Quarter of Belgian MPs’ Questions Written by AI

Nearly a quarter of the written questions that Belgian members of parliament submit to the federal government are generated entirely by artificial intelligence, according to an investigation published on 23 September. A further one in five questions is partly AI-written, and the practice is rising sharply, prompting the speaker of the Chamber of Representatives to warn that it risks hollowing out the legislature.

The findings, first reported by Het Nieuwsblad and carried nationally by VRT NWS and HLN, place Belgium at the centre of a fast-spreading European debate over how generative AI is reshaping the day-to-day work of democratic institutions.

A Quarter of Questions Generated by AI

The investigation analysed hundreds of Dutch-language written questions filed with the Chamber between 2022 and early 2026, using the AI-detection tool Pangram. Of a recent batch of 491 questions, 22.5 percent were fully generated by AI, while 20.4 percent were partly AI-generated. Only 57 percent were entirely human-written, HLN reported.

Ministerial answers were also affected: 7.7 percent were fully or partly written by AI. A control sample drawn from hundreds of questions and answers submitted in 2022 was found to be entirely human-written, indicating that the practice is recent and spreading rapidly rather than a long-standing feature of parliamentary work.

De Roover Warns of a Hollowed-Out Parliament

Chamber President Peter De Roover, of the Flemish nationalist N-VA, has become the most prominent voice of concern. In an interview on VRT NWS’s De Ochtend programme, he described the trend bluntly: “It is growing exponentially.”

De Roover warned that AI makes it “very easy to submit a whole series of questions that are not really meaningful and serve to score points in media rankings that track who asks the most questions.” He described the shift as the work of a “fast-growing minority” of MPs, and in remarks reported by Gazet van Antwerpen framed the stakes in stark terms: “If AI takes over the parliament, we can close the tent.”

The speaker acknowledged the difficulty of banning the technology outright. “Impossible. The whole world uses it, so why should parliament not be allowed to use AI? ChatGPT is also an encyclopedia, a search engine, a translation computer,” he said. But he left the door open to procedural restrictions if the volume of questions continues to overwhelm the system. “The right of parliamentary questions is sacred, but if we see the system clogging up because too many unnecessary questions are being asked, we must sit around the table to tackle it,” he said.

MPs Defend Their Use of AI

Members of parliament have offered sharply differing accounts of how they use the tools. Arthur Orlians, of the recently rebranded Anders party, confirmed he uses AI but not to draft questions. “I use it, like so many others… I use it for translations and to plough through legal texts, but not to draft questions,” he told VRT NWS. He warned that the real danger is an MP who does not understand his own interventions: “There is nothing more harmful than an MP who does not understand what he is intervening on and reads from a piece of paper.”

Eva Demesmaeker, also of N-VA, said she uses AI only at the end of the drafting process to check for ambiguities, spelling and language errors, describing it as an extra staff member that lightens the workload. “I hope everyone learns to work with it sensibly and that rules are not necessary,” she said. Vincent Van Quickenborne, of Anders, admitted to using AI “occasionally” and “in a hybrid way.”

Orlians pushed back on the notion that AI represents an unprecedented shortcut, noting that parties have long relied on internal study services to prepare questions and speeches. “Previously they had questions prepared by the study service. Is that so much better?” he asked. He opposed new rules outright: “There are already enough rules. Every MP does their job as they see fit. The voter will determine whether he or she did it well or not.”

A Cross-Border Pattern

The Belgian findings mirror a parallel investigation in the Netherlands. A NOS analysis of 2,500 documents containing more than 28,000 individual Dutch parliamentary questions found that a quarter were at least partly written with AI, rising to 26 percent in the most recent six-month period from 17 percent the half-year before.

In the Dutch case, AI developer Maarten Sukel told NOS that while large language models can help legislators frame better questions faster, the risk lies in outsourcing the thinking itself. “If you are working on a dossier you know less about and you let AI help you come up with parliamentary questions, that is a risk. You have to keep immersing yourself and not hand over the thought process,” he said. Tech thinker Marleen Stikker framed the deeper concern as one of sovereignty: “We are outsourcing our democratic systems to American AI companies” whose models, she added, are “not neutral, but coloured.”

Analysis: The Question of Accountability

The Belgian episode exposes a tension that parliaments across Europe will increasingly confront. Generative AI is now a routine tool of professional life, and legislators argue, reasonably enough, that refusing to use it would leave them worse equipped than the citizens and businesses they oversee. De Roover himself concedes that a ban is unworkable.

Yet the investigation points to a subtler problem than simple cheating. The volume of written questions has long been a proxy for parliamentary activity, and media league tables reward MPs who file the most. When drafting becomes nearly costless, that incentive structure produces a flood of low-value questions that ministers cannot answer at pace — a dynamic Demesmaeker confirmed when she noted that some committee questions remain unanswered for months because ministers “really receive too many questions.”

The question of accountability is equally pointed. If an MP submits a question generated by a tool he does not fully understand, the constitutional right of parliamentary oversight risks becoming a formality rather than a genuine act of scrutiny. As the GVA analysis put it, generating questions with AI displays “either intellectual laziness or intellectual incompetence.”

What to Watch

No formal rules currently govern the use of AI in the Belgian Chamber, and the leading figures interviewed are divided on whether any are needed. De Roover has signalled he is prepared to discuss procedural limits if the system continues to clog, while Demesmaeker hopes voluntary restraint will suffice and Orlians rejects regulation entirely.

The next test will be whether the upward trend in AI-generated questions continues through the coming parliamentary term — and whether the Netherlands’ parallel findings push both countries toward common disclosure standards. For now, the Belgian investigation stands as one of the clearest data points yet that generative AI has quietly entered the machinery of representative democracy.