Analysis: AI wrote almost a quarter of Belgian parliamentary questions
An analysis of 491 questions classified 22.5 per cent as entirely AI-generated, but detection is not firm evidence of authorship.
.jpg){kind=link}
Almost a quarter of the written questions examined from Belgian parliamentarians contained text that, according to a media analysis, was entirely AI-generated. The result draws attention to the responsibility of elected representatives, but does not automatically indicate how much thinking was delegated to a chatbot.
The analysis examined 491 written questions in Dutch to the Belgian federal government. According to reports, the detector used, Pangram, classified 22.5 per cent as entirely AI-generated and 20.4 per cent as partly AI-generated.
Answers from ministers were also examined. These reportedly contained traces of AI, either wholly or partly, in 7.7 per cent of cases. The analysis was conducted on documents from 2022 through 2026 and compared older parliamentary documents with newer questions.
The figures should be read with caution. AI detectors identify patterns in language, but cannot establish with certainty who wrote a text or how much human checking preceded it. A classification as ‘AI-generated’ is therefore not proof that a parliamentarian did not devise or check the content themselves.
Parliamentarians use AI in different ways. Some deploy systems for translations, searching through legislation or improving language. Others use them when drafting a first version. The difference between editorial assistance and outsourcing content is legally and politically important, because MPs remain responsible for their questions.
Speaker of the House Peter De Roover warns that AI could make it easier to submit large numbers of questions. If questions are produced faster than ministers and civil servants can answer them, a growing volume of documents of little significance could place the parliamentary right to ask questions under pressure.
At the same time, written questions are not a competition in prose. A well-formulated question can also be produced with digital tools, provided the parliamentarian understands the facts, purpose and consequences. Transparency about the use of AI may therefore be more important than an absolute ban.
The Belgian debate is part of a broader shift within government. AI can speed up administrative work, but makes it harder to see who drafted a text and who bears responsibility. Rules will therefore probably need to address not only the use of software, but also oversight, attribution of sources and ultimate human responsibility.
One story, several perspectives
What is established
- A media analysis examined 491 Belgian parliamentary questions.
- An AI detector classified 22.5 per cent as entirely and 20.4 per cent as partly AI-generated.
- The analysis was not an official audit of the Belgian parliament.
- Parliamentarians remain responsible for their parliamentary work.
Left
Arguments AI can make public institutions more accessible and efficient, but human oversight and transparency must remain compulsory. Citizens should be able to know when policy texts and answers have largely been drafted by systems.
Values Democratic oversight, accessibility, workers’ rights and protection against opaque power.
Consequences Greater openness may increase trust, but additional rules could disadvantage smaller groups with less support.
Centre
Arguments What matters most is not whether a chatbot was used, but whether the parliamentarian checks the content and is politically accountable for it. Practical guidelines, source checking and recording AI use are likely to be more effective than a total ban.
Values Responsibility, practicality, quality and institutional continuity.
Consequences A proportionate framework could permit useful applications without parliamentary processes being overwhelmed by unchecked text.
Right
Arguments Elected representatives are chosen to study issues and ask questions themselves. Widespread use of AI could hollow out the right to ask questions and increase bureaucracy; rules or disclosure are therefore justified.
Values Personal responsibility, efficient government, institutional authority and limiting bureaucracy.
Consequences Stricter standards could limit the number of questions, but also discourage useful digital assistance.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The percentages and responses have been confirmed by several public media outlets. The text avoids the incorrect conclusion that a detector can establish authorship or outsourced thinking beyond doubt.
- confirmed The analysis examined 491 written parliamentary questions in Dutch. — This figure is mentioned by The Brussels Times and Anadolu Agency. source
- confirmed 22.5 per cent was classified as entirely AI-generated. — The same media analysis reports this percentage. source
- confirmed 20.4 per cent was classified as partly AI-generated. — Reported by The Brussels Times and VRT Nieuws. source
- confirmed AI detection does not automatically prove who wrote a text. — This is a methodological limitation of detectors and is expressed cautiously in the article. source
Editor's note
The percentages come from a media analysis and an AI detector, not from an official parliamentary audit. The text therefore carefully distinguishes between classification, actual use and responsibility.Sources
- Belgian lawmakers are increasingly using AI to write questions for government — The Brussels Times
- Nearly quarter of Belgian parliamentary questions generated entirely by AI — Anadolu Agency
- Kwart van parlementaire vragen door AI geschreven — VRT Nieuws via Headliner
More on this in Dutch media
- Trouw — „kunstmatige intelligentie”
- NU.nl — „kunstmatige intelligentie”
- De Telegraaf — „kunstmatige intelligentie”