EU Forces Google to Share Search Data

EU set to force Google to hand rival search engines its query data by July 27

Brussels' binding decision under DMA Article 6(11) will pipe Google's ranking, query, click and view data to Ecosia, Qwant, OpenAI and Perplexity — over Google's privacy objections.

The European Commission has until July 27, 2026 to sign a binding order that forces Alphabet to share the raw plumbing of Google Search — ranking, query, click and view data — with any online search engine or AI chatbot that asks for it, on FRAND terms and via API, for at least five years. The measure, drafted under Article 6(11) of the Digital Markets Act, is the first hard test of whether Brussels can engineer competition into a market it already regulates. The angle worth watching: the same anonymisation debate that has quietly stalled EU privacy law for a decade is now the last lever Google has to narrow the remedy — and the biggest beneficiaries are not the European search startups the DMA was written for, but American AI firms.

What Brussels is actually ordering

The Commission's preliminary findings, adopted on 16 April 2026, are unusually prescriptive. According to the official case summary, Alphabet must give any third-party search engine in the European Economic Area access to "all the Search Data on par with the data collected by Alphabet for the purpose of optimising its OSE services." That parity principle is the load-bearing phrase. It means Google cannot ship a stripped-down feed to rivals while keeping the richer version for Gemini or its own ranker.

The delivery mechanism is an API, refreshed daily, sending only new records rather than a full dump each pull — a design that mirrors how Google feeds its own systems. Access must run for at least five years per beneficiary, or for as long as Google Search remains a designated gatekeeper service, whichever is longer. Pricing is capped: incremental cost plus a return that "shall not exceed Alphabet's weighted average cost of capital." This is not a licence to monetise the dataset; it is a compliance duty priced at cost-plus.

Crucially, the Commission has declared AI chatbots with search functionality — read: OpenAI, Perplexity, Anthropic, Mistral — eligible "data beneficiaries." The 16 April Commission press release makes that inclusion explicit. This is the point where a competition file quietly becomes an AI-industrial-policy file.

The privacy trapdoor Google is trying to widen

Article 6(11) of the DMA, as published in the EU Official Journal, requires that any query, click and view data constituting personal data "shall be anonymised." Recital 61 goes further: the data must be "irreversibly altered." That single word — irreversibly — is where the fight now sits.

Google's position, briefed heavily in the run-up to the July decision, is that irreversible anonymisation of raw search queries is either technically impossible or so aggressive it destroys the dataset's usefulness. The company is not wrong on the risk. The 2006 AOL release of 20 million supposedly anonymised search queries from 650,000 users produced within days a New York Times story identifying "user 4417749" as a specific Georgia widow. Two decades of academic literature since — from Narayanan and Shmatikov's foundational Netflix Prize deanonymisation to a 2026 arXiv paper showing LLM agents can now re-identify anonymised search histories using only public web evidence — has hardened the finding: search logs are one of the most re-identifiable data types in existence.

The Commission's technical fix, developed with the European Data Protection Board, is layered. Direct identifiers and IP addresses are suppressed, long queries and queries containing rare words are dropped, location and device type are generalised, and session data is limited to refinements of prior queries. Contractual clauses ban re-identification attempts, onward sharing and any use outside search. Brussels' argument, set out in the Joint EDPB–Commission guidelines on DMA/GDPR interplay, is that technical measures reduce re-identification risk to "residual" and contractual measures push it to "insignificant."

Mark MacCarthy, in a June 2025 Brookings analysis of the parallel US antitrust remedy proposed by the Department of Justice, warned that the AI race will create pressure to exploit any shared search dataset for personalisation: religion, politics, health, sexuality. His verdict on the DOJ's proposed safeguards — "woefully lacking" — is the argument Google is now recycling in Brussels. The Commission's package is materially tougher than the DOJ's. But every additional privacy safeguard is also a data-quality cut, which is precisely what Google wants.

Who actually benefits

The DMA was sold politically as a lifeline for European search. Ecosia, the Berlin-based non-profit, and Qwant, the French privacy engine, have lobbied for Article 6(11) enforcement for two years, arguing that Google's scale advantage — nine in ten European queries, according to a December 2025 European Parliamentary Research Service briefing — makes independent ranking economically impossible. Bing's own filings to the Commission, cited in a DMA gatekeeper case decision, estimated its share at just 3.6% of European search clicks and concluded that even full access to Google's query data would raise Bing's share by only 0-5%.

That is the awkward part. The measure that European search startups wanted most may move their market share barely at all. The players positioned to extract real value are the AI chatbots the Commission explicitly waved in. OpenAI, Perplexity and Google's own Gemini currently dominate a fast-growing generative search layer. A 2026 Cambridge Data & Policy study of 14,000 LMArena conversations found Gemini and Perplexity's Sonar visit around ten relevant web pages per query but cite only three to four. Those systems need exactly the click-and-view signals Article 6(11) will make available: what users actually asked, what URLs Google's ranker thought were relevant, and what users clicked.

The historical parallel that reframes this: in 2010–2012, the FCC's data-portability push for telecoms created scale for competitors but redirected surplus toward over-the-top platforms — Skype, WhatsApp — that were never in the regulator's line of sight. Brussels' Article 6(11) file looks structurally similar. The stated beneficiary is European search; the actual beneficiary is US AI infrastructure that will use European search demand as free training signal. The Bruegel working paper published on 1 July 2026 on DMA Article 6(7) makes precisely this argument for the parallel Android AI-interoperability file the Commission is running.

What the parallel US case tells us

The US Department of Justice's remedy proposal in United States v. Google, filed in the D.C. District Court, contains a nearly identical data-access remedy. The May 2025 status report disclosed the technical specifications of Google's existing European Search Dataset Licensing Program: a k-threshold of at least 30 unique signed-in users worldwide over 13 months for any query to be released, and an m-threshold of five users for each query–result–device–country combination in a given quarter. In practice, this excludes an enormous share of the long tail — exactly the queries where Ecosia, Qwant and AI chatbots most need training signal.

The Commission's 16 April draft measures require Google to abandon that quarterly aggregation for daily API delivery and to widen the query set, but the k- and m-threshold logic will survive in some form. Whether the final thresholds sit close to Google's current programme or materially below it is the single most important variable in the 27 July decision. It determines whether "parity" is a legal fiction or a real transfer.

"Data is a key input for online search and for developing new services, including AI. Access to this data should not be restricted in ways that could harm competition." — Teresa Ribera, Executive Vice-President for Clean, Just and Competitive Transition, European Commission press release, 16 April 2026

The forecast, and what would break it

Named forecasts from analysts tracking the file converge on three probabilities. First, the Commission will adopt a binding decision by the 27 July statutory deadline — the Commission's own case page states the six-month clock is hard. Second, Google will challenge the measures at the General Court in Luxembourg within two months of publication, most likely on Article 8 (right to private life) and proportionality grounds, drawing on the ECJ's own case law tightening the definition of personal data. Third, the anonymisation thresholds in the final decision will land closer to the Commission's preliminary text than to Google's counter-proposal, because the EDPB has publicly signed the interplay guidelines and cannot now retreat.

Enforcement history supports the direction of travel. On 2 July 2026, the European Court of Justice upheld the €4.1 billion Android fine against Google, reported by The Economist. The European Parliament's DMA state-of-play briefing records that Apple and Meta were fined €500 million and €200 million respectively in April 2025 in the first non-compliance decisions. Brussels is not in a mood to soften.

Forward catalysts to watch:

• 27 July 2026 — Commission adopts binding Article 6(11) measures on Alphabet; text becomes the reference for every future gatekeeper data-sharing dispute.
• Late September 2026 — Deadline for Google's General Court appeal filing; whether Alphabet seeks interim measures signals how disruptive it thinks compliance will be.
• Q4 2026 — First data flows to non-Google search engines and eligible AI chatbots via the mandated API; expect the first user re-identification claims within weeks, filed by Max Schrems' NOYB or a Member-State DPA.

Diplomat View

The politically saleable framing of Article 6(11) — European regulators help European search challengers dislodge an American gatekeeper — is not what the file will actually produce. The mandated dataset is too aggregated to move Bing, Ecosia or Qwant more than a few percentage points, and Google's real scale advantage lives in click-through data on personalised results the DMA cannot force it to share. The dataset is, however, exactly what generative-AI search agents need to reduce hallucination and improve retrieval — a competency gap where OpenAI and Perplexity, both eligible under the Commission's own definition, are currently spending billions to close.

The forecast: the 27 July decision holds, Google appeals but does not win a stay, and by mid-2027 the visible market impact of Article 6(11) is measured not in European search share but in the citation and grounding quality of US AI chatbots serving European users. Falsifiers: if the final decision materially raises the k- and m-thresholds above Google's existing licensing programme, or if the General Court grants interim suspension, the remedy becomes cosmetic and this forecast is wrong. If, conversely, EU privacy regulators subsequently block AI chatbots from accessing the dataset on GDPR grounds — a NOYB complaint is the obvious trigger — Brussels will have engineered a competition remedy that benefits nobody it named. Both scenarios are plausible; neither is the base case.

The Bottom Line

The EU is not really forcing Google to help Europe's search underdogs — it is forcing Google to subsidise the next generation of American AI search. Article 6(11), as drafted, will transfer the most re-identifiable dataset in commercial computing to eligible AI chatbots on cost-plus terms, and the privacy safeguards, however serious, will be tested in weeks. Watch 27 July, then watch who files the first complaint.