EU Forces Google to Share Search Data as AI Reshapes Competition

The European Commission is moving to compel Google to hand over the ranking signals, query logs, and click data powering its search results, making that information available to rival platforms and, according to proposed measures summarized by ITIF, potentially to AI chatbot services as well. The proceeding opened in January and has since grown into a dispute about whether the data regulators want to open up still describes the thing worth competing for.

The product Google is building looks less and less like the one the Digital Markets Act was written around. AI Overviews, agentic checkout, query fan-out across live data sources these are not incremental updates to the blue-link model. The DMA's data-sharing obligation targets behavioral signals from that old model. Whether those signals reach the new competitive bottleneck is the central question the proceeding has not resolved.

The specification proceeding is ongoing. What follows covers what the Commission is proposing, what Google has actually shipped, and why the competitive picture is harder to read than either side suggests.

What the Commission opened in January

The Commission opened a formal specification proceeding in January 2026, tied to its prior designation of Google as a gatekeeper and its search platform as a core platform service under the DMA. Article 6(11) of that regulation requires Google to share ranking, query, click, and view data with rival search engines on fair, reasonable, and non-discriminatory terms, per ITIF's formal comment filed three weeks ago in opposition to the Commission's proposed measures.

Earlier this month, the Commission issued detailed proposed measures covering eligibility, data scope, anonymization, pricing, and data acquisition terms, per the same filing. ITIF submitted those comments against the proposal. Where the filing characterizes what the Commission's measures require, that framing reflects a critic's reading though the underlying procedural facts are consistent across the available record.

What EU forces Google to share under Article 6(11)

The parity principle sits at the core of the proposed measures. According to ITIF's summary, Google would be required to share all query, view, click, and ranking data it collects to optimize its search services, covering both paid and organic results, with no degraded or lower-frequency feed offered to competitors.

That data would need to be delivered via API or equivalent tool, at the same frequency Alphabet accesses it internally, for a minimum of five years, per ITIF's filing. On pricing: Google would charge recipients the incremental cost of making the data available plus a reasonable return, with smaller companies capped at incremental cost only.

The recipient definition is notably broad. The proposed measures explicitly state that Google shall not exclude AI chatbot services from eligibility, provided they meet the regulatory definition of an online search engine "even if the OSE is provided as part of a broader service," per ITIF's account. That language could bring services like ChatGPT Search into scope, depending on how regulators apply the classification.

For a traditional search challenger, the dataset has obvious value: years of behavioral signal on what users clicked, what they skipped, and how content performed against real queries. For an AI-native competitor building a conversational product, that signal is useful input. Whether it addresses the right bottleneck is a different question.

Why AI changes the stakes for Google search data sharing

The product at the center of this dispute looks significantly different from the one Article 6(11) was written around.

AI Overviews, now used by more than one billion people globally, place an AI-synthesized answer above the traditional results list, according to Google's March 2025 announcement. The ten blue links are still there. They're just no longer the first thing a user encounters.

AI Mode goes further. Rather than returning links, it uses what Google calls a "query fan-out" issuing multiple related searches concurrently across data sources including the Knowledge Graph, real-time shopping inventory, and live web content then synthesizes the output into a structured response, per the same announcement. Google introduced AI Mode as an experimental Labs product in March 2025 and began rolling it out broadly in the U.S. a year ago. At I/O 2025, Google announced agentic capabilities including event tickets, restaurant reservations, and local appointments described as coming to users in the weeks and months following the announcement, with an agentic checkout feature available for users who opt in through Google Pay, per Google's I/O 2025 update.

The commerce layer has moved beyond announcements. Google's Universal Commerce Protocol enables checkout directly inside Search and the Gemini app, with Etsy and Wayfair already live for U.S. shoppers and Shopify, Target, and Walmart announced as coming soon, according to Google's Ads and Commerce blog from three months ago. Google describes this as agentic commerce becoming reality, not a roadmap item.

The data the DMA requires Google to share captures how the old search engine behaved. A competitor that acquires it could improve a traditional blue-link ranking system. It does not transfer Google's query fan-out infrastructure, its Knowledge Graph, its Shopping Graph, its Gemini models, or the closed-loop behavioral signals generated inside AI Mode interactions. The Commission's proposed remedy may target a bottleneck that has already moved.

The privacy tradeoff at scale

The proposed measures would require Google to modify shared data to reduce re-identification risk and to layer administrative, contractual, and organizational safeguards on top of any technical anonymization, per ITIF's filing. The intent is user protection. The structural problem is that the data's competitive value depends on its granularity, and those two goals pull against each other.

ITIF's filing submitted as a formal comment opposing the proposal's scope illustrates the identification problem: a search for a rare medical condition submitted from a small town can be effectively identifying even without a name attached, because context alone narrows the field. That framing comes from a critic of the proposal. The underlying technical tension, though, is not invented.

The same filing argues that privacy-enhancing techniques such as k-anonymity substantially degrade the dataset, and claims Google's own DMA-related privacy filters have already excluded 99% of queries from what's currently being shared, per ITIF. That figure comes without disclosed methodology or independent corroboration. If it is directionally accurate, it shows how quickly the privacy-utility tradeoff can render a mandated dataset commercially thin.

The scale itself follows from the proposal's design. ITIF characterizes the proposed measures as contemplating the daily sharing of hundreds of millions of Europeans' sensitive behavioral data with third parties, sustained for at least five years, across a recipient pool that may include AI chatbot services alongside conventional search engines. That is a meaningful exposure profile even if every safeguard works as intended.

Whether AI competition makes the remedy more or less necessary

The market the DMA is trying to contest is already under competitive pressure. Google's U.S. search market share fell below 90% in 2024 for the first time since 2015, with independent forecasts projecting the traditional search market could shrink by 25% by 2026, per ITIF's filing. Those figures are reported through ITIF, not sourced directly, and the market definitions are unclear. Read as directional, they suggest the regulation is arriving as the segment it targets is already contracting.

OpenAI's CFO was quoted in ITIF's filing as saying AI is "blowing open the search markets," pointing to the company doubling its search share from roughly 6% to 12% in six months. The methodology behind those figures is not disclosed. The U.S. court overseeing Google's antitrust remedies reached a parallel conclusion independently, writing that "the emergence of GenAI changed the course of this case" and naming it "front and center as a nascent competitive threat," per ITIF's account.

Google's engagement numbers point in a different direction for its own trajectory. AI Overviews are used by more than one billion people, and in the U.S. and India, queries where AI Overviews appear are driving over 10% higher usage for those query types, per Google's I/O 2025 update. Those numbers come from Google and are not independently audited, but the direction is consistent with a platform deepening engagement even as it loses share in traditional keyword search. Gemini 3 became the default model for AI Overviews globally earlier this year, adding follow-up questions directly from search results and a more fluid transition into conversational AI, per the Google Search blog.

Both dynamics are real. AI is generating organic competitive pressure on the blue-link model, which complicates the case for regulatory intervention. At the same time, Google is using its existing advantages brand, distribution, data scale, and the financial capacity to deploy frontier models to extend its position into the AI era rather than cede it. The DMA's data-sharing remedy targets one dimension of that picture. Whether it reaches the dimension that matters most remains unanswered.

What happens next

Several consequential questions remain unresolved as the proceeding continues. Whether AI chatbot services will ultimately be classified as eligible recipients. What privacy thresholds will be set and how they interact with the dataset's commercial utility. Whether the incremental-cost pricing structure will make the data economically viable for mid-sized competitors to actually use, rather than simply access.

The gap between the behavioral dataset the Commission is opening up and the infrastructure currently driving Google's competitive position the models, the transaction layer, the closed-loop signals from AI Mode interactions is the thing to watch as those questions get answered. The specification proceeding is ongoing. That gap already exists.