You're eleven minutes in and you already know. The answers are landing short, the energy is flat, and some part of your brain has quietly started drafting the apology email while another part keeps nodding.

Most advice about saving a bad podcast interview shows up too late to help. It's written for the edit: cut the dead sections, build a clip show, or accept the loss and move on. Good advice, wrong moment. The interview is still happening. You have forty minutes of recording time left and a guest who is right there, willing, and fixable.

This is for hosts who record interviews and want a plan for the middle of one going wrong. Not editing tricks. What you actually say next.

First, stop calling it "bad"

"Bad interview" is a useless label because it points at a feeling instead of a cause. You can't fix a feeling. You can fix a mechanism.

Almost every interview that's going badly is going badly in one of four specific ways, and the fixes are close to opposites. Deploy the wrong one and you make it worse. Ask a rambler an open-ended question and you lose another six minutes. Interrupt a nervous guest to tighten things up and they close down completely.

So before you change anything, spend sixty seconds working out which one you're in.

Mode 1: The closed-off guest

Answers run under fifteen seconds. You get facts, not stories. There's a small silence after each one where you can hear them waiting for the next question like it's a job interview.

Usually this is nerves, not reluctance. They've been told a podcast is a Big Deal, they're performing "professional," and performing makes people terse.

Mode 2: The rambler

The opposite problem and much more common with experienced guests. One question, nine minutes, four topics, no landing. You're not learning anything because there's no shape to hold on to.

Mode 3: Pitch mode

Everything routes back to the product, the book, the funding round. The guest is media-trained and running a script. The tell is that answers sound finished — polished in a way that spontaneous speech never is.

Mode 4: The mismatch

The rarest and the only genuinely bad one. You booked them for a topic they turn out not to know much about, or the premise of the episode is wrong. No conversational technique fixes this, and pretending otherwise wastes an hour.

The moves, one per mode

Here's what to actually say. These are close to verbatim, because "gently steer them back" is not usable at speed.

For the closed-off guest, drop the question format entirely. Questions invite answers; answers can be short. Instead, hand them a scene to walk into: "Take me to the day you decided. Where were you sitting?" Concrete, sensory, and almost impossible to answer in eight words.

Then shut up.

The single highest-return move here costs nothing and most hosts can't do it. After they finish, wait two full seconds before you speak. NPR's Rachel Martin puts it plainly: sometimes the best follow-up question is to say nothing, because people fill the space themselves. Two seconds feels like a decade in your headphones. It reads as completely normal on tape. If you only take one habit from this piece, take that one.

The other one is admitting you don't know things. NPR's training team calls it feigning ignorance — asking the obvious question you technically know the answer to, so the guest gets to explain rather than confirm. Nervous people relax enormously when they're the expert in the room.

For the rambler, get specific and get shorter. Broad questions are fuel. Cut the fuel. Swap "how do you think about hiring?" for "what's the last hire you got wrong?" A question with one possible answer produces one answer.

And interrupt. Politely, warmly, but genuinely interrupt. Transom's guidance for radio producers is blunt about this: gently nudging people gets you better tape. Try "Can I stop you there, because I want to get this one thing on the record properly." Nobody has ever been offended by that sentence. Hosts avoid it out of politeness and end up with ninety unusable minutes, which serves the guest far worse than a clean interruption would have.

For pitch mode, ask about the failure. Media training covers the wins. It rarely covers "what did you get wrong in the first year, and what did it cost?" Ask for a number, a date, a name. Specificity is the thing scripts don't have.

If it keeps bouncing off, name it. Lightly: "I'll be honest, I think our listeners have heard the polished version. What's the bit you'd only tell someone over a drink?" That works more often than it has any right to.

For a genuine mismatch, stop recording. Say the real thing: "I think I booked us against the wrong topic and that's on me. Can we reset and talk about X instead?" Half the time you get a better episode than the one you planned. The other half you reschedule, which costs you a slot and saves you an unpublishable file.

What makes any of this possible

The rescue depends on things you set up beforehand, and this is where most bad interviews are actually lost.

You need a run-of-show you can read at a glance rather than a linear list you're marching through, because rescue means jumping to a different part of the plan mid-sentence. You need a few sacrificial questions parked at the bottom — the ones designed to open someone up, kept in reserve. And you need to have done a pre-interview, even a ten-minute one. Transom's producers treat the pre-interview as the place you find out where the stories are, so you're not discovering on tape that your guest is a nervous talker.

Do that and a bad interview becomes a navigation problem instead of an emergency.

Where an AI producer helps

This is the part of the job Liyam was built for, because the hard bit isn't knowing the techniques. It's executing them while you're also hosting, watching the clock, and managing a human being.

Liyam runs as a producer alongside your recording. It generates follow-up questions in real time based on what's actually being said, which matters most in exactly the moment your prepared list has stopped working and your own attention is fully occupied. Your run-of-show and questions stay on screen, organized by topic, so jumping to a different segment is a glance rather than a scramble. Optional pacing prompts keep you on structure without scripting you, and a live fact check surfaces sources on screen when something needs checking, so you don't lose the thread.

There's also a prep call agent that sits in on your prep call and drafts the outline, segments, and questions from that conversation. Which quietly solves the upstream problem: most rescues fail because the plan underneath was thin.

You can try it free.

When to give up

Some interviews should die. Stop if the guest is distressed, if the audio is failing in a way you can't fix, or if you've run all four plays and thirty minutes still produced nothing you'd want to listen to.

Ending early is not a failure. Publishing it would be.

Ending early is also a different decision from ending on time, which every episode needs whether it went well or not. If the conversation is fine and you are simply out of runway, how to end a podcast interview covers the wind-down signal and the closing question.

FAQ

Can you fix a bad podcast interview in editing?

Partly. Editing fixes structure, pacing, and tangents. It cannot manufacture a story the guest never told, which is why an in-the-room rescue is worth more than any post-production work. Cut hard, and be honest with yourself about whether what remains is worth an episode.

Should I tell the guest the interview isn't working?

Usually yes, and more directly than feels comfortable. Framed as your problem rather than theirs ("I don't think I'm asking the right questions here"), it resets the room without embarrassing anyone. Guests almost always sense it already.

How long should I wait before trying to turn an interview around?

Roughly ten minutes. Earlier than that you're reading nerves that would have settled on their own. Much later and you don't have enough usable time left to rebuild the episode.

What if the guest is just boring?

Boring is nearly always Mode 1 or Mode 3 wearing a disguise. Genuinely uninteresting people are rare; badly asked questions are not. Try the scene-based question and two seconds of silence before you conclude it's them.

The short version

Diagnose, then act. Closed-off guests need scenes and silence. Ramblers need narrower questions and a polite interruption. Pitch mode needs failure and specifics. A real mismatch needs you to stop and say so.

The skill isn't rescuing interviews. It's noticing at minute eleven instead of minute fifty.

If you want a producer handling the questions, the structure, and the fact-checking while you concentrate on the person in front of you, Liyam is free to start.