AI search atomizes our information, warns govt designer • The Register

Those who rely on artificial intelligence to summarize official material may get a misleadingly narrow or incomplete version of it, a senior designer for the UK government has warned.

The country's Department for Education's digital services are seeing more traffic from AI-mediated search and fewer actual page visits, according to head of design Mark Edwards. "At first, this felt like progress; faster access to information is not something to resist," he writes in a blogpost on the GOV.UK website. "But as we looked closer, a more complicated picture emerged."

If AI-mediated agents or answers become the dominant entry point, we need to be sure that people who lack confidence or familiarity are not disadvantaged further... One problem is that AI tools only answer the question they are asked: "They meet users where they already are, which can limit discovery and reinforce gaps in understanding," Edwards writes.

For example, a teenager leaving school may not know to search specifically for information on options such as apprenticeships, T-levels or vocational pathways. Things may not be as bad for a clueless school-leaver as Edwards fears. An AI overview generated by Google in response to a search on the question "What are my options leaving school?" from a UK IP address included all three of the options he mentioned as well as A-levels, university and getting a job. It drew on sources including a GOV.UK blogpost designed to answer this question, as well as graduate careers sites and advice pages on websites run by charities, a local authority and an individual school, all of which appeared fairly reliable.

But even if AI tools can provide detailed answers to broad questions, those writing official websites will need to do so in future partly for people who never visit.

"We now need to design with the expectation that much of what we publish will be read indirectly, atomised, summarised or reinterpreted by systems we don't control," writes Edwards. This makes qualities including clear, well-structured content in plain English even more important, but also means thinking about whether a paragraph lifted out of context is safe and accurate.

Edwards adds that is also likely to mean testing what material looks like after it has gone through AI tools and designing it to be interpreted by both machines and people. "If AI-mediated agents or answers become the dominant entry point, we need to be sure that people who lack confidence or familiarity are not disadvantaged further," he says.

The Government Digital Service is piloting its own AI-driven chatbot, which only uses material from GOV.UK, and reckons this is now accurate in 90 percent of its answers. Users find its 10.7 second average response time is too slow, however, so the service is considering breaking up answers to provide an initial reply more quickly.

This shift toward AI-mediated information access represents a fundamental change in how government services reach citizens. Rather than users navigating to specific pages and reading complete documents, AI systems are increasingly serving as intermediaries that extract, summarize, and present information in atomized form.

The implications extend beyond simple convenience. When information is fragmented and presented out of context, there's potential for misinterpretation or loss of nuance. A policy document's full meaning might depend on understanding its broader context, which could be lost when AI tools extract only the most directly relevant paragraphs.

Edwards' concerns highlight a growing challenge for digital service providers: designing content that remains accurate and useful whether read directly by humans or processed by AI systems. This requires a new approach to information architecture that considers both human readers and machine interpretation.

The speed at which this transition is occurring is notable. The Department for Education's experience shows a clear trend toward AI-mediated access, suggesting that what Edwards describes as a "complicated picture" may become the standard way many people interact with government information.

For government services specifically, this shift raises questions about accessibility and equity. While AI tools might make information more accessible to some users, they could potentially disadvantage others who might benefit from the fuller context provided by traditional web pages. The challenge is ensuring that the move toward AI-mediated access doesn't create new barriers for those who need government services most.

As AI continues to reshape how we access information, the experience of the UK's Government Digital Service offers valuable insights into the challenges and opportunities this transition presents. The key will be finding ways to harness the benefits of AI-mediated access while preserving the accuracy, context, and accessibility that government services must provide.