TL;DR
A team from Edinburgh Napier and Sweden’s Chalmers University of Technology found AI-assisted content in 243 of the 4,200 Westminster speeches they examined, about 5%, and the share has been creeping up over time, LBC reported on 15 October. The team looked for turns of phrase and pairings of words that large language models favour over human writers. Party made no meaningful difference, and the Swedish chamber looked much like Westminster.
Using AI to draft a speech is legal for MPs, and nothing in Commons rules forbids it. The researchers’ concern is accuracy and disclosure.
What the researchers found
Dr Peter Barclay, a computer scientist at Edinburgh Napier, warned that AI can shift the intended tone of a document “or even potentially introduce outright errors”. He contrasted routine use by politicians with polling in which only 4% of the public said they would trust AI to make decisions in government.
The study does not name individual MPs. Only one has said publicly that his office uses AI: Labour’s Steve Yemm told the Economist “my office uses general-purpose AI productivity tools within parliamentary guidelines”. The research found no AI-attributed text in speeches by a group of senior figures that included Kemi Badenoch, Sir Ed Davey and Nigel Farage.
In Sweden, close to one in ten motions in the 2025-2026 session were produced with AI help, the study found.
The case for disclosure
Andrea McGlinchey, a fellow Napier researcher, said the work looked only at evidence of undisclosed use in the language of parliamentary texts, not at regulation, but that the results point to a need for “greater transparency”. She is calling on ministers to say openly where AI and algorithms are used in official duties.
Looking forward
In our view, detection that relies on language patterns can only flag likely AI phrasing, so any individual case stays contestable. That makes a disclosure norm more practical than detection. Parliament already lets members use the tools; a simple statement when a speech or motion was drafted with AI help would address the accuracy worry Barclay raises without banning anything. The same question is arriving in every UK organisation that publishes in its leaders’ names.