Investigation: 68 MPs read out AI-generated texts
An analysis of contributions to debates indicates chatbot use by several parliamentary groups, but detection provides no conclusive proof.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
.jpg){kind=link}
At least 68 MPs this year read out texts that, according to an AI detector, had largely been written by a chatbot. Investico and De Groene Amsterdammer report this on the basis of their investigation.
The researchers analysed 2,715 prepared opening contributions from 303 plenary and committee debates between 1 January and 30 September. Only texts that, according to the Pangram detector, had been written with AI to more than half counted. Interventions and motions were left out of consideration.
In 187 of the 303 debates examined, the analysis found that at least one MP had delivered a contribution meeting that threshold. In total, this involved 297 contributions. The marking does not prove that a text was largely or entirely written by a chatbot: according to the description of the method, the detector identifies language patterns that occur more often in AI-generated texts.
According to the investigation, use was spread across several parliamentary groups. Former Pro MP Esmah Lahlah, now an alderman in Amsterdam, topped the list. She was followed, among others, by PVV MP Rachel van Meetelen, D66 MP Fatimazhra Belhirch and VVD MP Wendy van Eijk. According to the analysis, BBB had the highest proportion of marked contributions: about 30 per cent.
According to Investico, many MPs approached by the researchers said they mainly used AI to shorten their own texts, make them easier to understand or improve them linguistically. CDA MP Harmen Krul told Investico that he used AI in preparation, but did not want to create the impression that his contributions came entirely from a chatbot. According to Investico, other MPs said they had since stopped using it.
The researchers note that AI-marked texts more often contain certain stylistic patterns, such as lists of three elements and formulations involving a contrast. Professors of language processing and political communication cited in the investigation warn that the issue is not only about style: when a model selects information or shortens an argument, it can also influence the political judgement involved.
The investigation comes at a time when the use of AI in politics is already being debated more frequently. The NOS previously reported that about a quarter of written parliamentary questions had been drafted with the help of AI. The new analysis concerns spoken contributions to debates and, according to the researchers, shows that the discussion is not only about efficiency, but also about the responsibility, oversight and recognisability of elected representatives.
One story, several perspectives
What is established
- Investico and De Groene Amsterdammer analysed prepared contributions to debates in the Lower House (Tweede Kamer).
- 68 MPs and 297 contributions exceeded the chosen AI-detection threshold.
- The method used can identify language patterns, but does not independently establish who wrote a text.
- MPs say they use AI, among other things, for shortening and editing.
Left
Arguments AI can reduce workloads, but elected representatives must make their substantive choices themselves. If commercial models determine which arguments are emphasised or removed, this may particularly disadvantage groups that already have less political power.
Values Democratic oversight, human responsibility, equality and transparency.
Consequences The left fears an impoverishment of debate and calls for clear rules, disclosure and adequate support for MPs.
Centre
Arguments One possible centrist assessment is that AI can be a legitimate tool for language editing, as long as MPs check the content and retain political responsibility. Guidelines, training and verifiable agreements are appropriate here rather than a total ban.
Values Carefulness, practicality, reliability and continuity of parliamentary work.
Consequences The centre stresses that rules should be proportionate: distinguish between editing and having political positions generated.
Right
Arguments One possible right-wing assessment is that MPs must be answerable for their own words, but that technology should not be distrusted in advance. A ban could disadvantage smaller parliamentary groups with fewer policy staff that use AI to cope with workloads.
Values Individual responsibility, efficiency, freedom of choice and equal access to tools.
Consequences The right fears bureaucratic restrictions and mainly wants action against factual errors, deception and outsourcing political judgement.
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 key figures and research method can be found in Investico’s publication and are described independently by NOS and NL Times. The text explicitly presents AI detection as an indication rather than conclusive proof.
- confirmed The analysis included 68 MPs whose contributions to debates were, according to Pangram, largely written by AI. — Mentioned by Investico and NOS. source
- confirmed The researchers analysed 2,715 contributions from 303 debates between 1 January and 30 September 2026. — The research method and figures are given in Investico’s methodology statement. source
- confirmed In 187 of the 303 debates, at least one contribution was classified as largely AI-written. — Investico research note. source
- confirmed According to the analysis, about 30 per cent of BBB’s contributions to debates were AI-marked. — Mentioned by Investico and NOS. source
- confirmed AI detection does not prove that a text was entirely written by a chatbot. — This is a methodological qualification that follows from the detection method described and the threshold chosen. source
Editor's note
It is certain that Investico and De Groene Amsterdammer analysed 2,715 contributions to debates and that 297 contributions exceeded their AI threshold. It remains uncertain whether the marked texts were actually written by AI, because detection software provides no conclusive proof; official parliamentary data on AI use are lacking.Sources
More on this in Dutch media
- de Volkskrant — „tweede kamer”
- RTL Nieuws — „tweede kamer”
- NOS — „tweede kamer”