EU Orders Google to Share Search Data in 2027, With Strict Limits on AI Training
The European Commission's DMA measures will require Google to provide eligible search engines and AI chatbots with anonymized Search data from January 2027.
The European Commission has adopted final Digital Markets Act measures requiring Google to share anonymized Google Search data with eligible third-party search engines and AI chatbots from January 2027. The decision is a significant intervention in how a major search platform's data can be used by rivals, but it includes a crucial boundary: recipients cannot use the shared data to train general-purpose AI models.
The measures aim to make it easier for qualifying competitors to develop and optimize search services and AI-enabled search tools using data categories that Google uses to improve its own search offerings. At the same time, the Commission has set privacy, access, pricing and audit requirements intended to prevent the mandate from becoming unrestricted access to user-level search behavior.
What the DMA decision requires
The Commission adopted the final specification measures on 16 July 2026 under the DMA. According to the European Commission's Alphabet specification proceedings on Google Search data sharing, Alphabet must finalize its anonymized dataset and pricing by January 2027, then begin providing access to eligible beneficiaries in that month.
The dataset is intended to include the types of information Google uses to optimize search. That scope includes ranking data, query text, metadata such as language and device type, displayed URLs, interactions with search results, and information about result positioning. In practical terms, access could help an eligible service understand how queries, results and user interactions relate within European search use cases.
The requirement is not open to every company that wants search data. The measures establish eligibility criteria for beneficiaries, including credible online search engines with an EU presence and adequate scale. Access must be offered on fair, reasonable and non-discriminatory terms, and an independent verification regime will conduct annual audits.
Privacy is central to the design. Google must use a multilayer anonymization approach, supported by safeguards and audit requirements intended to limit residual re-identification risks. That means the policy is not a mandate to release identifiable search histories or a raw replica of Google's internal data systems.
| DMA specification measure | Who it is designed to support | Key timing | Core limitation or safeguard |
|---|---|---|---|
| Anonymized Google Search data access | Eligible third-party search engines and AI chatbots | Dataset, pricing and sharing from January 2027 | Multilayer anonymization, eligibility checks, audits and no general-purpose AI training |
| Android interoperability improvements | Third-party AI assistants | Targeted for use in 2027, with broader availability expected by July 2027 | Access concerns core operating system features and voice activation, rather than Search data access |
The Commission issued a separate specification decision on Android interoperability. It is intended to let third-party AI assistants access core Android operating system features, be activated by voice and perform actions in apps for users. While both decisions concern competition in Google's ecosystem, the Android measure addresses platform functionality, whereas the Search measure addresses access to anonymized data.
Why the rules matter for search and AI services
The strongest immediate effect is likely to be on search competition and AI-enabled search experiences, not on the foundation-model race. Search data can be valuable for improving how a service interprets queries, evaluates result relevance and learns from aggregate interactions with results. The Commission's measure explicitly connects the shared data to the development and optimization of third-party search services and AI-enabled search tools.
However, the prohibition on using the data to train general-purpose AI models is material. It prevents eligible recipients from treating the dataset as a broad corpus for building or improving a general-purpose model. Developers and organizations assessing the opportunity will need to separate search-specific optimization work from general model training workflows, and establish governance that reflects the permitted purpose.
That distinction creates several practical requirements for potential beneficiaries:
- Eligibility preparation: Organizations will need to meet the Commission's access criteria before they can receive the data.
- Purpose controls: Technical and organizational controls will need to keep the data within permitted search and AI-enabled search uses.
- Privacy governance: Recipients must account for the anonymization safeguards and the residual re-identification risk the regime is designed to limit.
- Commercial planning: Access will be subject to a pricing framework rather than guaranteed free availability.
- Audit readiness: The annual independent verification regime makes compliance an ongoing operational issue, not simply an onboarding step.
For Google, the decision formalizes a new obligation around an asset that has historically supported search quality at exceptional scale. For eligible competitors, it may reduce one barrier to building search products that need richer evidence about queries, rankings and result interactions. It does not eliminate the other difficult parts of operating a search service, including infrastructure, product design, retrieval systems and trust.
The Android decision could have a separate competitive effect. AI assistants that can be invoked by voice and carry out actions across apps may become more viable alternatives on Android devices, depending on the implementation of the interoperability requirements. The Commission expects the changes to be used in 2027 and more broadly available by July 2027.
Organizations considering AI-enabled search or assistant experiences will need to treat these measures as both a product opportunity and a compliance design problem. Scalevise can help teams assess AI architecture, data governance and workflow integration when adapting products to new platform and data-access rules.
Frequently Asked Questions
When must Google begin sharing anonymized Search data?
Alphabet must finalize the anonymized dataset and pricing by January 2027, and sharing with eligible beneficiaries is due to begin in January 2027.
Who can receive Google Search data under the EU measures?
Eligible third-party online search engines and AI chatbots can seek access. The measures include eligibility criteria such as a credible service, EU presence and adequate scale.
Can recipients use the shared data to train general-purpose AI models?
No. The measures explicitly prohibit using the shared Google Search data to train general-purpose AI models.
What information will the shared dataset include?
The specified data types include ranking data, query text, metadata such as language and device type, displayed URLs, user interactions with search results and result-positioning information.
What changes are planned for Android AI assistants?
A separate Commission decision is intended to improve Android interoperability so third-party AI assistants can access core operating system features, be voice activated and perform actions in apps on behalf of users.
Conclusion
The EU's DMA specifications create a defined route for eligible search and AI chatbot providers to access anonymized Google Search data from January 2027. The opportunity is deliberately narrow: it supports competing search and AI-enabled search services, while privacy safeguards, audits, pricing rules and the ban on general-purpose AI training set firm limits on how the data can be used. Alongside the Android interoperability requirements, the measures could expand room for rival services in Google's European ecosystem without removing the operational and governance demands of building them.