The EU Just Forced Google to Share Its Search Data
On 16 July 2026 the European Commission adopted a final decision that forces Google to hand over anonymised Search data to rival search engines and, crucially, to AI chatbots with search functions. It is enforcement of Article 6(11) of the Digital Markets Act, and it aims squarely at the moat that has kept Google above 90% market share in Europe for decades.
If you work in search, discovery or growth, this is not a niche compliance story. It is the first serious attempt to loosen Google's grip on the raw material that makes search good: user query data. I think it matters far more than most people have clocked.
What actually happened
The Commission opened specification proceedings on 27 January 2026 because Google's own compliance proposal was, in the regulator's words, not working. Google was removing between 90 and 100% of unique search queries from the dataset it offered, and it excluded AI chatbots that provide search from the pool of beneficiaries entirely.
The result was predictable: no meaningful uptake. After two years of talks, the Commission decided a formal specification was the fastest route to a dataset that actually meets the law. The final decision and measures now spell out exactly how Google must anonymise and share the data.
The timeline in plain terms
| Date | Event |
|---|---|
| 27 January 2026 | Commission opens specification proceedings under Article 6(11) DMA |
| 16 July 2026 | Final decision adopted, setting out the measures Google must implement |
| Within 6 months of sharing | First independent compliance audit of each beneficiary |
| Annually thereafter | Recurring compliance audits |
| Every two years | Biennial review of the measures themselves |
Timeline compiled from the European Commission's DMA developer portal.
Who gets the data, and what they can do with it
Eligibility is not open season. Only search engines with verified investment plans to improve their services qualify, and they must pass an independent audit before access is granted. That deliberately includes AI chatbots offering search functions, closing the exact loophole Google tried to open.
The permitted uses are tightly drawn. Beneficiaries can use the data to improve three things: query understanding (intent, spelling, autocomplete, related queries), ranking and retrieval (including the grounding that keeps AI answers accurate), and indexing (understanding which sites to prioritise).
What they cannot do is telling. They may not train general-purpose AI models on it, use it for consumer profiling or advertising, or systematically replicate Google's results instead of building their own. In my opinion this is the smartest part of the ruling: it hands over the ingredient without licensing a copy of Google's kitchen.
How the anonymisation actually works
This is where the detail earns its place, because the whole thing lives or dies on privacy. Google must apply technical measures before sharing anything, in three main steps.
- Build a haystack. Strip direct identifiers (usernames, IP addresses) and other attributes like timestamps and input format, so queries cannot be readily tied back to one user.
- Suppress the secrets. Remove records containing rare terms such as full names, passwords, street addresses or bank details, plus unusually long queries.
- Apply k-anonymity. Generalise metadata so every user sits in a group of at least 1,000 users sharing the same location, device type and query language. That 1,000 is the strict floor: 95% of users will sit in groups of at least 29,000.
On top of that sit contractual measures: ringfenced environments, no linking to other datasets, no re-identification attempts, limited retention, and yearly audits. The approach follows the draft joint guidance from the Commission and the EDPB on how the DMA and GDPR interact.
What this changes for your team
From my observation, the practical shift is about optionality. More credible search alternatives mean more places your content can be discovered, and, as the Commission notes, more traffic sources and greater ad market competition for businesses.
If rival engines and AI chatbots get better at query understanding and grounding, the answer engines your customers already use could improve faster. That reinforces a point I keep making: optimising for a single dominant engine is a shrinking strategy. This is the moment to widen your understanding of who your audience actually is and where they search.
Concrete actions: audit your visibility across non-Google surfaces now, ensure your structured data and content are machine-readable for grounding, and treat AI chatbots as a discovery channel rather than an afterthought. If your stack still funnels everything to one source, this is your cue to rethink your data foundation for a multi-engine world.
The moat is not gone. But for the first time, the drawbridge has a public specification for lowering it. Watch which chatbots and engines qualify first, because that list will tell you where discovery is heading next.
Tags