Canadians Split Over AI in Federal Consultation

The federal government held a 30-day public sprint in October 2025 to inform a refreshed national AI strategy, and published a summary of the consultation (Government of Canada). The government's summary reports it processed more than 64,600 responses submitted by over 11,300 participants (Government of Canada; BetaKit). Innovation, Science and Economic Development Canada, or ISED, reported using digital tools and an in-house pipeline of large language models, with human oversight, to identify and summarize common themes in the submissions, as described in BetaKit's reporting (BetaKit).

BetaKit ran an independent, keyword-based code analysis of the consultation submissions to quantify how often topics appeared in the dataset. That analysis found references to AI's economic benefits in 35.6% of entries and references to AI harms in 34.6% of entries. The four most prevalent themes BetaKit identified were economic growth, ethical harms, environmental harms, and productivity (BetaKit).

Privacy Commissioner Philippe Dufresne testified to the House of Commons that results from the consultation show deep public skepticism about the technology, especially generative AI, and pushed for a focus on trust and privacy protections. "The value of this innovation will be maximized when it is accompanied by trust," he said, as reported by Global News (Global News). The government launched an AI Strategy Task Force and the October 1 to October 31, 2025 national sprint to gather input from industry, academia, civil society, and Canadians, according to the government news release (Government of Canada).

BetaKit's method is a keyword-based coding approach applied to a large corpus. Industry-pattern observations: keyword-frequency methods scale to tens of thousands of submissions but typically miss semantic nuance, sarcasm, and context-dependent sentiment unless supplemented with human annotation or more sophisticated NLP pipelines. BetaKit also notes the government summary did not publish the prompts or the detailed prompt-engineering used by the internal LLM pipeline, which limits reproducibility of the government's synthesis (BetaKit).

Observed patterns in similar public consultations show a recurrent trade-off between capturing broad participation and preserving granular transparency about analytic methods. When governments or large organizations use automated tools to summarize public comments, independent verification becomes harder without published prompts, model details, or example annotations. For policy and governance practitioners, that gap can undermine public trust in the interpretive layer between raw submissions and distilled recommendations.

• •Whether the government or ISED publishes a methodological appendix, sample prompts, or aggregated metadata that allows researchers to reproduce or audit the thematic breakdown (Government of Canada; BetaKit).
• •Parliamentary committee testimony and follow-up from the Privacy Commissioner for specific recommendations on privacy safeguards and data-use limits for training models (Global News; parliamentary records).
• •How the AI Strategy Task Force cites consultation themes when it issues its own recommendations, and whether task-force outputs include technical annexes describing data handling and synthesis methods (Government of Canada).

For practitioners handling large-scale public feedback, this episode highlights two operational priorities in public-facing analytics: documentable preprocessing and explainable synthesis. Industry-pattern observations: teams that publish codebooks, sampling strategies, and prompt examples enable more productive external critique and increase confidence among stakeholders. BetaKit's numeric breakdown supplies a useful baseline for comparative analysis, but it does not replace a reproducible pipeline for auditability (BetaKit).

What remains unsourced in public reporting is the exact prompt set and model configuration ISED used when the in-house LLM pipeline assisted synthesis. BetaKit requested further detail from Minister Evan Solomon's office, according to BetaKit's article; the public record in the scraped reporting does not include a complete methodological disclosure (BetaKit).