Initiative overview
The UK government runs approximately 500 public consultations annually to gather citizen input on policy decisions. Analysing tens of thousands of written responses manually is slow, expensive, and creates a bottleneck in democratic policymaking. The process requires an estimated 75,000 working days per year, costing £20 million, and often takes months, delaying policy action and making frequent public engagement prohibitively expensive.
Consult uses AI to automatically identify themes across thousands of consultation responses and categorise individual submissions, whilst keeping full human oversight at the core of the process. The tool employs a two-step approach: first generating themes from the responses, then mapping each response to relevant themes. Human analysts validate the themes and review AI categorisations, maintaining control over the analytical framework whilst eliminating time-consuming manual sorting.
Results, outcomes and impacts
Across seven consultations in 2025, Consult saved approximately 8,000 hours and over £250,000, whilst enabling faster policy responses to public input. For example, Consult analysed over 50,000 responses to the Independent Water Commission review in two hours (work that would have taken months manually). Evaluations have also found no evidence of additional bias.
Analysis time reduced from months to hours (50,000 responses in two hours). Cost reduced from £100,000 per consultation to £240. Potential savings: £20 million annually across all UK government consultations. Performed better than the human-to-human baseline (83% vs 55% agreement). Sixty per cent of AI suggestions were accepted without changes by expert reviewers.
Governments can run more consultations, respond faster to public input, and allocate analyst time to interpreting findings rather than manual categorisation, leading to more effective policymaking.
Challenges and lessons learned
Measure AI against realistic human baselines, not perfection. Human experts only agreed 55% of the time; AI achieving 83% exceeds typical human performance. Build public trust proactively through transparency. Consult was the first UK tool to complete an independent "Social Readiness Assessment" with public input on AI use in democratic processes. Design for human oversight at critical stages. AI categorises responses but humans validate themes and review mappings; never fully automate analysis of democratic input.



























