The difference between a podcast that gets booked again and one that doesn't is rarely the audio quality. It's whether the host showed up having read the work. Not the headlines. The actual work — the second book, the long-form essay from three years ago, the obscure interview where the guest said something that contradicts what they say in public now. That kind of preparation makes a guest lean in. The lazy first-question prep makes them coast.
This guide is about building a guest-research workflow that produces real preparation, repeatedly, without burning your weekends. The agent does the slow reading and synthesis pass. You do the second pass — the one where your taste decides which threads are actually worth pulling.
What guest research is for
The job of guest research isn't to know everything about a person. It's to find the two or three angles that other hosts haven't explored, that the guest will actually want to talk about, and that connect to what your audience cares about. That's a synthesis problem on top of a reading problem.
The reading problem is annoying. A serious guest has a book, a few major essays, twenty interviews, a Substack, and a public-speaking history. Reading all of it before each episode isn't realistic if you're recording weekly. Most hosts compromise by skimming, and the conversation suffers.
The fix isn't to read less. It's to put the material into a vault and let the agent do the first pass — the one that maps the territory, surfaces the contradictions, and lists the questions every other host already asked. You spend your prep time on the second pass, where your judgment matters.
The guest dossier — one page per guest, deepening over time
A "Guests" section in the vault holds the running record. One page per upcoming or past guest. The page accumulates over months. For recurring guests or people you might book again, the dossier is years deep by the third interview.
The dossier holds: bio, the work you've gathered, your own observations, the questions you actually want to ask, the questions you've decided not to ask, and after the recording, the post-interview notes about what landed and what surprised you.
The work you've gathered is the bulk of it. Bio. Recent essays. Their book if there is one. A couple of past interviews. A long-form video where they were unguarded. Drop everything on the page. PDFs auto-convert to markdown via docstrange so the book chapter you saved becomes searchable text. Audio (a past podcast they did) transcribes with speaker labels so you can see what they actually said versus what the host's framing suggested they said.
The structure pairs naturally with the broader podcast workflow described in AI notes for podcasters — guest research is one slice of the show's notes system.
The agent's first pass — the reading you'd do anyway
Once the material is on the page, ask the agent to do the unglamorous reading: "Read everything on this guest's page. Pull the three threads I should ask about that don't show up in any other interview I have here. List the questions other hosts always ask so I know to skip them. Note any contradictions between their published positions over time. Surface anything they've said in passing that might be the seed of a real conversation."
You get a starting point. Not a finished prep doc — a draft to react to. Some of the suggested threads are obvious; you ignore those. One is genuinely interesting; you pull on it. Two are wrong but trigger your own thinking about what the right thread actually is.
The first pass is the work that makes the difference between competent prep and the kind of prep that makes a guest pause and say "nobody's asked me that before." It's also the work most hosts don't do because the reading takes hours and the synthesis takes more hours. With the agent doing the reading and the structural draft, the synthesis is the hour you actually have.
The second pass — your taste, your questions
The second pass is yours. It's where the prep becomes a real conversation plan rather than a checklist of clever questions.
Read the agent's draft. Cross out the threads that don't actually interest you. Add the ones the agent missed because they require knowing your audience or your style. Rewrite the questions that the agent phrased in a generic way. Add the unscripted parts — the moment you want to leave open, the place where you want the guest to talk for ten minutes uninterrupted.
The output is a working prep doc on the dossier page. Three to five questions you actually want to ask. A loose order. A few backup questions in case one thread dries up. A short list of references you might want to mention. That's it. The doc doesn't need to be longer.
For recurring guests, this pass also asks: what changed since last time? "Pull what they've said in the past year on the topic we covered last time. Has their position shifted? Is there new material that should reshape the next conversation?" The conversation gets better because it's actually building on the previous one, not starting from scratch.
The contradiction file — where the real questions live
The most interesting questions for a podcast are often the ones that surface a contradiction the guest hasn't been asked about. The book says one thing in chapter three; the recent essay says something different. The interview from 2021 took a hard position; the interview from 2024 is softer. The Twitter thread last month said something that doesn't quite square with the book.
The agent is good at finding these. "Read everything on this guest's page. List every place where their published position has shifted over the past five years. For each shift, suggest what the most generous interpretation might be."
You get a list of real conversational openings. Not gotchas — openings. The good ones get used: "I noticed in your 2019 essay you said X, and in your recent book it sounds more like Y. What changed?" The guest gets to do the work of explaining their own evolution, which is the kind of moment podcasts exist for.
The contradiction file is also where the agent earns its keep against the hours-of-reading problem. A human host re-reading three years of a guest's work to find these moments would spend a day. The agent does it in a minute. You read the list, make judgment calls, and write the questions that make the conversation feel earned.
Pre-interview reading — the bit that's still on you
The agent does the broad reading. There's still a smaller pass that the host has to do personally — usually the most recent piece of work, often the book, sometimes a single long-form essay that the conversation might revolve around.
Read it the day before. Take notes on the dossier page. Mark the passages you might want to quote. Type three lines about what surprised you, what you'd push back on, what felt unfinished. Those three lines are the most useful part of the prep doc on the day of the recording.
For the rest of the work — the past interviews you haven't time to listen to, the secondary essays, the older book — the agent's first-pass synthesis is enough. You're not the world's expert on the guest. You're the host who's done enough preparation to ask a real question.
The post-recording note
After the recording, type three lines on the dossier page. What landed. What surprised you. What you wish you'd asked. What the guest mentioned that's worth following up on later.
Three lines. Two minutes of typing. Over the run of the show, those post-recording notes become some of the most useful pages in the vault. They're the only place where you've honestly recorded what the conversation actually was, beyond the polished post-published narrative.
For recurring guests, the next prep starts from the post-recording note: "Pull the post-recording notes from the last conversation with this guest. What did I say I'd follow up on? What's still unresolved? What new direction did the conversation suggest?"
The back-research, recurring asset, and a boundary on AI
A long-running show accumulates real guest research. After fifty episodes, you have fifty dossiers. Many of those guests will come back, or be referenced by future guests. For new guests who connect to past ones — "Pull every past guest who has worked in or written about regulatory policy. What's the running thread across their views?" The agent reads across the dossiers and surfaces connections you wouldn't spot manually. For seasonal review, ask the agent to read across a quarter of dossiers: "What kinds of voices have I leaned heaviest on? Where am I underweight?" For the broader content workflow this fits into, see how to build a content calendar from your notes and content repurposing across platforms.
A practical note: the agent is doing reading and synthesis on material you provide. It's not validating that quotes are real or that the guest hasn't been misrepresented. The host's job is to verify before quoting, especially when surfacing a contradiction. The agent also doesn't replace the human work of actually being interested in the guest. A prep doc that's purely AI-generated reads like one — flat, a little generic, missing the small specific question that comes from genuine curiosity. Use the agent to compress the reading. Do the synthesis yourself.
A calmer way to prep
Podcast prep doesn't have to feel like a Sunday-night cram. The cram comes from doing all the reading the day before the recording. The shape that holds is small ongoing capture (drop the new article into the dossier the day it comes out) plus a focused thirty-minute block the day before recording, where the agent's first-pass synthesis is already done and you're doing the second pass that requires your judgment.
Try Docapybara free — sign up, pick your next guest, drop in their bio plus one past interview and one of their published essays, and ask the agent for the three threads no other host has pulled.