AI music for podcasts: the intro is where this pays for itself

A theme at the top, fifteen to thirty seconds, the same every week.

A sting between segments, five to fifteen seconds, used two or three times an episode.

An outro under the credits, fifteen to thirty seconds, often the theme again.

That list is short and it is finished. Everything else in a podcast is people talking.

Which means the whole music requirement for a show is two or three tracks, generated once, at 20 credits each.

Sixty credits and an afternoon buys the entire sonic identity of a show. The 300 free credits with no card cover fifteen attempts at getting it right.

Listeners learn a theme through repetition. It becomes the sound of your show somewhere around the tenth episode.

Changing it resets that, which is why a theme that is slightly imperfect and consistent beats a better one introduced in month six.

So spend the attempts up front. Generate five, sleep on it, generate five more, then commit.

The track you loved in the first hour is often the one you tire of by episode ten. Novelty is loud, and it fades.

Save the description that produced it. Regenerating from the same words gives you a sibling track for a special episode or a season change.

There is no editing after generation, so a theme that is nearly right needs a new description rather than a fix.

Speech occupies roughly 500 Hz to 4 kHz. Consonants sit at the top of that band, where they are quietest and easiest to mask.

A bed under two people talking for forty minutes costs clarity the whole way through and adds atmosphere nobody asked for.

It also fatigues a listener wearing headphones, which is how most of your audience is listening.

Podcast listeners are usually doing something else, so anything competing with the words costs comprehension rather than mood.

If a segment genuinely needs underscoring, keep it to a short passage. Describe the track as sparse, no midrange, low end and high percussion only.

The ceiling is 120 seconds anyway, so an hour of bed is not available even if you wanted one.

A theme carries an expectation. A driving electronic opening promises pace, and a slow acoustic one promises reflection.

If the promise and the show disagree, listeners feel it in the first ten seconds of every episode for a year.

So describe the conversation rather than your own taste. Two people arguing gently about books is a better brief than any genre name.

Suggestion chips write into the description rather than replacing it, so start from a genre and then add what the show is actually like.

Ask for the tempo explicitly. Slow tempo around 70 BPM and fast tempo around 140 BPM both work as instructions.

Then play it to somebody who has not heard the show and ask what they expect. That answer is worth more than your own opinion of the track.

A show that changes its opening in month three has taught its audience nothing. The twenty credits was never the reason people delay it.

Sixty credits and one afternoon removes an item from the list of small unfinished tasks that quietly stop shows from launching.

Generate the sting in the same session while the description is still fresh. Coming back in a fortnight produces a track that does not match the theme.

Four plain limits sit around that. Nothing runs longer than two minutes, and joining pieces is work you do elsewhere.

Stems, ducking and a mix somebody signs off on belong to a full audio editor.

A specific recognisable song for a theme is a rights negotiation rather than a generation problem.

And there is no retiming or trimming of a returned track. A close-but-wrong result means a new description and another 20 credits.

The track is generated for you rather than licensed from a catalogue. Confirm the usage terms that apply to your own account before it fronts a public show.

Keep the description, the date and the file together. A question about a theme running for three years arrives eventually.

Then leave it alone. The value of a theme is entirely in how many times the same one gets heard.