Study: AI Can Explain Your Government Benefits but Can't Apply For You
You'll still have to click through the paperwork yourself — in all 166 countries.
Governments are increasingly meeting citizens through a chatbot instead of a website. A new study introduces RADAR, a measuring stick that asks two simple questions about that AI in 166 countries: can it correctly explain a public service, and can it actually get you there? The researchers call the first "informational legibility" — can the AI tell you, accurately and from an official source, how to renew a license in your country? The second they call "agent operability" — can an AI agent (software that takes actions for you, like filling in forms or booking appointments) actually complete the process on your behalf?
The headline finding is blunt: AI is much better at describing government than at reaching it. In every one of the 166 countries studied, the explaining score beat the doing score. And here's the surprising part — the gap did not shrink as countries got richer. Money alone doesn't fix it. National income and language explain only part of the pattern, and several governments performed far better or worse than their budgets would predict.
The two problems also have different causes and different cures. Whether an AI can describe your government well depends on how much your country's main administrative language shows up in the giant piles of text used to train AI models — something a government can't change quickly. Whether an AI can actually reach the service depends on how strong that country's web presence is, which a government can change right now. Existing rankings of "digital government" miss this second problem entirely, which means many countries look better on paper than they are in practice.
For you, the practical takeaway is this: if you ask an AI assistant how to claim a benefit, renew a passport or register a business, you'll likely get a decent answer. But expect to do the last mile yourself — the clicking, the uploading, the waiting. RADAR's authors frame that as good news: because reach is fixable, governments at any income level can act now, so public services become not just described by AI but genuinely doable through it.
- Researchers scored AI against government services in 166 countries — and describing a service beat actually completing it in every single one.
- Wealth didn't close the gap: rich and poor countries alike struggle to let AI agents finish the job, from filing forms to booking appointments.
- The fix differs by problem — describing depends on language data a government can't change fast, but reach depends on web presence a government can improve now.
Why It Matters
AI may soon answer your government questions, but you'll still do the paperwork yourself.