French publishers v Google AI Overviews: what the Search Console opt-out does not fix

In one sentence: Any website can already remove its content from Google’s AI Overviews and AI Mode in Search Console, but it then loses all the traffic those features send, and nothing changes for AI training; that price of leaving, and the lack of payment, is what five French publisher organisations have now put before the European Commission.

On 29 October 2026, five French trade bodies for publishers, the CPA, FNPS, GESTE, SEPM and SPIIL, filed a complaint with the European Commission about Google’s practices around AI Overviews and AI Mode, according to the FNPS statement (in French). Their joint statement, published by the SPIIL (in French), objects to a French launch on 22 July 2026, a date confirmed by Google France, carried out without consulting publishers and without giving them transparency or real control over how their content is used.

Most coverage stops there. It does not mention that since 31 August 2026 Google has offered a setting that lets any site leave these features. What that setting does, and what it does not, explains what the complaint is really about, for publishers anywhere in the European Economic Area.

A complaint to feed an investigation already under way

The complaint does not start a new case. The five organisations want to contribute to the Commission’s existing investigation with evidence gathered from French publishers since the launch, says the joint statement; it publishes no traffic figures and no detailed list of grievances.

That investigation is case AT.40983. On 9 December 2025, the Commission opened proceedings against Google and Alphabet for a suspected breach of Article 102 of the Treaty on the Functioning of the European Union, the abuse of a dominant position. Its opening decision covers two practices: using publishers’ web content to generate AI Overviews and AI Mode summaries “without offering these web publishers the possibility to refuse Google’s use of such content without losing access to Google Search and without appropriate compensation”, and using YouTube videos to train Google’s generative AI models on the same terms. The decision “covers the entire EEA”, and it stresses that opening proceedings is not a finding of infringement.

One clarification, because the two are often confused: this is competition law, not the Digital Markets Act. Silicon.fr (in French) discusses DMA scenarios, but the December 2025 decision cites only Article 102 and Regulation No 1/2003. The question of whether a chatbot can infringe the press publishers’ right is a separate one, now before the Court of Justice: see the two tests proposed by the CJEU Advocate General.

Google’s four controls, and what each covers

Google points to a control in Search Console. Tested with UK sites from 3 June 2026, it has been available to all websites worldwide since 31 August 2026, according to Google’s announcement and the Search Console help page. It sits alongside three older levers that are easily mixed up.

The Google-Extended documentation is explicit: the token “does not impact a site’s inclusion in Google Search nor is it used as a ranking signal”. It governs training of future Gemini models and grounding in Gemini Apps and Vertex AI, not AI Overviews. The same page says Google intends to support the shared AIPREF training signal in early 2027; Le Latent has explained that signal (in French). Treated like Google-Extended, it will not remove anything from Search either.

Before 31 August, only snippet controls were left. “AI features and your website” points to nosnippet, data-nosnippet, max-snippet and noindex to limit what Search shows from a page: they work page by page, but they also shrink the ordinary result, or remove it with noindex.

What the opt-out does not fix

The Search Console help page sets out the limits, and they match the publishers’ grievances.

• Leaving means losing all AI traffic. An excluded site gets no “traffic or impressions from these features”, and Google warns that content from other sites “will still be available in those features, and it may appear similar to yours”. The answer stays on screen, without you.
• Inclusion is the default for every property.
• Training is not covered. The control “doesn’t affect AI training”: for that, Google refers to Google-Extended, and content “may still be used to power Google Search as a whole”, according to the same help page.
• No page-level setting. The choice is made per Search Console property, a domain or a URL prefix, with child properties inheriting from their parent; the help page describes no setting for a single page.
• No payment is offered, which is precisely what the Commission’s decision examines.

Then there is measurement. GESTE argues that existing tools cannot isolate clicks and impressions from AI Overviews, according to Silicon.fr. Search Console’s generative AI performance report counts impressions, by page, country, device and date; its help page does not mention clicks. Without clicks, a publisher cannot put a figure on what leaving would cost.

What the UK already requires

The most useful comparison comes from London. On 3 June 2026, the Competition and Markets Authority imposed a publisher conduct requirement on Google for its general search services, covering outputs shown to anyone located in the UK. The CMA notice requires:

• controls covering both training and grounding, in Google’s AI search features and in its broader generative AI services;
• for grounding in AI search features, opt-outs at directory level and at page level;
• “clear and detailed metrics on user engagement” with publishers’ content in AI search features (paragraph 7 of the notice);
• clear, accurate attribution, with a clear means for users to reach the source;
• no downranking or other penalty in ordinary search for publishers who use the controls.

The requirement comes into force six months after the 3 June 2026 notice, so in December 2026, and the page-level opt-out nine months after it, in March 2027, according to the same notice. It creates no right to payment, but it gives UK publishers what the worldwide opt-out does not: refusing training in the same move, and measuring engagement. Nothing equivalent exists in the EU so far.

What a site owner can do now

1. Measure before choosing. Open the generative AI performance report and find the pages that appear in AI Overviews.
2. Do not rely on robots.txt to leave AI Overviews. Google-Extended and the shared training token govern training, not Search. Objecting to AI training as a person, rather than a publisher, is a different procedure, covered in how to object across the EU, service by service.
3. Weigh exclusion. The control is in Search Console under Settings, then “Search generative AI”; content is generally excluded within one to two days, according to the help page. For a single sensitive page, nosnippet or data-nosnippet remain the only option, at the cost of the ordinary snippet.

What we do not know yet

The complaint is not public, and the joint statement gives no traffic loss figures. Google had not published a detailed response, according to Silicon.fr. The Commission has set no timetable: opening proceedings only means it will deal with the case as a priority, according to its decision.

AI-generated news sites are another front: in early November 2026, Google removed nine French ones from its Discover feed, as Le Latent reported (in French), and French Wikipedia will block sites flagged by the French outlet Next, as explained in how French Wikipedia will block AI-generated news sites. More EU and French regulation stories are in the Society section.