Government AI chatbot goes live across GOV.UK App - PublicTechnology

The UK Government Digital Service has completed a 2.5-year development cycle to integrate an AI-powered chatbot directly into the GOV.UK App, which now serves 563,000 registered users. The chatbot, branded as GOV.UK Chat, represents a significant infrastructure decision: after initially building on OpenAI's technology, the team migrated to Anthropic's Claude large language model, signaling a strategic shift in the government's approach to generative AI procurement and vendor selection.

The system's scope is substantial, with the chatbot trained on 80,000 pages of government guidance drawn from GOV.UK's broader ecosystem of 700,000 pages. This architectural choice—grounding responses in structured government documentation rather than relying on general-purpose training data—reflects a deliberate attempt to constrain hallucination risk and maintain authoritative, policy-aligned outputs. The Department for Science, Innovation and Technology has reported 90% accuracy in testing, though the methodology and baseline for these claims warrant scrutiny from an engineering perspective.

For senior technology leaders evaluating similar citizen-facing AI implementations, this deployment offers practical lessons in vendor flexibility and long-term maintenance planning. The decision to swap underlying models mid-project suggests the government prioritized model reliability and safety characteristics over switching costs—a calculus that differs markedly from many enterprise deployments. The integration into an existing mobile application with half a million users also underscores the operational complexity of rolling out generative AI at scale within regulated public-sector environments, where user expectations around accuracy and consistency are particularly high.