Search queries are messy little confessions, and that is exactly why this case deserves more than a bureaucratic shrug. Woodway Assurance's independent assessment says Google's search-query data could be treated as anonymized for sharing under the Digital Markets Act, but only if the European Commission tightens parts of its July 2026 framework. That is not a clean bill of health. It is a conditional finding with some uncomfortable math sitting right in the middle of it.
A Competitive Mandate Meets Human Reality
The DMA requires Google to share certain search data with eligible rival search engines so they can build and improve their own services. Competition advocates will see the point immediately: data advantage is market power, and hoarding it can freeze out smaller players. Fair enough. But search data is not a bland spreadsheet of clicks. It can reveal anxieties, medical concerns, relationships, jobs, finances, politics, and plenty else people would never volunteer in public.
The Commission's decision requires Google to anonymize the data before it is shared and imposes safeguards on recipients. Google brought in Woodway and other outside teams for evidence about whether that approach protects users. Woodway says it independently designed and ran its work, while retaining responsibility for its analysis and conclusions. That independence matters, though investors and policymakers should still read this as an assessment of a particular process and data environment, not a universal promise that search data is harmless once labels are stripped away.
The Number That Should Stop Complacency
Woodway's motivated-intruder exercise targeted 98,501 people across six countries. Testers used publicly available facts about named individuals, including professions, interests, activities and often email addresses, then tried to find those people in Google's search records. Google verified each match. The reported re-identification rate was 0.094%.
At first glance, that sounds tiny. And under Woodway's predefined residual-risk threshold of 0.47%, it falls below the line. But each verified match took roughly three hours and cost about €105. That is the part that sticks in the throat. A cheap attack does not need to work on everyone to cause real harm; it needs to work on somebody worth targeting. Woodway itself says that cost would not realistically deter a re-identification attempt. No sensible reader should glide past that sentence.
Thresholds Need Teeth, Not Just Decimal Places
Woodway's central recommendation is practical rather than theatrical. The Commission should specify explicit criteria for tools that detect direct identifiers in queries, including names and email addresses. It should also add protections against sensitive facts being inferred about people. Those are not cosmetic edits. If direct identifiers slip through, or if combinations of seemingly ordinary searches make a person distinguishable, the whole data-sharing arrangement gets shakier.
The firm also argues for pairing intruder testing with quantitative risk modelling. That strikes me as the grown-up approach. An intrusion test shows what determined people managed to do under defined conditions. Modelling forces the parties to state assumptions about potential attackers, attack frequency, controls at recipient organizations, and scenarios that were not tested. Neither tool is magic. Together, they can expose where someone has quietly assumed away the risk.
Low observed risk is not the same thing as no meaningful risk, especially when search histories can carry highly personal context.
Why This Matters Beyond One Data Release
For Google, the immediate issue is compliance with a competition rule that collides with a privacy obligation. For competing search engines, access to useful data could help narrow an enormous quality gap. For privacy-tech providers, this is a reminder that defensible evidence—not glossy assurances—will increasingly be part of the commercial conversation. Woodway's EviData platform sits directly in that lane, offering privacy-risk assessment, privacy-preserving transformations, and evidence generation.
Still, the broader lesson is not that every company should rush to monetize anonymized data. It is that regulators need to say exactly what “anonymized enough” means in a setting where motivated outsiders can cheaply test the claim. The Commission has a workable opening here, according to Woodway, but the gaps are real. Tighten identifier detection, address inference risk, document assumptions, and audit the recipients. Anything less leaves a market-opening policy carrying privacy baggage it does not need.