Guest Brief

guest-brief

The inversion this skill exists for

Every research tool answers "what exists about this person". That output is the direct cause of cringe questions, because the most findable material is the most repeated material. Ask from it and you ask what they answered on four podcasts last month. They give you the rehearsed version, and the listener can hear a rehearsed answer from across the room.

So this brief is built inside out. Its first section is what not to ask, and that section is measured rather than asserted.

This is prep, not leverage. The point is a better conversation, not a dossier.

Gate 1. Refuse before you research

Run this before writing anything. Two separate failures with two separate responses.

A substantive source contains the guest actually answering at length: roughly five minutes of speech or 800 words. A red carpet hit, a listicle, a press release, a label bio, a quote lifted into someone else's article and a social post are not substantive sources. Do not pad the count with them.

Every brief opens with this header, whether it passed or not:

SOURCES   7 substantive (of 19 found)
SPAN      2023-04-11 to 2026-08-02
MIX       4 video, 2 audio, 1 print

The span matters more than the count. Four sources all from 2023 passes a naive threshold and is still useless. Print the span so the reader can overrule the gate.

The mix matters because most current press is video. A saturation list built only from print silently under counts and looks identical to one that did not. If video is zero, say so in the header.

The unit of measurement: count the story, not the question

This is the part that makes the number real rather than decorative.

"So what made you start?", "talk me through the origin story" and "how did this all begin?" are one question wearing three costumes. Counting question wording sees three separate items and misses the most saturated thing in the file.

Cluster by the answer, not the prompt. The unit is the anecdote or claim the guest reaches for. If they tell the same story, it is saturated regardless of how it was pulled out of them. That is observable in a transcript and robust to phrasing.

For each cluster record four things:

• The story, in one line, in your own words.
• N of M, how many substantive sources it appears in.
• Most recent, the date it was last told.
• Where they stop. If they tell the same story nine times and cut it off at the same point every time, that stopping point is the most interesting thing in the brief. It is almost always the question.

The seven sections

1. Saturation. Do not ask these.

The clusters, ordered by N of M descending. Anything at 50% or above is a hard do-not-ask. Below that, note it and move on.

Present it as the stories they will reach for, not as a ban list. Knowing a story is coming lets you ask around it, or ask for the part they always skip.

2. Drift

What they said two years ago set against what they say now. People are most interesting at the exact point of a changed mind, and nobody asks about it because it requires having read the old thing.

"No drift found" is a legitimate output. Say it, and list the sources you checked underneath. Do not manufacture a change of mind because the section looks empty. That is the main way this section fails.

3. Contradictions in the public record

Where two sources genuinely disagree, or the bio and the work pull apart. Not gotcha material. Usually where the actual conversation is. Each one carries both sources with dates so the decision to use it can be made in the room.

Hard exclusion, no exceptions. Changes to identity, name, pronouns, gender, sexuality, or an account of their own past are never contradictions and are never presented as findings here. A public record that disagrees with itself on these points usually reflects someone who was closeted, unsafe, using a different name, or surviving. Route anything of this kind to section 5 as context, flagged as something the host probably already knows. It is not a contradiction. It is a life.

If applying this exclusion empties the section, the section is empty. Say so.

4. The unasked

Derived rather than found. What is structurally absent from the coverage:

• A period of years nobody covers.
• A collaborator who is in the early record and vanishes from the later one.
• The job before the famous thing.
• A failure everyone skips past to reach the win.

This section is inference, so label it as inference. Each item states what is missing and which sources you would have expected it in.

5. Live wires

What is genuinely raw right now: ongoing legal matters, a bereavement, a public fallout, health. Also the destination for anything excluded by section 3.

Not a list of things to avoid. A list of things to know before sitting down, so the choice to go there is deliberate rather than accidental.

6. Shared ground

Real overlap between the host and the guest: same scene, same city, same venue, mutual people, a night you were both at.

Do not fetch this from brand or bio files on your own initiative. Ask the host what the overlap is, or leave the section empty. An invented connection is worse than no connection, and this is the one section where the host is a better source than the internet.

7. The running order

Ten questions, in an order, not a ranking. Ranking by novelty puts the strangest question first, which is a bad interview. Warmth has to come before novelty.

Every question carries one line underneath saying why it is not saturated: either the cluster it deliberately sidesteps, or the stopping point it walks into.

Banned question shapes

These are the podcast equivalent of "delve". None of them may appear in section 7 in any phrasing.

• What advice would you give your younger self
• What does [the scene / the work / drag / the community] mean to you
• Talk me through your journey
• If you could tell people one thing
• What's next for you
• How did you get started, and every origin story variant
• What has been the biggest challenge
• Who are your inspirations
• Describe yourself in three words
• Did you always know
• What would you say to someone who

The failure they share: each one hands the guest a shape they have already filled in dozens of times, so the answer arrives pre assembled.

Gate 2. Self check before output

Both must pass. If either fails, fix the brief and run them again.

1. No question in section 7 appears in any saturation cluster at or above the threshold. Check every question against section 1 by hand, not by intention.
2. Every factual claim in the brief carries a source and a date. A claim without one is removed, not softened.

Anything in the brief that could embarrass the host if wrong goes through the fact-checker agent before the brief is handed over.

Calibration. Two poles, never one.

The saturation threshold can invert its own purpose. Set it too aggressively and every question with any traction is banned, leaving obscure trivia, which is its own bad interview. Nobody wants a host who has clearly done homework and forgotten to be interested.

Calibrate against two existing episodes: one the host rates and one the host finds flat. Run the saturation measure over both.

• If it flags most of the good one, the threshold is too aggressive.
• If it cannot separate the good one from the flat one, it is not measuring anything and the number should not be trusted.

A control with one pole is a vibe with a number attached. Do not calibrate against a single episode.

What this needs

This skill does not fetch. It assumes the sources have already been gathered, either by you or by a research pass before it runs. The brief is only as good as what is behind it, so two capabilities are genuinely required.

A way to read full pages, not snippets. Snippet level search results cannot support a saturation count, because the count needs the guest actually answering at length.

A way to read video and audio. Most current press is video. A saturation list built only from print silently under counts and looks identical to one that did not.

How people usually satisfy those:

If video cannot be read on a given run, do not silently fall back to print. Report it in the MIX line and state that the count is a floor rather than a total.

Output

Two artefacts.

1. The full brief, markdown, all seven sections plus the header, read the night before.
2. One page, holdable in the room, carrying only sections 1, 5 and 7. The saturation list, the live wires, and the running order. Nothing else fits and nothing else is needed while recording.

Write the questions as plain content. Do not write them in the host's voice. The host phrases them. A skill writing podcast host questions in someone else's voice produces exactly the cringe this brief exists to prevent.

Never use em dashes or en dashes anywhere in the output.