Government Agencies Need to Publish for AI, Not Just Humans
When 20 million Americans lose their jobs each year, many turn to the web for answers: unemployment benefits, food assistance, health insurance, taxes. They no longer search exclusively for official government websites. Instead, millions now ask generative AI tools like ChatGPT, Gemini, and Claude - and treat the confident paragraphs these systems return as reliable guidance.
The problem: when official information is outdated, fragmented, or locked inside PDFs and interactive calculators, AI systems retrieve whatever is easiest to find. A plausible-sounding answer from a secondary source can nudge someone toward a missed deadline or rejected application.
How Government Websites Became Invisible to Machines
Over two decades, government agencies designed websites for humans clicking through browsers. Eligibility rules live in interactive calculators. Policy guidance hides in PDFs. Requirements span multiple screens.
That design works for people. For generative AI systems, it's functionally invisible. These tools can only summarize information they can access and interpret. They favor sources that are easy to extract.
In an AI-first internet, public-facing policy becomes whatever an AI system can retrieve most easily.
What Governments Should Do Now
Agencies need to treat machine-readable publishing as civic infrastructure. Several concrete steps help:
- Publish program rules, eligibility guidance, required documents, and deadlines on stable web pages - alongside PDFs and interactive tools, not buried inside them
- Add visible "last updated" lines and notes on "what's changed" when policies shift
- Coordinate between program teams who control rules and digital teams who control publishing
- Treat official guidance as mission-critical
These tasks require coordination. But the alternative costs more: missed deadlines, incorrect paperwork, and friction for people who can least afford it. Those small errors create real delays. Understaffed help lines field more calls. Caseworkers spend more time on back-and-forth. Costs land on agencies, not just applicants.
Early Action in California
Marin County, California, has begun assessing how often generative AI tools cite official county sources versus secondary sites. The work could inform replicable guidance for governments at every level.
The Reality Governments Face
Agencies cannot choose whether AI mediates public services. That's already happening. But they can improve the odds that the tools people use surface - and cite - official, accurate guidance.
For government professionals, this means treating information architecture as a service delivery problem, not just a web design problem.
Your membership also unlocks:
