There are two ways to discover you said something wrong on your podcast. One is during the recording, when fixing it costs you eight seconds. The other is three days after publication, when a listener emails you, and fixing it costs you a correction, a re-upload, or a public note that stays attached to the episode forever.
The Fact Checker agent is built for the first case. This guide covers what it actually checks, how to read what it gives you, and how to handle a correction on air without wrecking the flow of the conversation.
What the Fact Checker Watches For
The agent is narrower than "checks everything you say," and the narrowness is deliberate. Its role in the interface: Checks specific claims, stats, dates, and attributions.
Four categories:
- Claims. Assertions of fact that could be true or false.
- Statistics. Numbers, percentages, magnitudes.
- Dates. When something happened, in what order.
- Attributions. Who said it, who did it, who was responsible.
What it does not check: opinions, predictions, jokes, hypotheticals, and anything that is not a factual assertion. "I think the second album is better" is not checkable. "The second album went platinum in 1994" is.
This distinction matters for how you use it. On a show that is mostly opinion, the agent will be quiet. On a show that is dense with figures and history, it will be busy.
Turning It On
Open the Agents menu in the recording control bar and toggle Fact Checker on. You can do this before you start or partway through a take.
The mid-session toggle is more useful than it sounds. Many shows have a factual segment and a discussion segment. Run it during the first, switch it off during the second, and your research panel stays readable.
Reading a Verdict Card
When the agent catches something, a card appears in the panel beside your transcript. It has four parts:
Fact Checker Likely incorrect "LeBron has more NBA championships than Jordan." Jordan has 6 titles; LeBron has 4. Basketball-Reference, NBA Finals history
- A verdict. The agent's assessment of the claim.
- The claim, quoted back so you know exactly what triggered it.
- The correction or context. What the record actually says.
- A source. Where that comes from.
The source line is the part to look at first. A verdict without a source you recognize is a verdict to treat cautiously. A verdict citing a reference you know to be authoritative on that subject is one you can act on immediately.
Cards are silent and visual. Nothing plays, nothing pings, and your guest sees nothing. This is entirely on your screen.
How to Correct on Air
This is the skill part, and it is worth thinking about before you need it.
If you made the error, just fix it. "Actually, let me correct myself, Jordan has six and LeBron has four." One sentence, no ceremony, move on. Hosts consistently overestimate how disruptive this is. Listeners hear a person who cares about being right.
If your guest made the error, be gentler and slower. You have options in descending order of confrontation:
- Fold it into a question: "Is that four or six? I want to make sure we have it right."
- Attribute it outward: "I think the number I've seen is four."
- Let it go and note it. Not every small error is worth interrupting a good answer for. Press M to drop a marker at that timestamp and address it in the edit or the show notes.
The last option is legitimate and underused. A guest telling a meaningful story who gets a year slightly wrong does not need to be stopped. A guest stating a wrong figure that the whole argument rests on does.
If the verdict contradicts something you are confident about, trust yourself and check later. The agent is a prompt, not an authority. Drop a marker and verify after the recording.
Why Live Correction Beats Post-Hoc Correction
There is real research here, and it is less obvious than you would expect.
Audience trust in media is under strain. YouGov's 2026 Trust in Media survey found trust declined for most outlets measured. And corrections themselves are not automatically trust-building: research finds that people generally trust journalists when they confirm claims to be true, but are more distrusting when journalists correct false claims.
That finding is often misread as an argument against correcting. It is not. It is an argument about framing and timing. Work on correcting misleading content finds that how a correction is presented, including adding context and signaling uncertainty, measurably affects perceived credibility.
The practical version for podcasters: a correction that happens inside the conversation, in the moment, as part of getting it right together, does not read as a correction at all. It reads as rigor. A correction issued afterward, detached from the episode, is the kind that costs you.
Live checking means you mostly never have to make the second kind.
Which Shows Benefit Most
News and current affairs. Obvious. Dates, figures, and attributions are the entire substance.
Sports. Records, statistics, and history get stated confidently and wrongly all the time.
Finance and business. Numbers about companies, markets, and valuations, often stated from memory by guests with a stake in them.
History. Dates and sequence are the whole discipline.
Health and science. Where being wrong has consequences beyond your credibility.
Interviews with founders and executives. Guests routinely state figures about their own companies that are outdated, rounded generously, or from a deck rather than a filing.
Where to leave it off: comedy, fiction, improv, and any format where statements are not meant literally.
Using It With Session Context
Fact Checker and Prep Desk work well together and do different jobs.
Prep Desk draws on material you uploaded to the session through the Context panel: notes, files, URLs, and guest bios. Fact Checker checks against sources it finds.
On a show covering a specific report or study, load the primary document into Context. Then when someone characterizes what the study found, you have Prep Desk returning what the document actually says alongside Fact Checker's independent read. That combination is considerably stronger than either alone.
What the Cards Are Worth After the Recording
Every card surfaced during the take is saved with the session and appears in Session Viewer, anchored to the transcript line that triggered it.
This has two practical uses:
Editing. Scan the verdict cards before you cut. Anything flagged that you did not address on air is a decision: correct it in the edit, note it in the show notes, or cut it.
Show notes. The sources behind your verdicts are citations you can put on the episode page. A show that publishes its sources is a show that builds a different kind of relationship with its audience.
A Realistic Expectation
The Fact Checker will not catch everything, and it will occasionally flag something that is fine. It works from the finalized live transcript, so a claim buried in crosstalk or a badly mangled name may not register.
Treat it as a second pair of ears that never gets tired and has no ego about being the one who noticed. That is genuinely valuable. It is not a guarantee of accuracy, and no tool is.
The host is still responsible for what goes out. The agent just makes it much harder to be wrong by accident.
Building the Habit
The agent works better when you build two small habits around it.
Glance at the panel during your guest's answers, not during your own questions. While you are talking you should be talking. While they are answering you have spare attention, and that is when the card about your last claim has arrived.
Decide your correction threshold before the recording. Different shows warrant different standards. A history show should probably stop for a wrong decade. A casual business chat probably should not stop for a rounded revenue figure. Knowing your own threshold in advance means you are not making an editorial judgment under time pressure, mid-conversation, with a guest waiting.
A useful default for interviews: correct anything the argument depends on, mark anything that does not, and never correct a guest twice in one segment. The second correction changes the temperature of the room in a way the first does not.
Say what you are doing, once. Some hosts tell guests at the top that they run live fact-checking. It sets a tone of shared rigor rather than gotcha, and guests tend to be more careful with figures when they know. It also means a mid-interview correction is expected rather than jarring.
Turn It On for Your Next Episode
Open the Agents menu, toggle Fact Checker, and record something with numbers in it. The first time a verdict card catches a figure you were about to let stand, the value is self-evident.
Start at app.podmod.ai.