The EU Wants Google to Open Up Search Data to Rival AI Engines

The legal foundation is already in force rather than being a new proposal. Article 6(11) of Regulation (EU) 2022/1925 requires designated search gatekeepers to give third-party online search engines access on “fair, reasonable and non-discriminatory terms” to ranking, query, click and view data from free and paid search. Query, click and view data that constitute personal data must be anonymized.

The current development, as framed by the Associated Press report, is about making access useful to rival AI engines and addressing access involving Google’s Gemini services. A binding decision adopted on 16 July 2026 now defines eligibility thresholds, anonymization rules, latency limits, pricing methodology and implementation milestones; final prices remain due by January 2027. Those details need confirmation before anyone treats the policy as operational infrastructure.

The competitive effect could still be substantial: shared behavioral data may weaken an advantage built from years of query and interaction feedback. But it is fuel, not a finished research index. Structured Data will help determine whether an engine can turn a promising URL into an attributable claim about the right person, organization, date and source.

Query reformulations can reveal that two phrases express a similar need. Click patterns can help identify results users investigate, while ranking histories can supply context about which documents were available and visible. These signals can narrow retrieval candidates, but clicks are affected by position, presentation, familiarity and existing popularity. The supplied research does not provide a peer-reviewed numeric estimate for click-position bias or dwell-signal quality, so no such figure can be responsibly assigned here.

The wider manipulation risk is concrete. A 2026 arXiv preprint reported a 91.4% promotion rate while testing strategically engineered context against AI-search ranking systems. The study’s scope cannot be generalized to every engine from the supplied summary, but the result is a clear warning: observed popularity or promotion is not proof that a source is correct.

Structured Data is machine-readable information attached to a page using a standardized vocabulary. Schema.org defines types such as Article, NewsArticle, ScholarlyArticle, Person and Organization. Properties including author, datePublished, dateModified, citation, sameAs and about give a retrieval system explicit facts and relationships that are harder to recover from an ambiguous page layout.

That semantic layer can help an engine resolve whether “Jordan” is a person, country or brand; connect an author to a stable identity; and preserve provenance when extracting a passage. It does not guarantee ranking, inclusion or citation. Syntactically valid markup that contradicts visible content can create a worse entity record, so a knowledge-graph specialist should review meaning as well as validator results.

Shared search signals could help ChatGPT and other answer engines find useful evidence beyond their current browsing, retrieval and licensing arrangements. The likely pipeline is straightforward: behavioral data narrows candidate sources, Structured Data identifies entities and provenance, a retrieval system extracts relevant passages, and the answer layer attaches citations to individual claims.

Retrieval parity would not produce citation parity. An SSRN working paper explicitly challenges the assumption that being number one on Google guarantees citation in AI search. The working paper found that Google rank strongly predicted AI citation and dominated the tested intent and page-feature predictors, while also showing that a number-one Google ranking did not guarantee citation.

The strongest systems will combine shared signals with independent crawling, publisher agreements, source-quality checks and citation interfaces that show which evidence supports which sentence. Citation presence alone is weak. A useful citation should support the attached claim, distinguish primary evidence from commentary and remain understandable after the source page is updated.

Europe needs to prevent Google’s feedback loop from becoming an unchallengeable moat without exporting the same biases to every rival. Article 6(11) supplies a legal base, while the 91.4% promotion result in the cited arXiv preprint shows why access must be paired with auditing rather than treated as ground truth. The July 16, 2026 measures define eligibility, privacy thresholds, latency, security safeguards, audits and a pricing formula, while Google must finalize its pricing offer by January 2027.

The bottom line: publishers should implement truthful JSON-LD now, while AI engines should build independent source evaluation instead of treating Google-derived clicks as truth. Watch the EU’s final eligibility, privacy, pricing and delivery rules; those details will determine whether the policy creates genuine retrieval competition or simply reproduces Google’s popularity patterns.