A privacy assessment commissioned by Google has concluded that the search data the European Commission is forcing it to share with rival search engines could count as anonymised - but only after Brussels closes gaps in its own rules. Woodway Assurance, a Canadian company that sells privacy risk software, said on November 18, 2026 in Brussels that its testers re-identified 0.094% of the 98,501 people they targeted in Google search data used to test the Commission's measures, a rate that falls below the 0.47% residual risk threshold it had set in advance.

The finding cuts both ways. According to Woodway, Google's end-user search query data "could be considered anonymized under established contextual practices, provided certain gaps in the European Commission's anonymization measures are addressed first." The company named two of those gaps: the absence of explicit performance criteria for the tools that strip direct identifiers, such as names and email addresses, out of search queries, and the need for further measures to stop sensitive information being inferred about individuals. The headline of Woodway's announcement put it more bluntly, saying the Commission's measures "need to be updated to ensure Google search data has low privacy risk."

Google's role is set out in the announcement itself. "Google engaged Woodway Assurance, alongside other external teams, to obtain third-party evidence on how well the EU Commission's approach would protect their users' privacy," according to Woodway. Google also supplied the data used in the test, and "Google's team verified every match." Woodway said it worked with the University of Ottawa and other collaborators, "independently designed and conducted its assessment and retained sole responsibility for analyzing the data, interpreting the results and reporting the findings."

The assessment is therefore evidence produced at the request of the company the rules bind, by a firm whose business is selling privacy risk assessments. It is not a finding by the Commission, a data protection authority or a court.

In Short

A privacy firm hired by Google tested Europe's rules for sharing Google search data with rival search engines and managed to identify 0.094% of the 98,501 people it went looking for, which it says is below its own 0.47% risk limit. The rules decide what Google must hand to competing search engines and AI chatbots that answer search queries, with sharing due to begin in January 2027. Woodway Assurance says the data can count as anonymous only if the Commission sets performance standards for the tools that remove names and emails from queries, and adds protection against sensitive details being inferred about people. Because Google commissioned the work, supplied the data and checked every match, the result is one party's evidence rather than a regulatory finding, and it arrives while the Commission's own measures remain open to amendment.

What the testers did

Woodway's assessment had two parts, according to the company. The first was a critical analysis of the Commission's anonymisation requirements. The second was empirical motivated intruder testing - controlled attempts to re-identify individuals - using search data and structured data made available by Google. Woodway said the testing ran under a protocol approved by a research ethics board, "with defined safeguards for re-identification testing, access to sensitive data and reporting of findings."

The method assumes an attacker with a purpose. "Motivated intruder testing, in which testers act like someone determined to identify people in the data, gives real evidence of where weaknesses remain," according to Woodway. The main test targeted 98,501 individuals across the six countries covered by the data. Testers gathered publicly available information about named people, "such as their professions, interests or activities and, often, contact details like email addresses," and then used it to locate those individuals' search records.

The approach mirrors the kind of attack the Commission's contractual measures prohibit. Under the decision, recipients are barred from attempting re-identification, for example "by linking the Search Data with auxiliary datasets." Woodway's testers did, in effect, what a rule-breaking recipient might attempt.

Woodway was explicit that a test of this kind has limits. "On its own, however, it cannot show that the risk of re-identification is low across all users and under all the ways the data might be accessed," the company said. It therefore combined the test results with quantitative risk modelling, which, according to Woodway, "combines the test findings with explicit assumptions about attackers, how likely attacks are, and how well the recipients' controls work." The company argued that "a defensible assessment needs both kinds of evidence, with uncertainty and untested scenarios stated openly." It also disclosed that it "used elements of this modelling to set the threshold against which it interpreted the results."

The numbers

Measure Figure Source
Individuals targeted in the main test 98,501 Woodway
Countries covered by the data 6 (not named) Woodway
Identity disclosure rate observed 0.094% of targeted individuals Woodway
Implied number of people re-identified About 93 PPC Land calculation (98,501 x 0.094%)
Average effort per verified match 3 hours Woodway
Cost equivalent per verified match €105 Woodway
Implied hourly rate behind the cost figure €35 PPC Land calculation (€105 / 3 hours)
Implied total effort for all matches About 279 hours, about €9,800 PPC Land calculation
Predefined residual risk threshold 0.47% Woodway
Observed rate as a share of the threshold One fifth PPC Land calculation (0.094 / 0.47)
Who verified each match Google's team Woodway

Two of Woodway's own sentences pull in different directions. On cost, the company wrote: "Costs this low would not realistically deter a re-identification attack." On the overall result: "In a quantitative model using a predefined residual risk threshold of 0.47%, the observed rate of 0.094% for identity disclosure falls below that threshold."

Both can be true at once. A few hours of work and roughly a hundred euros per match is cheap, so cost is not the barrier. What keeps the measured rate low, on Woodway's account, is how rarely the attacks succeeded at all - around one target in a thousand - combined with the contractual and audit controls that the model assumes recipients operate under. The 0.094% figure covers identity disclosure only. Woodway did not publish a comparable rate for inferences about sensitive attributes, which is one of the two gaps it identified.

Two gaps in the Commission's measures

Detector performance

The first gap concerns the software that finds and removes direct identifiers - names, email addresses and similar strings that users type into a search box. Woodway said the Commission should specify "explicit criteria for the performance of tools to detect direct identifiers, such as names and email addresses, in search queries."

The decision summary published in the Official Journal requires Alphabet to "suppress direct identifiers and other identifying attributes from search records." It does not set a performance standard for how reliably that suppression must work. The full decision text is more detailed, but the detector duties it imposes are about disclosure. PPC Land's review of the full decision in September found that Alphabet must tell the Commission which detectors it will use, with confidence thresholds and reasons for choosing them, and must report the detectors' performance statistics every quarter. That review did not describe a minimum accuracy level the detectors must reach. Woodway's first recommendation targets exactly that space.

Detector accuracy matters in both directions. A detector that misses names leaves people exposed in the shared data. A detector that flags too much strips out ordinary queries, which shrinks the dataset rival search engines are paying for. As Damien Desfontaines, a privacy engineer who advised the Commission, explained in September, the detectors sit at the start of a pipeline that then deletes any record containing an entity searched by fewer than 50 distinct users in the past year.

Sensitive inferences

The second gap is about what can be learned about a person even when they are not named. Woodway called for "further measures to protect against sensitive information being inferred about individuals."

That concern maps onto the European Data Protection Board's own test for anonymity. In Guidelines 02/2026 on Anonymisation, adopted on July 7, 2026, the board set three criteria: no record isolation, no linkage and no inference. Woodway's 0.094% figure addresses the first two, by measuring whether testers could single out and link specific people. Its second recommendation points at the third.

The Commission's technical measures already blur some attributes. According to the decision summary, Alphabet must "generalise location and user interaction metadata to ensure indistinguishability among users," limit sessions to chains of up to three refining queries, and "coarsen durations (e.g. click-back time, and hovering and viewing durations) into time intervals instead of precise time data." Woodway did not say which additional measures it has in mind.

A test ordered by the company under the obligation

This is not the first time Google has put privacy evidence in front of the Commission on this file, and the Commission rejected the earlier attempt. In May 2026, Google claimed to have re-identified several individuals in under two hours using one week of Irish search data. In its final decision, the Commission described the exercise as a partial assessment rather than red teaming, run without the contractual security controls the decision imposes. DuckDuckGo's Aurélien Mähl later said in a podcast interview that the filters Google tested did not reflect the Commission's measures in full.

Woodway's design addresses that point. Its model explicitly includes "how well the recipients' controls work," the same layered structure the Commission uses: technical measures that reduce the risk "to a residual level," and contractual measures that "further reduce the residual risk of re-identification of end users to an insignificant level." The two exercises used different data and are not directly comparable. But where Google reported several re-identifications within two hours, Woodway's testers averaged three hours per verified match and succeeded for about one target in a thousand.

Woodway's chief executive framed the outcome as broadly supportive of the Commission's approach. "Organizations sharing data need confidence that individuals' privacy is sufficiently protected. Our methods and solutions draw on decades of pioneering research and practical application to assess privacy risks relative to established standards and precedents while maintaining data utility," said Dr. Khaled El Emam, founder and CEO of Woodway Assurance. "Our findings illustrate that the EU Commission's approach can protect personal data with appropriate safeguards, while highlighting certain gaps that deserve attention."

The independence claims sit alongside facts that limit them. Google picked the assessor, provided the data, and confirmed each match. Woodway said Google engaged it "alongside other external teams," but did not say who those teams are or whether their results have been published. The call for updated measures is also consistent with the position Google took when the decision was adopted: Google objected on July 16, arguing that the measures put European users' privacy at risk.

What the July decision requires

The rules Woodway tested were set by a Commission implementing decision adopted on July 16, 2026, under Article 8(2) of the Digital Markets Act. A summary of that decision, case DMA.100209 - SP - ALPHABET - Article 6(11), was published in the Official Journal of the European Union on August 14, 2026 as C/2026/4443, under Article 44 of the regulation, which requires the Commission to publish the main content of its decisions while protecting business secrets.

The underlying obligation is short. According to the summary, Article 6(11) requires a gatekeeper "to provide to any third-party OSEs, at its request, with access on FRAND terms to ranking, query, click and view data in relation to free and paid search generated by end users on its OSEs. Any such query, click and view data that constitutes personal data must be anonymised." Alphabet was designated a gatekeeper on September 5, 2023, with Google Search listed as a core platform service.

The Commission did not mince words about Google's previous offer. "One of the key issues raised by third parties is Alphabet's extreme anonymisation approach, which removes the large majority of unique queries from the dataset," the summary states. Third parties also reported missing data types, "e.g., URLs displayed in advanced search blocks, granular location data and user interaction data," a substantial lag and short duration, pricing they considered not FRAND, and Alphabet's rejection of "operators of AI chatbots with search functionalities as beneficiaries of the data." PPC Land's review of the full text found that Google's old filter removed between 90 and 100 percent of unique queries, while the Commission's entity-based method removes between 10 and 20 percent.

The decision specifies measures in five areas.

Area What the decision specifies, according to the summary
Eligibility Which third parties may receive the data and how Alphabet assesses applicants, subject to Commission supervision
Data scope Query, view, ranking and click data, shared on par with what Alphabet collects to optimise its own search engine
Anonymisation Technical measures plus contractual measures to reduce re-identification risk "to an insignificant level"
Pricing terms FRAND pricing set at incremental cost plus a reasonable return on capital employed
Data acquisition and testing Testing samples, eligibility review, audits, suspension and termination, timeline and pricing process

Eligibility

Beneficiaries must operate an online search engine in the European Economic Area, and the measures state that this "includes AI chatbots that provide OSE functionalities." Applicants must also show Alphabet that they meet the conditions below.

Requirement Wording or threshold in the summary
Sanctions Not subject, directly or indirectly, to restrictive measures or other sanctions under EU law
Third-country control Not subject, "de jure or de facto, to the control of a third country that poses a serious and structural non-technical cyber security and/or data protection risk"
Scale or credibility "at least 50 000 monthly users, and have operated an OSE for at least two years in the EU or have received at least EUR 50 million in investments"
Data transfers Processing in the EEA or an adequate country; otherwise Alphabet must run a transfer impact assessment and must find the applicant ineligible if Chapter V GDPR protection cannot be guaranteed
Public security Alphabet may ask the Commission under Article 10 DMA to exempt it from sharing with individual undertakings on public security grounds

Data scope

Data type Definition in the summary
Query data "queries entered by end users into Google Search on any access point, any modifications made by end users and Alphabet to initial queries, and any other metadata about the users and their queries"
View data "all URLs and the description of visual elements which are displayed on search engine result pages, served to end users in response to their queries and are viewed by them (i.e., they generate an impression), no matter their type, format, structure or tab in which they are included"
Ranking data Each element's absolute whole-page position, its relative ordinal position among elements in the same area, and "any other information on the URL's relative and absolute position, prominence and visibility on the results page"
Click data "all data on user interaction with the results page as displayed on the user's device, with any SERP element, including the type, timing, order and duration (where relevant) of user actions, or lack thereof"

Each category is defined "on par with the data collected by Alphabet for the purpose of optimising its OSE service." Alphabet must also exclude invalid traffic "applying the same methods and process it uses itself in its ordinary course of business."

The summary's view data wording covers all URLs on the results page "no matter their type." It does not mention that, according to PPC Land's earlier reporting on the full decision, paid-result URLs are excluded from the dataset and recipients may not use it to build advertising or profiling products.

Anonymisation

The summary lists five technical measures, which Alphabet must apply while releasing data daily. According to the text, Alphabet must "inter alia (i) suppress direct identifiers and other identifying attributes from search records, (ii) suppress long queries and queries that contain rare words and combination of words, (iii) generalise location and user interaction metadata to ensure indistinguishability among users, (iv) limit session data to sets of maximum three queries where each is a refinement of the previous query and (v) coarsen durations (e.g. click-back time, and hovering and viewing durations) into time intervals instead of precise time data."

The thresholds behind those measures appear in the full decision rather than the summary. As Desfontaines set them out, records are deleted when they contain an entity searched by fewer than 50 distinct users in the past year, at least 1,000 signed-in users must share the same region, device category and inferred language - a k-anonymity style floor - and queries longer than the 95th percentile for their language are removed.

The contractual layer is extensive. Recipients must process the data "within a segregated processing environment under their effective control," and are prohibited from "attempting re-identification (e.g., by linking the Search Data with auxiliary datasets), using the dataset outside of the purpose of optimising online search engine services, retaining the data beyond 13 months, and sharing the dataset onward with any third-parties." They must encrypt data at rest and in transit, and pass "an independent audit prior to accessing the Search Data and annual monitoring audits to maintain such access." The Commission defines the scope of those audits.

The decision draws on the joint Commission and EDPB guidelines on how the DMA and the GDPR interact. The summary cites the version published for public consultation.

Pricing

According to the summary, "absent comparable market-based benchmarks, FRAND pricing corresponds to incremental cost-based pricing, where eligible beneficiaries pay only the incremental costs incurred by Alphabet for sharing the Search Data, plus a reasonable return on capital employed for that purpose. Exceptional circumstances may justify an additional margin." The pricing and access sections of the summary still refer to "the proposed measures," even though the summary describes a decision already adopted.

How access works under the decision

According to the Official Journal summary, the decision sets out the path from application to data in the following sequence.

  1. Application and eligibility review. A third-party search engine applies, and Alphabet reviews its eligibility against the criteria above, under Commission supervision. The review "complements the annual audit process."
  2. Data transfer check. Alphabet verifies whether the data will be processed in the EEA or a country with an adequacy decision, and otherwise runs a transfer impact assessment.
  3. Testing samples. Alphabet must provide three types of testing sample: "one small testing sample provided free of charge, a synthetic data sample, and a 5 % sample of the Search Data subject to beneficiaries meeting all conditions for accessing Search Data."
  4. Pre-access audit. The recipient passes an independent audit before receiving the full data, with the scope and assurance objectives defined by the Commission.
  5. Pricing. Alphabet finalises its pricing terms and communicates them to applicants and then to eligible beneficiaries, under a process the decision specifies.
  6. Delivery. Alphabet shares the data "in daily batches and with daily data time stamps," via the method and latency it uses internally "to the greatest extent that is technically feasible within a timeframe of no less than seven days."
  7. Duration. The beneficiary chooses how long to receive data, "up to five years from the moment when the Search Data becomes effectively accessible."
  8. Ongoing control. Annual monitoring audits continue, Alphabet may request ad hoc audits outside the annual window, and it may refuse, suspend or terminate access "to ensure strict compliance with contractual measures."

The summary also provides for a timeline within which Alphabet must make compliant data available, without stating it. PPC Land reported in July that the anonymised dataset was due to be finalised in November 2026, with final pricing and the start of data sharing in January 2027.

How the decision was reached

The summary describes a six-month specification process that followed more than two years of talks. The Commission has held "an extensive regulatory dialogue" with Alphabet on Article 6(11) since October 2023. It opened proceedings under Article 20(1) on January 27, 2026, and on April 16, 2026 sent Alphabet its preliminary findings and published the proposed measures for consultation. The Commission "received over 140 contributions," including from business users and end users, and shared them with Alphabet.

Alphabet responded to the preliminary findings on May 18, 2026. The EDPB also weighed in: its letter of May 5, 2026, containing 33 comments on the measures, was later criticised by Brussels lawyer Peter Craddock as inconsistent with the board's own anonymisation guidance. The Digital Markets Advisory Committee issued a positive opinion on July 15, 2026, and the Commission adopted the decision the next day, within the six-month limit the regulation sets.

Why it matters for search marketing

The data-sharing order is an attempt to change who can compete for search queries in Europe, and the anonymisation rules decide how useful the data is. Google's share of European search is around 90 percent, as PPC Land noted in July. The decision's stated aim is to let independent search engines "optimise their own search services." Much of the data's value lies in rare queries, and rare queries are where anonymisation filters cut deepest.

Any change prompted by Woodway's findings would land in that zone. Stricter performance criteria for identifier detectors, or new protections against inference, could remove more queries or coarsen more metadata, reducing what rival search engines and AI chatbots receive. Looser filters would push the other way. Recipients have already voiced the opposite worry: DuckDuckGo said in October that Google is not externally audited on the filters and costs it reports, while recipients face independent audits before and during access, and that Google might apply the filters more aggressively than required.

The Commission left the door open. As PPC Land reported, the Commission said in July that it may amend the anonymisation measures in light of market developments, including independent third-party evaluations, and the decision provides for an evaluation within two years of adoption. An assessment commissioned by Google is the kind of evidence that process anticipates, which is why its framing matters. A conclusion that the Commission's approach can protect personal data "with appropriate safeguards" supports the Commission. The Woodway headline, which says the measures "need to be updated," points the other way.

For advertisers, the effect is indirect. The dataset is not a targeting tool, and paid-result URLs are excluded from it. But if better data helps rival engines and chatbot search products win queries, more European search activity could move to inventory that Google Ads does not sell. The pace of that shift depends on uptake, pricing and filtering decisions that are still being settled.

Who Woodway is

Woodway Assurance describes itself as "the Canadian technology company behind EviData, an automated, AI-enabled platform that helps organizations use data for AI, analytics and data sharing by assessing privacy risk, performing privacy-preserving data transformations and generating objective evidence." It is a spin-off from the Electronic Health Information Laboratory at the CHEO Research Institute, which El Emam founded and directs, and it builds on more than 20 years of research by El Emam and the laboratory team, according to the company. In 2026, the laboratory received the Privacy and Human Rights Award from the Global Privacy Assembly, the international forum of data protection authorities, and the human rights organisation Access Now, according to Woodway.

The full report is published on arXiv, according to the company. Woodway has scheduled a webinar on its findings for December 2 at 11 a.m. ET, 5 p.m. CET. The listed speakers are El Emam; Dr. Sophie Stalla-Bourdillon, co-director of the Brussels Privacy Hub at Vrije Universiteit Brussel; Dr. Rob van Eijk, principal at Team Blaeu; and Woodway data scientist Ghislain d'Entremont.

What is not stated

Woodway's announcement does not name the six countries, the time period of the search data tested, the volume of records, or which version of the anonymisation pipeline Google prepared for the test. It gives no figure for attribute or inference disclosure, does not say whether Google paid for the assessment beyond engaging Woodway, and does not identify the other external teams Google engaged.

Key facts

Item Detail
Case DMA.100209 - SP - ALPHABET - Article 6(11)
Decision Commission Implementing Decision of 16 July 2026 under Article 8(2) of Regulation (EU) 2022/1925
Official Journal summary "Summary of the Commission Decision of 16 July 2026 relating to a decision pursuant to Article 8(2) of Regulation (EU) 2022/1925", C/2026/4443, OJ C, 14.8.2026
ELI http://data.europa.eu/eli/C/2026/4443/oj
EUR-Lex reference of the summary CELEX 52026DMA1002093
Legal basis Regulation (EU) 2022/1925 (Digital Markets Act), Articles 6(11), 8(2), 8(5)-8(7), 20(1), 44; Commission Implementing Regulation (EU) 2023/814
Gatekeeper designation September 5, 2023
Proceedings opened January 27, 2026
Preliminary findings April 16, 2026
Consultation contributions Over 140
Alphabet response May 18, 2026
Advisory Committee opinion Positive, July 15, 2026
Woodway announcement "Woodway Assurance finds European Commission anonymization measures need to be updated to ensure Google search data has low privacy risk", Brussels, November 18, 2026, 07:00 GMT
Woodway report arXiv 2611.21689
Commissioned by Google, "alongside other external teams"
Collaborators University of Ottawa and others
Main test 98,501 targeted individuals, six countries, 0.094% identity disclosure, 3 hours and €105 per verified match, 0.47% threshold
Woodway webinar December 2, 2026, 11 a.m. ET / 5 p.m. CET

Timeline

Summary

Who: Woodway Assurance, a Canadian privacy risk company led by founder and CEO Dr. Khaled El Emam, working with the University of Ottawa, at the request of Google. The European Commission set the measures under review, and Alphabet is the gatekeeper bound by them.

What: Woodway's motivated intruder test re-identified 0.094% of 98,501 targeted individuals in Google search data used to test the Commission's anonymisation measures, below a 0.47% threshold it set in advance, at an average of 3 hours and €105 per verified match. Woodway said the data could be considered anonymised only after the Commission sets performance criteria for identifier detectors and adds protection against sensitive inferences.

When: Woodway published its findings on November 18, 2026. The Commission adopted the measures on July 16, 2026, and published the decision summary in the Official Journal on August 14, 2026. Data sharing is scheduled to begin in January 2027.

Where: The European Union and European Economic Area. The test covered data from six countries that Woodway did not name, and the announcement was issued in Brussels.

Why: Under Article 6(11) of the Digital Markets Act, Google must share anonymised search data with rival search engines, including AI chatbots with search functions, on FRAND terms. How strictly that data is anonymised determines both the privacy risk to users and how useful the data is to competitors, and the Commission has said it may amend the measures in light of independent evaluations.