Every answer America.gov gives links to its source. Your company's answers should too.

The best decision in America.gov is not the AI model. It is a rule.

Every answer links back to the official agency page it came from. Axios, Engadget and FedScoop all describe the same pattern: the chat gives you a plain-language answer, and underneath it, the source.

That one rule does three things at once. It lets you check the answer. It tells you who is responsible for the underlying fact. And it keeps the AI honest, because an answer that has to point somewhere cannot drift too far from what that somewhere says.

An answer without a source is an opinion with good grammar

AI writes fluently whether it is right or wrong. That is the whole problem. A made-up number and a real one look identical in a chat window.

Sourcing is the only defense that scales. Not "trust the AI." Not "have someone double check everything." Just a rule that every factual answer carries a pointer to where it came from.

We learned this from money

At Sneeze It, one of our agents does forensic finance. Its first rule is written in plain words: never state a financial fact without naming its source. Its second is that an estimate never becomes a fact by being repeated.

We wrote those rules because we watched the opposite happen. A number gets said in a meeting. Someone writes it down. Three weeks later it is in a slide, and nobody remembers it started as a guess.

AI speeds that cycle up. An assistant that reads last month's notes will repeat the guess back to you as a fact, and it will sound more certain than the person who first said it.

Where to apply the rule

Your scorecard. Every number should say where it came from and who entered it. A KPI that an agent pushes from the source system is better than one somebody typed in from memory.
Your AI tools. When you set up an assistant for your team, choose the ones that cite. Turn off answers that do not.
Your meeting notes. A decision should point to the meeting where it was made. A to-do should point to who agreed to it.
Your website. Structured facts about your business, published once and read by every assistant, beat a dozen pages that each say something slightly different. More on that in part six.

Where OTP fits

In OTP, a number on the scorecard carries its owner, and an agent can push its own KPI straight from the system it measures. To-dos come out of the meeting where they were proposed and confirmed, so every commitment points back to the moment it was made.

What to do this week

Take your top five scorecard numbers. Write where each one comes from next to it. If the answer is "someone types it in," note who.
Ask your AI assistant a question about your business. If it does not tell you where the answer came from, ask it to. See what happens.
Adopt the rule out loud. In your next meeting, any number without a source is a question, not a fact.

Frequently asked questions

Does America.gov cite its sources?

Yes. Coverage of the launch describes every answer as drawn from official government websites, with links to the original agency pages.

How do you stop AI from making up numbers about your business?

Require a source on every factual answer, feed the AI from systems of record rather than notes and chat, and treat any number without a source as a question to check, not a fact to repeat.

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 4 of 10. Read this post as markdown: /blog/america-gov-every-answer-needs-a-source.md.