Part of the AI Regulation News hub.
A Commission implementing regulation adds ten mandatory generative AI variables to the EU household ICT survey, four on whether and where people use the tools and six on why non-users do not, so that from reference year 2027 every member state collects comparable official statistics
Almost nobody reads statistics implementing regulations, which is why this one is worth reading. It decides what the European Union will officially know about generative AI adoption, and the categories it chose are a statement about what the Commission thinks matters.
Bottom line: A binding implementing regulation specifying the technical items of the data set for the EU survey on ICT usage in households. Ten variables cover generative AI: four on use, being any use, private use, work use and use in formal education; and six recording why a non-user did not use the tools.
Who this affects: National statistical institutes, which must collect and transmit these variables; and anyone who relies on official comparable figures for AI adoption in the EU, including policymakers, market analysts and anyone drafting an impact assessment.
Issue date: Adopted 29 July 2026. Enters into force on the twentieth day following publication in the Official Journal. Covers reference year 2027.
What changed: Generative AI use, and non-use, become mandatory official statistics across the EU with defined categories, rather than something measured inconsistently or by private surveys.
Analysis: The disaggregation is the substance, and the non-use side is the part to notice. Splitting use into private, professional and formal education separates the three policy debates. But the regulation also requires every non-user to be asked why, with data protection, privacy, security or safety concerns and ethical concerns as named options. That turns refusal into an official statistic, which is a different and more politically consequential number than adoption.
Primary sources: Commission Implementing Regulation (EU) 2026/1864, EUR-Lex
- Instrument
- Commission Implementing Regulation (EU) 2026/1864 specifying the technical items of the data set, establishing the technical formats for transmission of information and specifying the arrangements and content of the quality reports on the organisation of a sample survey in the use of information and communication technologies domain
- Citation
- Regulation (EU) 2026/1864; document reference C/2026/5305; CELEX 32026R1864
- Authority
- European Commission
- Legal basis
- Regulation (EU) 2019/1700 of the European Parliament and of the Council, the framework regulation for European statistics on persons and households
- Jurisdiction
- European Union, with EEA relevance
- Status
- Adopted 29 July 2026. Enters into force on the twentieth day following publication in the Official Journal.
- Bindingness
- Binding in its entirety and directly applicable in all Member States.
- Reference year
- 2027
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1864
The four variables
The regulation sets out the technical items of the data set for the household survey on the use of information and communication technologies. Buried in a long schedule of variables are four that concern generative AI, each recorded as ticked, not ticked or not applicable.
IUAI records internet use of any generative artificial intelligence tools in the last three months. It is the headline adoption measure.
IUAIPR records use of generative AI tools in the last three months for private purposes, and is asked only of respondents who answered yes to IUAI. IUAIWP records use for professional, meaning work, purposes. IUAIFE records use for formal education, such as school or university.
A further reference appears elsewhere in the schedule. The variable on online learning material, IUOLM, expressly includes AI tools when used specifically for learning among the examples of material other than a complete online course, alongside video tutorials, webinars, electronic textbooks and learning apps.
The six variables about people who do not use it
The adoption items are only half of the generative AI module, and the other half is the more unusual one. Six further variables record the reasons a respondent did not use generative AI tools in the last three months, and they are put to individuals where IUAI equals 0.
They are: the respondent did not know that such tools existed; had no need; did not know how to use them; had concerns about data protection, privacy, security or safety; had ethical concerns; or had another reason. None of the six is marked optional.
The ethical concerns item is defined rather than left open, and the definition is worth quoting because it tells you what the Commission thinks the objections are: perpetuating of stereotypes, no accountability, and unauthorised use of intellectual property.
Measuring refusal, and its reasons, is a deliberate choice. An adoption figure supports a narrative about uptake; a refusal figure disaggregated into ignorance, no need, lack of skill, privacy concern and ethical objection supports arguments about why uptake is what it is. From 2027 those arguments will have official numbers behind them, which is a different position from the one they occupy now.
Why a statistics regulation is worth reading
Instruments like this are usually treated as administrative plumbing, and they are rarely covered. That is a mistake in this instance, because measurement decisions determine what later arguments can be made with evidence.
Until now, claims about generative AI adoption in Europe have rested on vendor telemetry, private survey panels and national exercises using incompatible definitions. Those sources are not comparable across countries and are not neutral.
From reference year 2027, member states will be collecting the same variables, with the same recall period of three months, under a binding instrument, transmitted in a common format with quality reports specified by the same regulation.
The practical consequence is that arguments about AI adoption in the EU acquire an official denominator. Anyone drafting an impact assessment, a consultation response or a national strategy after that point will be expected to work from these figures.
What the categories imply
The choice of disaggregation is not neutral, and it maps onto the live policy questions rather precisely.
Separating professional use from private use allows the workplace debate to be measured on its own terms: the questions about employment, productivity, worker consultation and workplace monitoring all turn on how many people use these tools at work, which no general adoption figure can answer.
Separating formal education use is the more pointed choice. Whether students use generative AI, and how many, is contested in every member state, and it is a question that institutions currently answer with anecdote and academic integrity statistics. A national figure collected on the same basis as every other country's changes the character of that debate.
What is not there is also informative. The variables record whether a person used generative AI and in which broad context. They do not record which tools, how often, for what tasks, or whether the use was disclosed. This is an adoption measure, not a behavioural one, and it should not be cited for propositions it cannot support.
What to do with this now
For a national statistical institute the obligation is direct: these items must be collected for reference year 2027 and transmitted in the specified format, with quality reporting as the regulation sets out.
For everyone else, the useful step is calendar planning. Results from a reference year 2027 survey will appear during 2027 and 2028, and any strategy document, market model or policy submission that will still be live then should anticipate being measured against them.
There is also an evidential point worth noting now. Where an organisation makes public claims about AI adoption in a European market, those claims will shortly be checkable against an official comparable series. Figures asserted today have a verification date attached to them, whether or not the person asserting them realises it.
The instrument itself is binding in its entirety and directly applicable in all member states, so there is no transposition step and no national discretion about whether to collect these variables.
The substance of this regulation is a measurement decision with a date on it. From reference year 2027 every member state must collect ten generative AI variables in the household ICT survey. Four cover use: any use in the last three months, private use, professional use and use in formal education, with AI tools also named among online learning materials. Six cover non-use, asking every non-user whether they did not know such tools existed, had no need, did not know how to use them, had data protection, privacy, security or safety concerns, had ethical concerns, or had some other reason. That gives the EU an official comparable series on both adoption and refusal, where previously there were only vendor figures and incompatible national surveys. Note the limits before citing it: these record no tool, frequency, task or disclosure. The instrument is binding in its entirety and directly applicable, adopted 29 July 2026 under Regulation (EU) 2019/1700, entering into force twenty days after publication in the Official Journal.
Source File
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32026R1864
Open the EUR-Lex page for CELEX 32026R1864 and confirm four things: the title identifying Commission Implementing Regulation (EU) 2026/1864 of 29 July 2026 on the data set for the sample survey in the use of information and communication technologies domain, adopted under Regulation (EU) 2019/1700; the reference to reference year 2027; in the schedule of variables, the four generative AI items IUAI, IUAIPR, IUAIWP and IUAIFE with their definitions and the condition that IUAIPR is asked only where IUAI equals 1; and the closing formula that the regulation is binding in its entirety and directly applicable in all Member States.
IUAI Internet use of any generative artificial intelligence (AI) tools in the last 3 months. Commission Implementing Regulation (EU) 2026/1864 of 29 July 2026, data set variables
FAQ
What does this regulation actually do?
It specifies the technical items of the data set, the technical formats for transmission and the arrangements for quality reports for the EU household survey on the use of information and communication technologies, for reference year 2027.
Which generative AI variables does it add?
Ten. Four on use: IUAI for any use in the last three months, IUAIPR for private purposes, IUAIWP for professional or work purposes and IUAIFE for formal education. Six on non-use, put to those who answered no to IUAI: IUAINUUNK did not know they existed, IUAINUNN no need, IUAINUUSE did not know how to use them, IUAINUSEC data protection, privacy, security or safety concerns, IUAINUETH ethical concerns, and IUAINUOTH another reason.
What counts as an ethical concern?
The regulation defines it in the variable itself, giving as examples perpetuating of stereotypes, no accountability, and unauthorised use of intellectual property.
Is it binding?
Yes. The regulation states that it is binding in its entirety and directly applicable in all Member States, and it enters into force on the twentieth day following publication in the Official Journal.
When will the data appear?
It covers reference year 2027, so collection takes place in that year and results follow in the usual publication cycle afterwards.
What does it not measure?
It does not record which tools were used, how often, for what tasks, or whether use was disclosed. These are adoption variables recording whether a person used generative AI and in which broad context.
Why does the disaggregation matter?
Because it separates work use, private use and formal education use, which are the three distinct policy debates. A single adoption figure cannot answer any of them; three comparable figures across member states can.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.