Citizens Advice has quietly built an AI chatbot that could transform how the charity's 21,000 advisers help millions of Britons each year – but the public will never see it. In a telling move that speaks volumes about the current state of AI, the organisation has deliberately kept its new tool behind closed doors, fearing the technology isn't reliable enough for the complex personal crises that walk through their doors daily.
The chatbot, revealed in a Computing UK report, serves as an internal research assistant for Citizens Advice staff, helping them rapidly sift through the charity's vast knowledge base of policies, precedents and guidance. Rather than replacing human advisers, it's designed to free them up for what they do best – listening to people's problems and offering nuanced support that considers their individual circumstances.
But here's where it gets interesting: despite the potential to serve clients directly and reduce waiting times, Citizens Advice's leadership chose to keep the AI firmly in the back office. The reason? Current AI systems have an alarming tendency to "hallucinate" – confidently presenting information that sounds credible but is completely wrong. When you're advising someone facing eviction, struggling with universal credit, or drowning in debt, there's no room for error.
This decision reflects a crucial tension playing out across Britain's public services and charities. AI promises remarkable efficiency gains and could theoretically help more people access support faster. Yet organisations serving vulnerable populations face a stark reality: get the advice wrong, and you could push someone deeper into crisis. The technology simply can't match human advisers' ability to read between the lines, ask the right follow-up questions, or recognise when someone needs emotional support alongside practical guidance.
Citizens Advice's approach reveals a mature understanding of AI's current limitations. Rather than rushing to deploy flashy public-facing technology, they're using it strategically – as a tool to make their human experts more effective, not to replace them. It's a blueprint that other organisations grappling with AI adoption would do well to study, particularly as the pressure mounts to integrate these systems into front-line services affecting real people's lives.