TL;DR
- Google updated its European Search Dataset Licensing Program documentation on 31 August 2026, publishing the eligibility rules, sample datasets, audit regime and dates behind the Commission's July 2026 DMA decision.
- Licensing agreements begin dispatching from 17 September 2026, and all three data samples become accessible from 16 November 2026.
- Applicants need at least 50,000 average monthly EU users plus two consecutive years of search service in the EU, or over 50 million euros in capital investment as a recent entrant.
- The 5 percent sample and the full dataset require an independent Level 1 audit first, then Level 2 reports, with the first due within six months of access and later ones annually.
- The programme is scoped to the European Economic Area, names no applicants, publishes no fees, and changes nothing about how Google ranks results.
Google has published the operating rules for its European Search Dataset Licensing Program, on a documentation page last updated on 31 August 2026. The page covers how eligible competitors can access and license search ranking, query, click and view data from Search in the European Economic Area, and it names the eligibility thresholds, the sample datasets, the audit procedure and the dates.
Search Engine Journal reported the update, noting that licensing agreements begin dispatching from 17 September 2026 and that data samples become available from 16 November 2026. The programme implements the European Commission's binding decision adopted in July 2026 under the Digital Markets Act, which requires Google to share anonymised search data with competitors, including AI chatbots that qualify as online search engines under the law.
What was already known, and what is new
The obligation itself is not news. Relevant Audience published a post on 17 July 2026 titled "EU orders Google to share Search data with rival engines and AI chatbots", covering the Commission's binding decision when it was adopted. At that point the headline thresholds and the general timeline were already public, and the story was a regulatory one: a gatekeeper told to open a dataset it had built from years of user behaviour.
What the 31 August update adds is mechanics. The decision now exists as a process with dates attached, an application route, three defined sample datasets, and a two-level audit regime that decides which applicants can touch the larger files. Anyone who read the July coverage already knows why this programme exists. The new documentation answers a narrower question: what an applicant has to do, and by when.
That distinction is worth holding on to, because the two stories will be blurred together in summaries. July was the ruling. August was the paperwork. Nothing in the paperwork expands or narrows the ruling. It operationalises it.
What ranking, query, click and view data means in practice
The four data types named in the documentation describe different parts of the same search event. Query data is what people typed. Ranking data is the order in which Google's system placed results for those queries. Click data records which results were selected. View data records what was shown and seen.
Put together, that is a record of demand and of how a mature ranking system responded to it. A rival search engine starting from a small user base faces the hardest version of a cold-start problem. Without traffic it cannot learn which results satisfy which queries, and without good results it cannot attract traffic. Query and click data from a much larger service is one way out of that loop, because it supplies behavioural signal that normally only scale produces.
The same logic explains why the Commission's decision reaches AI chatbots. An assistant that answers questions by retrieving and ranking documents is doing search work, whatever the interface looks like. Under the DMA definition applied here, a chatbot that qualifies as an online search engine can apply on the same terms as a traditional engine. That is the part of the decision with the longest reach, and it is why the programme is being read by people working on AI search visibility rather than only by people who follow competition law.
Who is allowed to apply
Eligibility is where the programme narrows. An applicant must meet the DMA's definition of an online search engine and must target users within the EEA. It must not be controlled by non-EEA state actors, and it must not be subject to EU sanctions.
Then come the size tests. Every applicant needs an average of at least 50,000 monthly users of its search service in the EU over the past year. On top of that, it must have been providing search services in the EU for the past two consecutive years, or, if it is a recent entrant, have received over 50 million euros in capital investment. Google may request supporting documents, and says it will respond to expressions of interest within seven calendar days.
Read that list as a filter rather than as a formality, because it is the part of the programme that decides who actually benefits. Fifty thousand average monthly EU users is modest by consumer internet standards, but it is far above what a weekend project or a small vertical engine reaches. The two-year operating history removes most new entrants. The alternative route, more than 50 million euros of capital, replaces one form of scale with another. A well funded AI assistant clears that bar without difficulty. A small independent engine with a loyal but modest audience probably does not, unless it has already been running for two years. The remedy is pointed at credible competitors, and the documentation defines credible in terms of users, tenure or money.
The three samples, and what they cost
Eligible applicants can request samples before committing to full dataset access. There are three of them. A free sample contains 1,000 rows. A synthetic dataset covering up to 10 million queries is available for a fee, as is a 5 percent sample of the full dataset. All three become accessible from 16 November 2026.
Fees are set on fair, reasonable and non-discriminatory terms, usually shortened to FRAND. Under the Commission's measures, fees are limited to the incremental costs of providing the data plus a specified rate of return. The documentation does not publish a price. Anyone quoting a euro figure for this dataset is guessing.
The ladder has a practical shape. A thousand rows is enough to inspect a schema and understand what the fields contain. A synthetic set of up to 10 million queries is enough to build and test a pipeline against before paying for real behaviour at volume. Only after that does an applicant have to decide whether the 5 percent sample or the full dataset justifies what comes next.
The audit regime is the real price of entry
Access to the 5 percent sample and the full dataset starts with an independent assessment. The two smaller samples require no audit, although Google still checks eligibility in advance.
An independent auditor examines the applicant's technical setup, data storage and workflows. The Level 1 report provides assurance that the applicant has credible plans to use the data for its own search product, and that its data protection controls are suitably designed. Designed is the operative word. Level 1 asks whether the controls are built correctly, not yet whether they work.
Continued access depends on the second level. Recipients accept ongoing monitoring by an independent assurance practitioner and must submit regular Level 2 reports testing whether the controls operate in practice. Search Engine Journal reported that the first Level 2 report is due within six months of access being granted, with later reports due annually.
Two further obligations sit around the programme. Google will have to maintain a public webpage listing the third-party search engines that access the dataset, and the Commission plans to review the measures every two years.
For a company weighing this up, the audit is the cost that never appears on an invoice from Google. Commissioning an independent auditor, documenting storage and workflows, then repeating an assurance exercise every year is a standing compliance function with staff attached to it. That is why the sample ladder matters more than it first appears. A prospective applicant can look at the free rows and the synthetic set first, and only then decide whether the full dataset is worth building an audit programme around.
The dates worth putting in a calendar
The documentation and Search Engine Journal's report give a short set of fixed points, and they are the only dates the source states.
| Date | What happens |
|---|---|
| 31 August 2026 | Google's programme documentation page last updated |
| 17 September 2026 | Licensing agreements begin dispatching |
| 16 November 2026 | All three data samples become accessible |
| Within six months of access | First Level 2 assurance report due, then annually |
| Every two years | European Commission reviews the measures |
What the documentation does not say
The page is scoped to the European Economic Area. It says nothing about Thailand or any other non-EEA market, and nothing in it suggests the dataset covers searches made outside the EEA.
It names no applicants. No company has been confirmed as having applied, been approved or been refused. It publishes no fee figures, only the principle the fees are set under. And it makes no claim about whether the data will change what any AI assistant shows a user, or when.
Nothing here changes ranking either. This programme does not alter how Google orders results, in Europe or anywhere else. It changes who else can obtain a record of how Google ordered them.
What this means for Thai marketers
Directly, nothing. A brand in Bangkok gains no access, no data and no ranking movement from this programme. Anyone pitching an EEA data angle to a Thai client is selling something the source does not support.
The interesting reading is second-order, and it is reasoning rather than reported fact. If qualifying engines and assistants obtain real query and click data from November onwards, the products built on that data could become better at working out what a query means and which pages answer it. Better retrieval inside an assistant would eventually show up as different citations in AI answers, including answers seen by users in Thailand, because these products are rarely built market by market. That chain has several unproven links in it, and Google's documentation says nothing about any of them. It is a hypothesis to watch, not a plan to act on.
The practical response is unchanged. Search visibility work in Thailand still runs on the same inputs it did last month: pages that answer real queries, technical hygiene, and content that survives being quoted out of context by a machine. Teams building for assistant surfaces should keep that work inside generative engine optimisation, which behaves the same way this week as last. Local ranking work under SEO in Thailand is untouched by a European licensing programme.
One thing worth tracking is the public list Google has to maintain. If it fills up with qualifying AI assistants rather than only traditional engines, that is the first observable sign that the remedy reaches the products most people now use to ask questions. If it stays short, the eligibility bar did the work critics expected it to do.
Frequently asked questions (FAQ)
What did Google publish on 31 August 2026?
Google updated the documentation page for its European Search Dataset Licensing Program, setting out eligibility rules, the sample datasets, audit procedures and key dates. The page covers licensing of search ranking, query, click and view data from Search in the European Economic Area.
Can an AI chatbot licence this data?
Yes, if it qualifies as an online search engine under the Digital Markets Act and meets the same eligibility tests as any other applicant. The European Commission's July 2026 decision explicitly covers AI chatbots that fall inside that legal definition.
How much does the dataset cost?
Google has not published a price. The documentation states that fees are set on fair, reasonable and non-discriminatory terms and, under the Commission's measures, are limited to the incremental costs of providing the data plus a specified rate of return.
What is the difference between a Level 1 and a Level 2 report?
Level 1 is a one-off independent assessment before access, confirming that the applicant has credible plans for the data and that its data protection controls are suitably designed. Level 2 reports come afterwards and test whether those controls work in practice, with the first due within six months of access and later ones annually.
Does this change rankings or traffic for a Thai website?
No. The programme is scoped to the European Economic Area, changes nothing about how Google ranks results, and gives a Thai brand no access to anything. Any effect on AI answers seen in Thailand would be indirect, unproven, and is not claimed anywhere in the source.
If your team is trying to work out which of these regulatory stories actually touches your search performance and which are background noise, that is a reasonable thing to get a second opinion on before rewriting a strategy around a headline.
