Use an AI agent to check facts during a live meeting

The job: flag, cite, don't rule

A decision can rest on a remembered number: "churn was about 4% last quarter" or "the contract renews in March." A fact-checking agent's job is to notice those claims and show what a source says.

It should never declare who's right. It points to evidence, and a person decides whether the meeting needs to correct course.

Why the agent needs checking too

Generative AI can state false things with confidence. NIST calls this confabulation and warns that it can produce made-up citations that make a wrong answer look justified.

So design the agent to show its work. Every flag should name its source, and the person reading it should open that source before relying on it.

Instructions you can adapt

Keep the scope tight. A starting point:

• Flag only specific, checkable claims: numbers, dates, names, quoted policies and commitments from earlier meetings.
• Check each against our saved FAQs and earlier meeting notes first. Use web search only for public facts.
• For each flag, give the claim, what the source says and a link or meeting reference.
• If you can't find a source, say "unverified". Don't guess.
• Ignore opinions, forecasts and anything said as a hypothetical.

Set it up in Bridge

Choose Create Agent from the sidebar. You can start from the Fact Checker template if it appears among the six starter templates Bridge shows, then rewrite the instructions for your team.

Test it on a meeting you know

Before using it live, open the Assistant on a completed meeting you remember well and @mention the agent. Ask it to check that meeting's claims. Compare its flags with what you know is true.

Look for two failures: claims it should have flagged and missed, and flags with weak or invented sources. Tighten the instructions until both are rare.

Keep people in charge

Proactive agents report privately within the Assistant, so the person who sees a flag chooses whether to raise it. Keep it that way while you build trust.

Agents can join and speak only in meetings hosted on Bridge, and participants see an AI badge on them. On Zoom, Meet or Teams calls you capture, flags stay private in the Assistant.

Tell people what the agent does and when it will speak up. Agents need a paid plan, and live features use Live AI credits.