ISED's Four-Model Classification Pipeline for Canada's Largest-Ever AI Strategy Consultation

Between 1 and 31 October 2025, Innovation, Science and Economic Development Canada (ISED) ran a public consultation on the country's next national AI strategy. It drew submissions from more than 11,300 participants answering 26 questions, generating over 64,600 individual responses — the largest public consultation in ISED's history.

To process that volume, ISED combined the commercial platform SimpleSurvey's natural-language-processing tools with an internally built classification pipeline that ran every submission through four different large language models — Cohere's Command A, OpenAI's GPT-5 nano, Anthropic's Claude Haiku and Google's Gemini Flash — to identify recurring themes across open-ended answers.

Independent legal scholars who reviewed the process, including University of Ottawa law professor Teresa Scassa, noted that ISED did not disclose the prompts, data flows, classification criteria or agreement thresholds used to reconcile the four models' outputs, raising questions about how reproducible or auditable the theme extraction was. ISED extended its own reporting timeline into 2026 given the volume of submissions, and the resulting 'AI for All' strategy was published in June 2026 drawing on the consultation's findings.

Read the full analysis: https://ised-isde.canada.ca/site/ised/en/public-consultations/engagements-canadas-next-ai-strategy-summary-inputs