The Thesis

France just turned “AI summaries hurt publishers” into a formal competition-law test case, in front of a regulator that has already fined Google twice for this exact pattern.

Google shipped the feature in France on July 22. The question now is not whether it owed a negotiation, but when the clock started.

The Signal

Three developments worth watching this week.

Signal 01
1. APIG filed, and the traffic numbers are the complaint’s spine.

What happened. On August 11, the Alliance de la Presse d’Information Générale (APIG), which represents nearly 300 French daily newspapers, asked France’s Autorité de la concurrence to intervene against Google’s AI Overviews and AI Mode. Google had deployed both features in France on July 22 without a license covering that use. The complaint rests on a specific figure: France’s audiovisual and digital regulator, Arcom, estimates that AI Overviews have driven a 33% to 38% decline in referral traffic to publisher sites in European markets where the feature is already live.

Why it matters. The filing attaches a named, quantified harm to a standing legal obligation. France’s 2019 related-rights law gave press publishers and news agencies protections over certain digital uses of their content, and in June 2022 the Autorité made Google’s publisher commitments legally binding, with terms that run through 2027 and require good-faith negotiation and a compensation proposal within three months of any new negotiation. Those commitments extend to any new Google service that makes fresh use of protected press content. APIG argues that AI Overviews are precisely that kind of new service, and that Google shipped the feature without ever triggering the clause.

Second-order effect. If the Autorité agrees that generative summarization is a distinct “use” requiring its own negotiation, every subsequent AI search feature that Google, or any platform, ships in France will inherit a licensing obligation that attaches before launch. The complaint stage disappears from the sequence.

Signal 02
2. The regulator has already fined Google twice for this exact pattern, and just ordered Meta to the table.

What happened. France’s competition authority fined Google €500 million in 2021 for failing to comply with its orders to negotiate with publishers in good faith; it fined Google another €250 million in March 2024 after finding that the company had breached four of the seven commitments in that settlement, including by training its AI model, then called Bard and since renamed Gemini, on publisher and news-agency content without notifying anyone. The same authority turned to Meta on July 8 of this year, ordering the company through interim measures to resume good-faith negotiations with French press organizations, APIG among the complainants, and to hand over within fifteen days the financial information publishers need to assess its payment offers. APIG’s new filing asks for the same treatment.

Why it matters. This regulator has shown that it will impose interim measures and fines on major platforms in this dispute category, and it did so five weeks ago on a complaint APIG itself helped bring. Against that record, the Google filing reads as the next entry in an established enforcement sequence.

Second-order effect. A third strike against Google under the same framework, aimed this time at generative summarization itself, would signal that regulators treat AI summaries of licensed content as a compensable use in their own right, separate from model training and separate from search indexing.

Signal 03
3. The French complaint joins a wider European front.

What happened. The European Commission opened a formal antitrust investigation in December 2025 into Google’s use of publisher content for AI purposes, and that investigation remains open. Italy’s communications authority, AGCOM, added its own pressure on April 30, 2026, asking the Commission to evaluate AI Overviews and AI Mode under the Digital Services Act after a complaint from FIEG, the Italian newspaper publishers’ federation. In the UK, the Competition and Markets Authority imposed a binding conduct requirement on Google on June 3, 2026, the first of its kind anywhere, and Google announced a Search Console setting that same day allowing site owners to exclude their content from AI Overviews and AI Mode at the domain level. The setting took effect for an initial group of UK sites on June 17 and began reaching site owners elsewhere in late July.

Why it matters. Google is now managing AI search licensing as a multi-jurisdiction compliance problem. The opt-out tool built under UK pressure is the same mechanism French publishers are being offered as their alternative to a license, and that substitution sits at the center of APIG’s objection, because opting out costs a publisher its search visibility entirely.

Second-order effect. Whatever remedy France reaches will become the template for Italy, the wider EU, and the UK, which could compress years of separate national proceedings into one fast-moving channel of precedent.

The Playbook

Four questions to ask before the next AI-licensing or content-strategy conversation.

Step 01
Map the exposure.

For any organization whose traffic or revenue depends on being discoverable through Google Search, identify what share of that traffic now arrives through AI Overviews and AI Mode, and establish whether the only options on the table are a paid license and an opt-out that costs visibility entirely.

Step 02
Watch the remedy.

APIG’s complaint is a referral, and a referral decides nothing on its own. The outcome to track is whether the Autorité opens a formal investigation, and what interim measures, if any, it imposes on the model of July’s Meta order. Google’s incentive is to run out the clock, and interim measures exist to take that option away.

Step 03
Track the “new service” argument.

The legal question turns on whether generative summarization counts as a use that requires its own negotiation under existing platform commitments. A win for APIG on that point travels well, because any jurisdiction with a comparable neighboring-rights or platform-commitment framework can borrow the reasoning.

Step 04
Separate the three obligations.

Training-data use, indexing, and generative summarization are being litigated as three distinct uses in this dispute, and a settlement on any one of them leaves the other two unresolved.

The Verification Test

Claim Under Test

“Regulators are moving to require platforms to pay publishers for AI-generated search summaries as a use in its own right, over and above payment for training and indexing.”

Test. Check whether the Autorité de la concurrence opens a formal investigation or closes the referral, and whether any resulting order names AI Overviews and AI Mode specifically or folds them into the existing News Showcase framework.

Pass criteria. A formal investigation opens, or interim measures land on Google tied specifically to generative summarization, on roughly the timeline the Meta order took.

Fail smell. The Autorité treats the complaint as already covered by the 2022 commitments and creates no new obligation, which would suggest that generative summarization does not yet count as a distinct licensable use in the EU framework.

The Metric

Two Methods, One Direction What publishers lose where Google’s AI Overviews are live ARCOM · REFERRAL TRAFFIC 33–38% DECLINE TO PUBLISHER SITES AHREFS · CLICK-THROUGH 58% LOWER CTR, TOP ORGANIC RESULT WHAT STILL REACHES THE PUBLISHER REFERRAL VISITS 62–67% remain CLICKS ON THE TOP ORGANIC RESULT 42% remain
Two independent methods measuring two different quantities. Arcom estimates the decline in referral traffic to publisher sites across European markets where AI Overviews are already live; Ahrefs measures click-through on the top-ranking organic result, comparing December 2023 with December 2025 across 300,000 keywords. Neither figure is a direct measure of the other, and neither isolates AI Overviews from other changes to the results page over the same period. Sources: Arcom, as reported by AFP via France 24, August 11, 2026; Ahrefs, February 4, 2026.
33% to 38%

What it measures. This range is the estimated decline in referral traffic to publisher websites in European markets where Google’s AI Overviews are already live, according to France’s audiovisual and digital regulator, Arcom. It is also the quantified harm at the spine of APIG’s complaint.

Why it matters now. The French estimate has independent company. An Ahrefs study of 300,000 keywords, built on aggregated Google Search Console data, found that the presence of an AI Overview correlates with a 58% lower click-through rate for the top-ranking organic result, comparing December 2023 with December 2025. Two different methodologies point the same direction: AI summaries are absorbing the click.

Source. Arcom’s estimate as reported by AFP via France 24, August 11, 2026, and Ahrefs, February 4, 2026.

The Lens — Horizon Search Institute

Responsible AI

Google’s domain-level opt-out is the first content-governance mechanism built specifically for generative search, and publishers read it as control offered without consent to the underlying use.

Human Performance

Independent click-through data now sits beside the regulator’s estimate in the record, and both treat the loss of the reader’s click as the measurable harm underneath the complaint.

Governance & Diplomacy

France’s Autorité de la concurrence has now moved against Google twice and Meta once under the same neighboring-rights framework, and the AGCOM referral plus the Commission’s open antitrust probe suggest that whatever remedy France reaches will set the template the rest of Europe uses.

Links Worth Your Time

Issue Credits
Author
Inaaya Firoz
Managing Editor
Ashwin Telang
Editor-in-Chief
David Lovejoy
Published by Horizon Search Institute, a registered trade name of HSI Research Foundation · EIN 42-1954110 · A Delaware nonprofit corporation · horizonsearch.org