AI Helped Write One in Five Commons Speeches, Analysis Finds
One in five words spoken by MPs in the House of Commons in the past year was written with help from artificial intelligence, an analysis by The Economist says.
The finding, reported Sept. 25, 2026, covers spoken contributions from September 2025 to September 2026 in selected English-speaking legislatures. It places Canada near the top of the group studied. The Globe and Mail
The Economist used Pangram, a software tool that estimates whether text was likely produced by AI, to review the material. Canada ranked second at 20 per cent, just behind Australia at 21 per cent. The rate in Ottawa was higher than in England, Scotland, Wales, Ireland, the United States and New Zealand.
In Ottawa, the figure covered speeches, questions and replies, and other interventions in debate. Parliamentary rules do not ban the practice. The Speaker of the House of Commons declined to comment.
The broader context here is how House speeches are normally made. Few MPs write every word themselves. Staff draft statements, party researchers write lines for Question Period, ministers read material prepared by departments, and backbench MPs rely on assistants and outside groups. AI fits into that chain. It does not change the point when a member rises, takes the floor, and takes responsibility for words that go into the official record.
In my view, the procedural point is narrower than the technology debate implies. The House controls order, time limits and language. It does not generally control authorship. A member may read a speech written by someone else if it follows the Standing Orders — the written rules of the House — and the Speaker's rulings on unparliamentary language and proper procedure. MPs are not required to disclose drafting help. The Speaker's silence fits that approach. It leaves any decision on guidelines to the House itself, through party leaders and the procedure committee, not to a ruling from the Chair.
Looking at what this means for Hill offices, the effects come down to workflow and accountability. AI can speed up preparation for debates, members' statements and late-evening adjournment debates. It can also flatten style, so offices using the same tools can start to sound alike. That counts in a Parliament where regional voice and personal credibility matter. A well-written question can still fall flat if the member cannot defend it in follow-up, and a vague ministerial reply can invite a tougher second question.
Looking at what this means for trust in the record, caution is needed on both sides. Detection results are not audits of individual MPs. They show patterns across a large volume of text. They do not separate a speech written entirely by a model from a staff draft cleaned up for grammar and structure. That difference matters to working MPs. Voters expect members to understand and stand behind what they say. Whether a paragraph started with a staffer, a stakeholder brief or a prompt will matter less than whether the facts hold and the member can answer for them.
