---
title: Two Google-Backed Studies Disagree on Whether EU Rules Keep Shared Search Data Anonymous
description: Google must share search data with rivals from January 2027. Two Alphabet-backed studies tested the EU's anonymisation rules and reached different verdicts.
author: Darie Nani (Editor-in-Chief)
date: 2026-11-18T21:00:23.352Z
updated: 2026-11-18T21:00:23.371Z
canonical: https://www.sovereignmagazine.com/article/google-search-data-anonymisation-dma-studies
image: https://cdn.nanimediahouse.com/berlaymont-european-commission-dma-602548.webp
categories: EU Focus
content_type: News
region: Europe
publication: Sovereign Magazine
schema_type: Article
about:
  - type: Organization
    name: Google
    description: Google is an American technology company, part of Alphabet Inc., best known for its search engine and for products including Android, YouTube, Chrome and Google Cloud. Its internal research on team effectiveness, Project Aristotle, found psychological safety to be the strongest factor behind high-performing teams.
    url: https://about.google
    sameAs:
      - https://en.wikipedia.org/wiki/Google
---

From January 2027, Google must share anonymised Google Search data with rival search engines, including AI chatbots that offer search, under a [16 July decision by the European Commission](https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1634) that applies the [Digital Markets Act](https://www.sovereignmagazine.com/article/eu-lawmakers-digital-markets-act-enforcement-us-pressure). Two research teams have now tested the Commission's anonymisation rules on real Google search data processed under them, and they disagree on whether the people who made those searches stay anonymous.

The Commission's rules strip direct identifiers such as Google usernames and IP addresses from each record, suppress records whose queries contain rare terms such as full names, usernames, passwords or street addresses, or that are unusually long, and generalise what remains so that each user sits in a group of at least 1,000 people with the same location, device type and query language.

Alphabet backed both studies. A team at Universitat Rovira i Virgili in Tarragona published its paper on 12 November, and its acknowledgements say: "Alphabet provided us with access to the data and the technical resources needed to conduct our analysis, as well as covered the expenses incurred in carrying it out." Woodway Assurance, a Canadian company that assesses privacy risk in data, published its study four days later. Woodway says Google engaged it, alongside other external teams, and that it designed and ran the assessment independently and kept sole responsibility for the findings. In its July decision, the Commission set aside earlier arguments from Alphabet, and third-party opinions Alphabet supported or commissioned, citing a potential conflict of interest, the Spanish researchers note in their paper.

On the day the decision was adopted, Kent Walker, president of global affairs at Google and Alphabet, [wrote](https://blog.google/company-news/inside-google/around-the-globe/google-europe/the-dma-should-not-undercut-security-privacy-for-europeans/) that "Europeans' private searches would be exposed to unfamiliar companies, without adequate anonymisation of the data and without user knowledge or consent."

## Spanish Researchers Say the Commission's Measures Fail All Three EDPB Anonymisation Tests

David Sánchez, Benet Manzanares and Josep Domingo-Ferrer titled their [paper](https://arxiv.org/abs/2611.17012) "Anonymous in Name Only". They worked on Irish and Spanish search data protected under the Commission's measures, which they describe as "the same data that competing search engines would have access to". They conclude that the measures "do not meet any of the three anonymization criteria established by the European Data Protection Board (EDPB)": records can still be singled out, linked to each other and used to infer facts about people.

In one attack, the team pulled 421,314 unique records containing the Spanish for "my name" ("me llamo", "mi nombre"). They used Google's Gemini model to rank the records by how rare the names were, checked the top 100 by hand and re-identified around a dozen people. Thousands of Gmail addresses, split into name and surname, got past the filters. Irish Eircodes, postcodes that identify a single property, passed the tool meant to catch identifying details, and the team could link directions searches from the same home Eircode across months. Every unique Icelandic-language query in the Irish data for March 2026 fell on one day, which the researchers say points to a single traveller.

Domingo-Ferrer's team says the Commission's approach "relies on unjustified heuristics rather than formal privacy guarantees, underestimates the risks inherent in free-text search queries, and sacrifices data utility without achieving meaningful protection." It argues that contractual limits on recipients "cannot replace effective technical anonymization".

## Woodway Re-Identified 0.094% of the People It Targeted, at About €105 Each

Woodway's team, led by founder and chief executive Khaled El Emam and working with the University of Ottawa, ran a "motivated intruder" test. Testers gathered public information about named people in six countries, such as their jobs, their interests and often their email addresses, and tried to find those people's search records. Google's team verified every match.

Of the 98,501 people targeted, 0.094%, or about 93, were re-identified. That is below the 0.47% limit Woodway set in advance, which assumed that five separate teams might attack the data. Every country's rate was below the threshold, though one region's was above it, as were the results of some test scenarios that rested on strong assumptions, such as counting every uncertain match as correct.

Each verified match took about three hours of work, equivalent to €105, not counting Google's verification or Gemini costs. Woodway says: "Costs this low would not realistically deter a re-identification attack."

Under its framework, Woodway rated the risk of identifying searchers as insignificant. Its [paper](https://arxiv.org/abs/2611.21689) also says the measures focus on the people doing the searching, "leaving other individuals mentioned in queries insufficiently protected", and do not explicitly address inference, where sensitive facts about someone are deduced from the data. The tools that detect names and email addresses in queries have no stated performance standard, and Woodway says their recall, the share of identifiers they catch, "was assumed rather than tested". It puts the minimum defensible recall at 0.88 and recommends 0.976. The paper says the controls and safeguards around the data "reflect good practice for managing residual risk."

> "Our findings illustrate that the EU Commission's approach can protect personal data with appropriate safeguards, while highlighting certain gaps that deserve attention."
> — Khaled El Emam, founder and CEO, Woodway Assurance

## The Commission Can Amend the Anonymisation Rules After Independent Evaluation

The Commission says Alphabet's initial compliance proposal "was removing between 90 and 100% of unique search queries from the dataset" and excluded AI chatbots, and that "as a result, there has been no meaningful uptake by potential beneficiaries." On top of the technical steps, recipients must keep the data in a ring-fenced environment, may not pass it to third parties and are banned from trying to re-identify anyone. In a passage of the decision that the Spanish team quotes, the Commission wrote that "while Alphabet's testing has shown examples of reidentification, these have not been at scale".

The measures are [reviewed every two years](https://digital-markets-act.ec.europa.eu/businesses-portal/data-access/alphabet-specification-proceedings-sharing-google-search-data_en). On 16 July the Commission said that "depending on future market developments, including based on independent third-party evaluation, the Commission may amend today's decision in particular as regards the relevant anonymisation measures." Walker wrote the same day that "a flexible, evidence-based process will be critical to account for these significant harms and adjust measures accordingly."

## Google Has Asked the EU General Court to Annul the Decision

Google has filed appeals at the EU General Court seeking annulment of this decision and of a second one adopted the same day on [Android interoperability for rival AI assistants](https://www.sovereignmagazine.com/article/google-eu-antitrust-fine-11-billion) (cases T-638/26 and T-639/26), according to an October analysis by the law firm Macfarlanes. Google has also asked for interim relief, whose scope has not been officially confirmed; Macfarlanes says public reports suggest Google wants a temporary suspension of the data-sharing obligation.

## FAQ

**Q: Who can receive the Google search data, and what are they barred from doing with it?**
Eligible online search engines, including AI chatbots with search functions, that have verified investment plans. They may use the data only to develop and improve their search services. They cannot use it to train general-purpose AI models, to improve unrelated services such as consumer profiling and advertising, or to systematically copy Alphabet's search results. Each recipient faces an independent audit before access, a first compliance audit within six months of sharing and audits at least once a year after that. Before sharing, Google can also assess whether a specific recipient poses serious cybersecurity and data protection risks.

**Q: What else is removed from each shared search record?**
Besides usernames and IP addresses, the Commission's rules remove precise query timestamps, advanced filters and input format from each record, so that different queries cannot readily be attributed to the same user. Bank account numbers are among the rare terms that trigger suppression. The grouping of users by location, device type and query language is a technique known as k-anonymity, and the Commission says 95% of users will be in groups of at least 25,000.

**Q: Does anonymisation take the shared data outside the GDPR?**
No. The Commission says the dataset still holds personal data about people other than the searchers, such as people who are searched for; its example is a query for "Kylian Mbappé". The sharing therefore remains subject to the GDPR, and recipients are controllers for that data.
