Most advice about fact-checking podcasts was written for narrative shows. Script, research, verify, record. That workflow is fine if you have a script.

Interview shows don't. Your guest says a number out loud, it's in the recording, and you have two options: catch it in the moment or catch it in the edit. Neither one is what the checklists describe.

This is a working process for hosts who record conversations. It covers what's actually worth verifying, how to handle a wrong claim while the tape is rolling without turning the episode into a cross-examination, and what to do when something slips through and goes live anyway.

Not every claim needs checking

The standard advice is to go through your transcript and highlight everything that could be a fact. On a 70-minute interview that leaves you with maybe 200 highlighted phrases and no plan.

Triage first. Three buckets:

Check it, always. Numbers, dates, dollar amounts, study findings, "the first / the biggest / the only" claims, anything about a named third party, and anything a listener could act on — medical, legal, financial. These are what get quoted back at you, screenshotted, and clipped.

Check it if it's load-bearing. Background details that hold up an argument. If the guest's whole point rests on "the industry doubled between 2019 and 2022," that's now a bucket-one claim even though it sounded like context.

Leave it. Opinions, predictions, personal memory about their own life, obvious hyperbole. You are not going to verify that their first job was miserable. Marking a guest's lived experience as "unverified" is a category error, and chasing it burns the time you needed for the real checks.

That sorting takes about ten minutes on a transcript. It's the step that makes the rest of this feasible.

Before you record: check what's checkable

Half the fact-checking work can happen before anyone is on mic, because you already know roughly what the guest will say.

Read their recent posts and interviews and pull out their stock claims. Every guest has four or five. The stat they always cite, the founding story, the "studies show" line. Check those in advance. Ten minutes with the original source tells you whether the study says what they say it says, and it's remarkable how often the number is real but the framing has drifted over the years of retelling.

Then keep the sources somewhere you can reach mid-recording. A single open tab with the primary links, not a folder you'd have to hunt through. The point isn't ammunition. It's that when the claim comes up you can go one level deeper instead of nodding.

If you send a pre-interview questionnaire, add one line: if you plan to cite any statistics or studies, send the source. Most guests will. The ones who can't find it usually stop citing it, which solves the problem quietly.

During the recording: the part nobody writes about

Here's where interview shows diverge from every fact-checking guide out there. A wrong claim in a live conversation isn't a research problem. It's a social one.

You have roughly three seconds to decide, and four ways to play it.

Ask for the source, neutrally. "Where's that number from? I want to look it up after." Zero confrontation, and it gives you the thread to pull later. Make this your default; it covers most situations.

Soften it in real time. "That's the ballpark I've seen too, though I think the more recent figure came in lower." You've flagged uncertainty for the listener without stopping the episode. Nobody is embarrassed.

Correct it directly. Reserve this for claims that could actually harm someone. Health, money, legal advice, a false statement about a named person. Say it plainly: "I want to jump in, because I think that's not right, and it matters." A good guest will thank you. A guest who won't take a correction on a harmful claim has told you something useful about the rest of the interview.

Let it go and fix it in post. Legitimate. Note the timestamp, keep going, and handle it with a correction later. Just decide it consciously instead of freezing.

What kills interviews is the fifth option: half-challenging. You mumble something skeptical, the guest gets defensive, and you spend four minutes negotiating instead of talking. Commit to one of the four.

One practical detail: whatever you do, say the timestamp out loud or write it down. "Note that" costs a second and saves you scrubbing through 70 minutes later looking for the moment you know is in there somewhere.

After: the transcript pass

Once you have a transcript, this goes faster than people expect.

Run your triage on it. Bucket one first, working from the primary source rather than a secondary write-up — the number in the news article is often not the number in the paper. Log each check as verified, wrong, unverifiable, or needs-the-guest, and keep that log. It becomes your record if anyone ever disputes the episode.

For anything that lands on wrong, you have three choices, in order of preference:

  1. Cut it, if it's incidental and the edit doesn't leave a hole. Cleanest fix.
  2. Correct it in the show notes, with the source linked. Works when the claim is woven into the conversation.
  3. Record a short correction and drop it at the top of the next episode. Best for anything material.

Go back to the guest on anything you can't resolve. "You mentioned X — I found Y, do you have a different source?" Most people appreciate not being published with an error attached to their name. And when a guest genuinely can't back up a claim you can't verify either, cut it. Unverifiable plus contested is not a coin flip.

Then write the sources into your show notes as a matter of habit. It's a real credibility signal, it takes five minutes when you already have the log, and it makes the next fact-check easier because the last one is written down.

Where a live co-pilot changes the math

The hardest part of everything above is the middle section. Checking a claim during a conversation means splitting your attention exactly when you can least afford to, which is why most hosts don't do it and handle everything in post.

That's the gap Liyam AI is built for. Its live fact check listens as you record and catches inaccuracies as they happen, bringing sources on screen without breaking momentum — so the claim gets verified while the topic is still open, not three days later in a transcript. When you want to go further into it, you can call up relevant articles, images, and clips mid-recording, and push them to a second screen if your setup shows visuals to the guest or the audience.

The practical difference is when the correction lands. Fixing an error in the moment gives you a natural exchange and a better episode. Fixing it afterward gives you a line in the show notes that a fraction of listeners will read.

It helps on the bookends too. Liyam's prep agent can sit in on your prep call and draft the outline and questions, which is where the pre-check research naturally attaches. And after you stop recording, it finds the most watchable moments for clips — worth noting for the reverse reason, because a clipped claim travels much further than the episode. The clips are what to check hardest.

None of it removes your judgment. Deciding a claim matters, and deciding whether to challenge it on tape, is the host's job. The tooling just means "I'd have to look that up" stops being a reason not to.

FAQ

Do podcasts have to be fact-checked?

There's no legal requirement and no regulator for independent shows. But you carry the same practical exposure as any publisher: defamation risk on claims about named people, platform policy risk on health and financial misinformation, and the reputational cost of being the show that repeats things. Interview shows carry it too. "My guest said it" isn't the shield hosts assume it is.

How do you fact check a guest without being rude?

Ask for the source instead of challenging the claim. "Where's that from? I want to link it in the notes" reads as diligence, not doubt, and gets you the same information. Save the direct correction for claims that could actually hurt someone, and when you use it, say why it matters.

How long does fact-checking an episode take?

For a typical interview, budget 20 to 40 minutes after recording once you're triaging rather than checking everything. Front-loading the guest's known claims before you record cuts it further, because the checks you'd otherwise do at the end are already done.

What if a guest's claim turns out to be wrong after publishing?

Correct it visibly and quickly. Update the show notes with the correction and the source, and if the claim was material, say so at the top of the next episode. Silent edits are the worse option — someone always has the original, and the cover-up is the actual credibility problem.

Bottom line

Fact-checking a podcast isn't about verifying every sentence. It's about knowing which claims carry weight, catching them as close to when they're said as you can, and being straightforward when one gets through.

Triage instead of highlighting everything. Check the guest's stock claims before you record. Have a move ready for the live moment so you're not deciding under pressure. And write down what you checked, so the record exists.

If you want the verification to happen while you're still on the topic instead of three days later, see how Liyam AI rides shotgun during the recording.


Further reading: Podcast Allies on fact-checking your episodes, Vocal Fry Studios' walkthrough of the fact-check pass, and the Columbia Journalism Review's interview on fact-finding methods.