Tested whether AI assistants can actually read the IRS, Social Security, and Student Aid websites. Two of the three block them outright.

People increasingly ask ChatGPT/Perplexity things like "how do I check my SSA benefits" or "what's the student loan forgiveness status." I got curious whether the sites those answers would need to come from actually let an AI agent in the door.

Ran the same audit engine I've been using on nonprofits and news outlets against 7 government service URLs. Method: one server-side fetch per public page, checks for agent-crawler access, structured data, and machine-readable signals — no live LLM prompting, same weights every time.

IRS.gov: 50/100 — blocks llms.txt and agents.md outright, no structured data
Medicare.gov: 70/100 — has schema.org, missing agent-specific metadata
USA.gov: 70/100 — same gap
IRS.gov/payments: 70/100 — same gap
SSA.gov: timed out — security layer blocks the fetch entirely
SSA.gov/myaccount: timed out — same block
StudentAid.gov: timed out — same block

So the three services people probably ask AI assistants about most — Social Security, IRS, student aid — are either partially readable or don't respond to the check at all. Not a knock on their security teams (blocking bots is a reasonable default) — just an honest data point about the gap between "citizens are already asking AI for this" and "the sites are built to answer."

Full methodology + reproducible: https://alexander-k-eliot.github.io/ai-visibility-check-free/government-benchmark/

Genuinely curious if anyone here has dealt with this from the inside — is the blocking deliberate policy or just default WAF config nobody's revisited?