'Ook veel van wat er in de Tweede Kamer wordt gezegd, komt van AI'

The use of artificial intelligence tools like ChatGPT and Claude by Members of the Dutch House of Representatives for social media posts and parliamentary questions has been identified in a study. The research involved scanning the spoken texts of 150 members using the Pangram program, which found that 68 members used AI tools to generate parts of their contributions over the past year. The study excluded contributions such as interruptions and motions from the count.
The distribution shows that the use of chatbots is popular across the chamber, particularly among right-wing factions like the PVV and Groep Markuszower, while parties like the SP, PvdD, and Volt reported no usage. Some members used AI for maidenspeeches, and in committee debates, a majority of spoken texts were partly generated with digital aids. Members who do use AI tend to employ it to refine sentence structure or shorten arguments, asserting they only use it as a tool to enhance their delivery.
There is notable variance based on the type of parliamentary interaction; AI usage was less prevalent during major speeches by party leaders but more common in behind-the-scenes committee discussions. Some members express concerns about the total delegation of content creation to AI, while others accept the practice as an efficiency measure, with some parties actively considering guidelines for its use.

* 68 members of the Dutch House of Representatives created texts using artificial intelligence tools like ChatGPT and Claude in the past year.
* Researchers analyzed the spoken texts of 150 Members of Parliament using Pangram to scan for AI usage.
* Interruptions and motions were excluded from the count of analyzed texts.
* The list of members who used AI included former MP Lahlah van Pro, Van Meetelen (PVV), Belhirch (D66), and Van Eijk (VVD).
* Right-wing factions, specifically the PVV and Groep Markuszower, were overrepresented among those using chatbots.
* The SP, PvdD, and Volt reported no use of chatbots.
* Some members used AI for maidenspeeches, which are usually personal narratives.
* Most AI usage was observed among backbenchers rather than more prominent representatives.
* AI tools were sometimes used to shorten arguments while retaining content integrity.
* Approximately 30 percent of contributions from the three BBB members came from ChatGPT or Claude.
* Party leader Bontenbal stated the CDA faction is developing guidelines for AI use.

The finding that AI is being integrated into the foundational act of parliamentary discourse reveals a shift in the nature of political contribution, moving from direct, unmediated speech to mediated, optimized output. The pattern of adoption seems tied to practical constraints rather than ideological alignment, as evidenced by the high usage among more pragmatic or less ideologically rigid groups. This creates an emergent dynamic where efficiency—reducing time spent formulating arguments—becomes a primary driver for adopting these tools, which simultaneously introduces systemic risks regarding authenticity and accountability.
The tension arises between individual pragmatism and collective norms. When some members view AI as a mere tool for brevity, while others fear the erosion of authentic representation, the system faces an internal conflict about what constitutes legitimate political expression in a deliberative body. The focus on backbenchers and behind-the-scenes work suggests that AI adoption may first manifest where direct public performance is less scrutinized than in high-profile debates. Furthermore, the emerging attempt by some parties to establish formal usage guidelines signals an awareness that unmanaged technological integration requires explicit governance to maintain democratic integrity.
What are the implications for cognitive sovereignty when the tools used to construct public opinion are opaque? If efficiency leads to automated expression, the value shifts from the speaker's unique perspective to the algorithm's capacity for synthesis. This forces a re-evaluation of where political truth resides—in the raw input, the final text, or the underlying intent that prompts the prompt. How do established institutions design accountability mechanisms when the source of the articulation is computationally augmented? What happens when 'efficiency' becomes the default mode of political communication?