GOV.UK Chat launches but charities warn it may exclude most vulnerable users
TL;DR:
- Technology secretary Liz Kendall has launched GOV.UK Chat, an AI assistant on the GOV.UK app intended to answer questions about benefits, schemes and entitlements without users “trawling hundreds of web pages”.
- Frontline charities including Nova New Opportunities and Policy in Practice say the chatbot’s design — English-only, iOS 16 / Android 10 minimum, GOV.UK-app-only — risks closing it off to the people in greatest need: those with limited digital literacy, second-language English speakers and households experiencing digital poverty.
- Policy in Practice estimates around £24 billion of UK benefits and financial support goes unclaimed each year because of stigma, system complexity and unawareness; the new chat tool is being pitched as a partial answer, but Amnesty International UK warns it must not be presented as a substitute for human-mediated support.
GOV.UK Chat is the highest-profile UK central-government generative-AI deployment yet aimed at the general public, and the launch coverage is unusually divided. Cabinet Office briefing has framed it as a productivity win for the citizen-facing side of digital government; Big Issue, Amnesty and frontline charities have framed it as a test of whether AI rollouts in welfare services are designed for the median user or the marginal one. Both framings are true, and the gap between them is the policy problem.
The accessibility critique is structural, not stylistic
Lizzie Cho of Nova New Opportunities is clear about who the tool serves well: “those who have higher levels of literacy, who know how to navigate and trust the systems around them.” Her concern — that “those most marginalised, most in poverty, with the most difficulties, who need the system for urgent needs might be excluded” — points at three concrete design choices. English-only output excludes second-language speakers who often need the most support claiming what they are entitled to. The iOS 16 / Android 10 minimum cuts out anyone whose only phone is older. And routing the experience exclusively through the GOV.UK app puts an extra trust hurdle in front of the Grenfell community and other groups whose distrust of councils and central government is well-documented.
Javier Ruiz Diaz, technology and human rights lead at Amnesty International UK, raised a sharper concern: the chatbot must not “make decisions, or give any information that could dissuade people from requesting social security”. DSIT has set strict privacy guardrails — the system blocks email addresses and phone numbers and does not require personal information to use — but a privacy guardrail is not an accessibility guardrail. The Ada Lovelace Institute’s briefing this week on UK AI productivity claims raised an adjacent concern: public-sector AI metrics overwhelmingly measure time and cost saved, rarely service quality or equity. GOV.UK Chat is the test case for whether DWP, HMRC and Cabinet Office can demonstrate the latter.
Looking forward
Expect select-committee scrutiny on whether GOV.UK Chat’s evaluation framework includes accessibility outcomes, not just usage and deflection metrics. For UK SMEs and govtech vendors selling AI into central government, the Cho-Barnett framing — “you always need a middle partner to transfer the trust” — is a procurement signal: contracts that bundle community-mediation and human-fallback provision will look more credible than pure-AI deployments. The reach of £24bn in unclaimed support is the prize on offer; the warning is that a chatbot which is great for digitally-fluent users may not move that number.