8 Audio Overview prompts that make the podcast worth listening to

All of these go in the focus prompt of the customize dialog (the pencil icon on the Audio Overview tile). They compose: pick a host dynamic, add a listener, set a length. If you have not used the dialog before, the full audio guide covers the basics of formats and lengths.

1. Give the hosts opposing jobs

Host A is 'The Expert' who explains the material. Host B is 'The Challenger' who pushes back: question the methodology, point out gaps in logic, play devil's advocate. Do not agree just to be agreeable.

The default hosts validate each other constantly, and after three episodes the "so true" rhythm starts to grate. Adversarial roles produce actual discussion, and as a bonus the Challenger tends to find the real weaknesses in your material.

2. Define the listener, not the topic

The listener is a first-year student who has never seen this material and is nervous about it. No jargon without an immediate plain-language explanation. Assume good intelligence and zero background.

"Make it simple" produces condescension. A concrete listener produces calibration. Swap in whoever you need: a CEO with four minutes, a skeptical engineer, your grandmother, a hiring committee.

3. Script a narrative arc

Start with the problem: [X]. Build toward [Y] as the surprising answer. End with what the listener should do differently tomorrow.

Without an arc the hosts wander through your sources in file order. With one, the episode has a beginning, a reveal and a takeaway, and you will remember it days later. This is the difference between information and a story, and it costs one sentence.

4. The hour-long version

Do not summarize. Go through every section in detail, discussing every concept, example and number in the sources. The listener has time and wants depth over brevity.

Combine with the Longer length setting. The hosts are summarizers by default; explicitly banning summarization changes their objective, and hour-long episodes from a dense document become normal. Commuters live on this one.

5. The five-minute cap

Be concise and stay within five minutes total. Spend no more than one minute on background. If forced to choose, keep the findings and cut the context.

The inverse hack. Also works per-topic: "one minute maximum on history, the rest on the current debate." More dependable than the Shorter setting alone.

6. The exam-prep episode

The listener sits this exam tomorrow. Cover only chapters 3 and 4. Quiz each other on the key definitions, walk through one worked example each, and end with the five questions most likely to appear.

Constraint plus format plus payoff. Students who pair this with the flashcards and quiz tools get a full revision loop: listen on the bus, drill at the desk.

7. Hold a language

This episode is ONLY in Portuguese. Do not switch to English at any point, including for technical terms; explain them in Portuguese instead.

Audio Overviews officially support 80+ languages via the output language setting, but the explicit "only" instruction keeps the hosts from drifting when your sources are in English. Language learners generate the same episode at two difficulty levels by adding "speak slowly and use common vocabulary" to one of them.

8. Prime it for interruptions

I will interrupt with questions using interactive mode. Treat them as genuine curiosity: go deeper on what I ask, not broader. If I push back on a claim, engage with my objection rather than smoothing it over.

Interactive mode (the Join button) lets you question the hosts mid-episode. Unprimed, they tend to give a polite non-answer and return to script. Primed, it turns into a tutorial. English only for now.

FAQ

Where do these prompts go?

In the focus prompt field of the Customize Audio Overview dialog, opened with the pencil icon on the Audio Overview tile before you generate.

Can I change the hosts' voices?

No. The voice pair is fixed. You can change their behavior, roles, language and pacing through the customize prompt, but not the voices themselves.

Do customize prompts work in every language?

The focus prompt works broadly, and audio generation supports 80+ languages. The length selector (Shorter/Default/Longer) and Interactive mode are currently English-only.