How Do You Appear in Google AI Overviews in 2026?

Google switched on AI Overviews in France on Wednesday 22 July 2026 at 7am. After two years of regulatory deadlock, the Gemini-generated panel now sits above the blue links on google.fr, across mobile and desktop.

To appear in Google AI Overviews, one thing matters: publishing an extractable passage that fully answers a precise question, on a site whose brand is mentioned somewhere other than its own pages. Everything else follows from that rule.

What surprises most people is that ranking first no longer does the job. Ahrefs analysed 863,000 keywords and found that only 37.9% of URLs cited in AI Overviews appear within the first ten SERP blocks. More than six citations out of ten come from elsewhere.

I’m Florian Zorgnotti, an SEO and GEO consultant. What follows is the protocol I’ve been running with clients since this morning: what to record before coverage scales up, which passages to rewrite first, and why the heaviest lever isn’t technical.

AI Overviews Have Landed in France: What Changed This Morning

Google activated AI Overviews and AI Mode on google.fr on Wednesday 22 July 2026 at 7am. The rollout is progressive, across mobile and desktop alike. Two years after the US launch, France was the last major market without them.

The blocker was neither technical nor linguistic. French already worked in Belgium and Switzerland. The lock was the press publishers’ neighbouring rights dispute, which cost Google two rulings from the French competition authority: 500 million euros in 2021, then 250 million in 2024. A letter sent to publishers in late June lifted that lock.

In practice, a Gemini-generated panel now sits above the blue links. It synthesises several pages and displays links to its sources. Google says it is selective: the summary appears on queries judged sufficiently precise or complex, not on simple searches.

I’ve tracked this file for eighteen months on behalf of clients. What strikes me today is the mismatch: almost every French guide published in recent weeks is written in the future tense, pinned to a 23 September deadline. They went stale this morning. The question is no longer how to prepare, but what to do in the next thirty days.

How Does Google Choose the Sources It Cites in an AI Overview?

Google works in two stages. A classic Search ranking first assembles a pool of candidate pages for the query. Gemini then evaluates that pool and keeps the most extractable, most reliable passages to compose the summary.

This two-tier mechanism explains the most widespread misunderstanding in the market. Ranking first gets you into the pool. It does not get you into the answer.

The figure that invalidates half the guides published in France

Ahrefs analysed 863,000 keywords and close to 4 million URLs cited in AI Overviews. The result: 37.9% of cited URLs appear in the first ten SERP blocks. The remainder splits almost evenly between positions 11 to 100 (31.2%) and pages outside the top 100 (31.0%).

A year earlier, the same measurement returned 76%. Ahrefs states openly that part of the gap comes from improved citation detection in its tooling, and that the two datasets are not strictly comparable. The direction remains unambiguous: more than six citations in ten now come from pages absent from page one.

In my view, this is the most important methodological reversal of the year, and almost nobody on the French side has absorbed it. As long as you think in terms of moving from position 6 to position 2 to earn a citation, you’re working the wrong lever. The right question is: across how many sub-queries on this topic am I present anywhere at all?

Why query fan-out changes your action plan

Gemini doesn’t simply process your query. It fans it out into sub-questions, then hunts for sources answering each one. A page ranking 40th on the head query can be cited because it answers a sub-question generated by that mechanism perfectly.

You therefore don’t win by optimising one page for one keyword. You win by covering a complete cluster, where each page handles one sub-question in a self-contained, complete way.

Which Signal Correlates Most Strongly With AI Overview Visibility?

Off-site brand mentions. Ahrefs cross-referenced data from 75,000 brands against mentions recorded in AI Overview responses. The correlation ranking is clear:

Read that last row carefully. Link building, the backbone of SEO for twenty years, comes last, at one third the strength of plain brand mentions. An unlinked mention in a trade publication outweighs a purchased backlink.

This finding isn’t isolated. It matches what large-scale citation analyses show: generative engines draw heavily on third-party sources rather than on brands’ own sites.

What this means for your budget

• A pure link-building budget yields poorly on AI citation. Reallocate part of it toward PR and industry speaking opportunities.
• Community and video platforms matter: YouTube ranks among the most-cited domains across verticals.
• A consistent business listing, reviews and an active LinkedIn presence feed branded search volume, the third signal in the ranking.

Do You Need Schema.org Markup to Be Cited by AI Overviews?

Markup helps machines parse your page, but no public study demonstrates that it raises your citation rate. An Ahrefs analysis of pages that added JSON-LD recorded no statistically significant change in AI citations.

This is where I part company with the French market consensus. Nearly every guide ranking for this query makes schema the centrepiece, with FAQPage and HowTo headlining. That recommendation is comfortable because it is easy to sell and easy to execute. It is not supported by the available data.

My take: do the markup, because it costs little, serves your classic rich results and can’t hurt. But don’t count on it as a citation lever, and above all don’t let a vendor bill you for a GEO engagement that amounts to dropping in JSON-LD. The cost-to-impact ratio is poor next to a mentions campaign.

What actually matters on the technical side

1. URL accessibility: the page must be reachable, unblocked, not noindexed. This is the absolute prerequisite.
2. Rendering without JavaScript: if your main content only exists after client-side execution, you drop out of the pool.
3. Preview control: directives limiting the snippet mechanically reduce your extractability.
4. Question-answer match: your passage’s phrasing should mirror the query’s.

How Should You Structure a Page to Be Extractable by Gemini?

A citable passage is a self-contained 40-to-80-word unit that fully answers one question without depending on the preceding paragraph. That is the working definition of a chunk.

You can test the rule easily. Take any paragraph from your page, strip it of its context, send it to someone unfamiliar with the topic. If they grasp the complete answer, the chunk holds. If they ask what this is about, Gemini will ask the same thing and look elsewhere.

The format that works

• A question heading using the phrasing real users type.
• A declarative opening sentence that answers in full, with no wind-up.
• Development with a dated figure, a named source, a concrete example.
• Explicit entity: name your brand inside the passage, not just in the site footer.

On length, one data point is worth knowing: Ahrefs’ analysis across a billion data points finds near-zero correlation between content length and citation probability, with a coefficient of 0.04. More than half of AI Overview citations point to pages under 1,000 words. Writing long doesn’t get you cited. Writing tight does.

What Protocol Should You Run in the First 30 Days?

Because the French rollout is progressive, you have a short window while competitors are still watching rather than acting. Here is the sequence I’ve been applying with clients since this morning.

Week 1: establish the baseline

You can never prove a gain without a starting point. Before coverage scales up, record the presence or absence of an AI Overview on your twenty most strategic queries. For each one note: AIO shown or not, sources cited, your presence or absence. That table is your reference for the next six months.

Week 2: isolate the queries at risk

In Google Search Console, filter the last four weeks and look for the characteristic pattern: impressions flat or rising, clicks and CTR falling. That scissor shape signals a panel capturing attention ahead of your link. Those queries become your priority queue.

Week 3: rewrite the first twenty passages

On the identified pages, don’t rebuild the article. Rewrite only the opening paragraph of each section so that it answers completely and immediately. This is the highest effort-to-result intervention in the whole protocol.

Week 4: open the mentions workstream

List ten industry outlets where your brand should be cited and isn’t: trade media, podcasts, specialist directories, interviews, partner case studies. An unlinked mention does the job. This workstream is slow, but it carries the most weight in the measured correlations.

How Do You Measure Your Presence in AI Overviews?

Three measurement tiers coexist, with very different costs and reliability.

Manual checks remain the most reliable and cheapest method. Private browsing, localised query, twenty target keywords, once a week. The drawback is that it doesn’t scale and depends on your discipline.

Google Search Console gives you the indirect signal described above, free, across all your queries. The data stays partial but it is official.

Tracking tools automate the checks and keep the history. That’s the only way to obtain an aggregated share of voice, which is the honest GEO metric. An isolated position on one query on one day means nothing: generative answers vary from run to run. Only aggregation across many checks produces a usable signal.

That’s the logic I built Cockpyt AI on with Laurent Séjourné: weekly measurement across ChatGPT, Gemini and Perplexity, share of voice against competitors, a Google Analytics 4 connection linking visibility to real traffic, and a monthly GEO audit that outputs a prioritised action plan rather than just a dashboard. Flat price at 29 euros a month billed annually, 14-day trial with no credit card.

The Mistakes That Keep You Out of AI Overviews

Five behaviours come up again and again in the audits I run.

• Building suspense. Three paragraphs of preamble before the answer drop you from selection, even from position one.
• Paraphrasing the existing SERP. If your page adds nothing Gemini can’t already find elsewhere, it has no reason to prefer you.
• Relying on markup alone. Dropping in JSON-LD and waiting for citations is a strategy with no demonstrated basis.
• Publishing without a visible date. Cited content measures as meaningfully fresher than classic organic results.
• Treating AI visibility as SEO. SEO remains the foundation; GEO adds a citation layer that obeys its own signals.

FAQ

Does ranking first on Google guarantee appearing in the AI Overview?

No. Organic ranking gets you into the pool of candidate pages; it doesn’t determine the final selection. Ahrefs’ 2026 measurements put the share of cited URLs sitting in the first ten blocks at around 38%. How you phrase your answer weighs more than your raw rank.

Do AI Overviews appear on every query in France?

No. The 22 July 2026 rollout is progressive, meaning two users may see different results on the same query for several weeks. Google says it reserves display for queries that are sufficiently precise or complex. Direct transactional and navigational searches remain largely unaffected.

Can a new site be cited in AI Overviews?

Yes, provided you target precise sub-questions rather than generic queries. Since most citations come from pages outside the top 10, a young but highly specialised site covering a narrow scope has a genuine window. On broad, competitive queries, expect several months of sustained effort.

Can you opt out of feeding AI Overviews?

Press publishers have an opt-out negotiated under neighbouring rights. That choice isn’t neutral: opting out doesn’t guarantee keeping your audience, since the summaries divert clicks regardless of your position. For a standard business site, the question mainly arises through preview directives, at the cost of reduced extractability.

What is the difference between AI Overview and AI Mode?

An AI Overview is a panel displayed passively above the classic results, without any user action. AI Mode is a dedicated tab opening a conversational interface, close to Gemini or Perplexity, where the user chains questions. Both rely on Gemini but behave differently and should be tracked separately.

How long before you see an effect?

On an established domain, a reworked page can be cited within weeks of updating. On a recent domain, expect several months. The brand mentions workstream, which carries the most weight in the correlations, produces results over a three-to-six-month horizon.

Does SEO become useless with AI Overviews?

No. Google retains the bulk of the French search market and a substantial share of queries display no AI panel at all. SEO remains the foundation that gets you into the pool of candidate sources. GEO adds the layer deciding whether you get cited once you’re in that pool.

Sources

• Johan Sellitto, “Google lance officiellement les AI Overviews en France !”, Abondance, 22 July 2026 — abondance.com
• Louise Linehan, “Update: 38% of AI Overview Citations Pull From Top 10 Pages”, Ahrefs, 2026 — ahrefs.com
• Ahrefs, “AI Overview Brand Correlation Study” (75,000 brands), 2026 — ahrefs.com
• Ahrefs, “Search Rankings and AI Citations”, 21 July 2025 — ahrefs.com
• “La recherche Google change de visage dès aujourd’hui, avec AI Overviews et AI Mode”, Clubic, 22 July 2026