In practice, AI Overviews can occupy the entire first screen on mobile. The user reads the answer, sees two or three sources and often decides not to click. The cited sources gain brand visibility and, to a lesser extent, the remaining click traffic.

There are three important differences from a conventional featured snippet. First, several sources contribute to the same summary, whereas a featured snippet extracts content from a single page. Second, Gemini rephrases the content rather than simply copying it: your page can be cited even if the exact wording does not appear on it. Finally, AI Overviews regularly adds interactive features (follow-up questions, comparisons and maps), making it a fully fledged place to find answers rather than a simple text block.

Google announced the launch of AI Overviews and AI Mode in France on July 22, 2026. AI Overviews appears when Google considers a generated answer useful for the query; it does not replace all conventional results.

The French launch has taken place. Use your historical data where it exists, then track your priority pages over comparable periods. If you are starting today, document this first measurement as your starting point; it cannot reconstruct the launch's impact.

AI Mode launched in France at the same time as AI Overviews. It lets users ask follow-up questions and explore the links offered in a conversational answer. Google's announcement describes these two experiences.

AI Overviews accelerates an established trend: zero-click search, where users get their answer without leaving Google. According to initial observations published by Ahrefs and several studies in the SEO press, pages ranked in conventional results may see their click-through rate fall by 15 % to 30 % on informational queries where AI Overviews has been deployed. The figures vary considerably by sector and query type, so caution is needed, but the direction is clear.

At the same time, Google announced several adjustments to preserve traffic to publishers: more visible links in the summary, clickable source cards and interactive formats. The publishing ecosystem remains under pressure, and the actual impact will be measured over six to twelve months after the French rollout.

The way users read is also changing. Someone who gets a concise answer at the top of the page no longer has a reason to compare ten results. They consult one or two sources if they have doubts, then move to a more specific intent (‘compare’, ‘pricing’, ‘reviews’). The discovery funnel becomes shorter, and brand visibility matters much earlier in the journey.

Three practical consequences for brands:

1. An answer-first response written for extraction. Each page addresses an explicit question. The first paragraph under the H2 answers it in 40 to 70 words and stands on its own: a reader (or LLM) who reads only that paragraph should understand the complete answer. The explanation, nuances and examples follow.

2. E-E-A-T demonstrated in practice. In May 2026, Google published its official guide to AI search optimization, formalising AEO/GEO criteria. Cited pages almost always have an identified author with a biography and dedicated page, external sources linked in the body, attributed numerical data, a visible update date, and a consistent entity identity across the page, the organisation and third-party profiles.

3. Schema.org structured data. At a minimum: Organization and WebSite at the root, Article or FAQPage on editorial pages, BreadcrumbList on deeper pages, and Person for authors with sameAs links to LinkedIn and other verifiable profiles. LLMs use these graphs to connect your content to a known entity.

4. Sound technical foundations. Server rendering (AI crawlers struggle with client-side JavaScript), semantic HTML, adequate Core Web Vitals performance, a robots.txt file that explicitly allows Google-Extended, GPTBot, ClaudeBot and PerplexityBot if you want to appear in other answer engines, and a clean sitemap. On llms.txt: Google clarified in May 2026 that Search does not use it — the file may serve other crawlers, but do not rely on it for Google.

5. Freshness and maintenance. Pages more than 18 months old without an update are regularly passed over in favour of recent pages. We add a visible ‘updated on’ block and actually update the content, not just the date.

6. Entity coverage and semantic content clusters. A page is not cited in isolation: it is cited because it belongs to a coherent collection of pages on a topic, connected by descriptive internal links. Covering a topic means explicitly addressing its sub-questions, each on its own page.

7. Off-site entity authority. Consistent mentions of the name, logo and contact details (NAP) in directories, the press and industry profiles. Up-to-date reviews on Google and industry platforms. For B2B brands, inclusion in the sector's reference lists and rankings.

8. Format variety. AI Overviews often draws on sources that combine structured text, comparison tables and numbered lists. A page containing only a long block of prose, without question-based subheadings or a table, is less suitable for citation than an equivalent page organised into scannable sections.

9. Consistency across several pages on the topic. LLMs cross-reference several pages from the same domain to assess its actual expertise. One well-written, isolated page on a topic carries less weight than five consistent, interlinked pages, each addressing a specific aspect. This is precisely the value of a semantic content cluster designed for GEO.

A simplistic claim is circulating: ‘SEO is dead; GEO takes over’. It is false and costly. AI Overviews relies on Google's index. If your website is not properly indexed, is slow or lacks schema, it will appear neither in the blue links nor in the AI summary.

The useful hierarchy is this: technical and on-page SEO remain the foundation. AEO (Answer Engine Optimization) structures content so that answer engines can extract it (featured snippets, AI Overviews, People Also Ask). GEO (Generative Engine Optimization) works on suitability for citation across all generative engines (ChatGPT, Perplexity, Gemini, Claude, Mistral) through entity authority and factual structure.

In practice, the three disciplines share 80 % of the work: indexing, schema, editorial quality and authority. The remaining 20 % varies by engine. The same page can target a Google featured snippet (SEO), a citation in AI Overviews (AEO) and a mention in a ChatGPT or Perplexity answer (GEO). The underlying content is the same; its organisation differs.

The operational consequence: there is no ‘miracle GEO tool’. What can actually be measured is citation share across a set of target prompts, checked at regular intervals. Solutions promising a GEO score without checking actual answers are selling empty promises.

For more detail, see our dedicated note: AEO vs SEO vs GEO: three disciplines, one objective.

On local queries, AI Overviews appears alongside the Google Maps pack but is beginning to replace it for some conversational phrasings (‘best accountant in Lyon’, ‘reliable SEO agency in Lille’). The engine answers with a summary citing a few businesses. If your firm or brand does not appear in that summary, you lose the prospect before they even look at the map.

The competitive challenge is shifting: it is no longer ‘rank first in my city’, but ‘be one of the three names cited when a prospect describes their problem in natural language’. A well-structured local brand, with a clear service page, consistent reviews and an up-to-date Google Business profile, can be cited instead of a larger competitor that AI finds harder to interpret.

For national B2B brands, the same logic applies to sector-specific queries: ‘best property management software’, ‘IT services company specialising in generative AI’, ‘outsourced HR firm for an industrial SME’. AI Overviews cites a few businesses; the others become invisible, even if they are still paying for SEA on the same query.

In our client work, this combines conventional local SEO with restructuring pages for AI extraction. Three public examples: Applewood, M2A and Campus Consulting.

Search Console provides a generative AI performance report for AI Overviews and AI Mode impressions. Google reports a global rollout as of August 31, 2026, with access limitations depending on the property and impression volume. This data is also included in the Web report: do not add it to that report's total.

Compare identical periods, countries, devices and aggregations. The chart is aggregated by property, or by URL when a URL filter is applied; the page table may therefore have a different total. Google's counting rules do not establish automatic double-counting of AI and conventional impressions. A CTR decline calls for an analysis of traffic and queries, not a blanket adjustment. Supplement Search Console with a documented search panel and the sessions and inquiries actually measured on your website.

When we take over work from another agency or an internal team, we almost always encounter the same mistakes. Here they are, in order of severity.

If you want to prepare your website before the full French rollout, here is the order we follow.

Timescales depend on the issue being corrected, recrawling and competition. Check crawling and indexing first, then track the corrected pages using a stable query panel. A launch or technical correction guarantees neither rankings nor AI citations.

On budgets, the mistake is combining SEO tools, a content agency and a technical freelancer without one person coordinating the whole. AI Overviews does not reward an accumulation of tools; it rewards a coherent system. It is better to invest in a team or partner covering audit, foundations, content and measurement from end to end than to spread budgets across three providers who do not communicate.

This is exactly the sequence Pharelia operates for its clients: data audit, foundations, SEO/GEO content production and measurement. Details of our engagement formats are on the offers page.

The questions we are most often asked about AI Overviews. See also the structured FAQ section below.