America.gov changed what it would answer during its own launch. Decide what your AI won't answer before yours.

On launch day, the Associated Press and Fortune tested America.gov with questions about elections and the Constitution. Early in the day, the chatbot answered them, drawing on official sources. Later, while the launch event was still going, it began declining what it called "political questions." No official explanation has been reported.

This post is not about which answers were right. People will disagree about that, and that is the point. It is about what happens when the rules for an AI are decided live, in public, after it is already talking to people.

Every AI has an answer policy. Most are unwritten.

Any AI you put in front of customers or staff has rules about what it will and will not say. The question is whether you wrote them, or whether they are whatever the model and the vendor defaulted to.

America.gov does publish some rules. It will not give financial, tax or legal advice, and answers carry a note saying so. That is a written policy, and it is a good one. The political-questions rule looked different from the outside: it appeared during the day, and the public learned about it by noticing the answers change.

When the rules change without notice, people stop trusting the answers they got before the change, and the ones they get after it.

The business version of this story

Your customer-facing AI will be asked things you did not plan for.

A customer asks your support bot whether you are going out of business.
A prospect asks your website assistant how your price compares to a competitor's.
An employee asks the internal assistant about a layoff rumor.
Someone asks about a lawsuit, a recall or a safety issue.

If nobody decided in advance, the AI will improvise. And when someone at your company sees the answer and panics, the rule will get changed on the spot. That is the America.gov pattern, just smaller.

Write the policy before launch day

A one-page answer policy covers most of it.

What the AI answers fully. Products, prices, hours, process, public facts.
What it answers with a source and a disclaimer. Anything close to legal, medical, financial or safety advice.
What it declines, and what it says instead. Named topics, with a fixed reply that points to a person.
Who owns the policy. One name. Changes go through them.
A change log. When the policy changes, write down what changed, when, and why. If it is customer facing, say so publicly.

Why ownership matters here most

An answer policy is a commitment your company makes to everyone who talks to your AI. Like any commitment, it needs an owner and a place where it gets reviewed.

That is the part AI does not solve. Somebody has to decide, somebody has to write it down, and somebody has to check next week whether the AI is still keeping the promise.

What to do this week

List five questions you would not want your AI to improvise on.
Write the reply you want for each one. Two sentences, pointing to a person.
Put one owner's name on the policy and add it to your next weekly meeting as a standing check.

Frequently asked questions

Did America.gov stop answering political questions?

According to the Associated Press and Fortune, America.gov answered several politically sensitive questions early on launch day and later began declining what it called political questions. No official explanation had been reported as of September 30, 2026.

What is an AI answer policy?

An AI answer policy is a short written document that says what a company's AI assistant will answer fully, what it answers with a disclaimer, what it declines, who owns the policy and how changes are recorded.

Sources

Ask your AI who owns what

OTP is readable from any MCP client. In Claude Desktop, Cursor or any MCP client, add this block:

"otp": {
  "command": "npx",
  "args": ["-y", "@orgtp/mcp-server"]
}

Restart the client. Then ask: "Use OTP to show me Sneeze It's chart. Who owns each seat, and what number is each one measured on?"

Start free at orgtp.com. Every seat is free, for people and agents. You pay only for the AI you use.

The series

All ten on one page: What America.gov teaches business about its knowledge problem.

Series: What America.gov teaches business about its knowledge problem. Part 8 of 10. Read this post as markdown: /blog/decide-what-your-ai-wont-answer.md.