Automated Guest Booking: Building a Matchmaker Agent Without Monthly Fees
A matchmaker agent reads the network you already have, ranks it against your audience, and hands you a draft invite you send yourself.
A matchmaker agent automates guest booking by reading your existing contact list, scoring every person against what your audience needs to hear, naming the specific conversation only you and that person could have, and drafting the invite in your voice for you to review and send. It replaces the decision that stalls most calendars: who's next, and what to say. You build it once, in an afternoon, and it runs against your own data instead of a subscription database.
The reference implementation is Pod Green Room, the guest booking and sponsorship tool Jacqueline V. Twillie built for solo hosts after producing more than 300 episodes of her own show. What follows is the architecture underneath that kind of tool. Settle one thing before you start: you're still paying for a platform. What changes is what you're paying for, logic you wrote and contact data you keep, rather than a per-seat tool that holds both.
A matchmaker agent is an AI agent that ranks people you already know against a written profile of your show and audience, then produces a per-person angle and a draft outreach message, all from names already in your list.
What does the agent need before it can rank anyone?
Two files and one written brief. Get these right and the build is easy. Get them wrong and no amount of prompting rescues it.
Your contact export. A CSV from LinkedIn or your email client, with first name, last name, company, position, and connection date where you have it. Real connections only. A purchased list produces confident nonsense. The agent will score strangers as warm.
Your show record. One row per published episode: guest, topic, date. This is what stops the agent from proposing a conversation you already ran in March.
Your show brief. Four things in plain words: who your audience is, what your show is about, the angle that makes it yours, and the two or three themes you're building this season around. This brief is the thing the agent scores against, so vagueness here becomes vagueness everywhere downstream.
What are the three specialists that do the work?
Split the job. One agent doing all of it produces mush. Scoring and writing need different instructions.
The pattern Jacqueline teaches is a manager agent holding your profile and handing work to named sub-agents. In her job search agents build, a Job Search Manager hands off to Scout, which finds roles, Matcher, which scores fit and names two reasons a hiring manager might pass, and Ghostwriter, which drafts outreach from real samples of your own writing.
The same three roles map onto guest booking:
- 01
The reader.
Takes your contact export and normalizes it. Fixes company names: "Acme" and "Acme, Inc." become the same employer. Drops rows where the company field holds something that is not an employer, which LinkedIn exports are full of. This is unglamorous and it determines everything after it.
- 02
The matcher.
This is where each contact gets a number: a score against your show brief, a one-line basis for it, and which season theme the person serves. It also returns a confidence level. A job title from four years ago should not be scored as if it were current.
- 03
The ghostwriter.
For the top-scoring contacts only, writes the specific conversation angle and a short invite in your voice, from two or three real samples of your own writing. One sample is not enough for it to copy your rhythm.
How do you score a guest without lying to yourself?
Score fit, and keep warmth separate.
This is the rule that makes the whole list trustworthy. If the score quietly rewards how much you like someone, the ranking tells you what you already believe and you have automated your own bias. Pod Green Room scores audience fit and strategic value as their own numbers, on evidence from the person's known role, and never folds warmth into that figure.
Two more constraints belong in the matcher's brief.
Never invent a fact. If the agent does not know what someone currently does, it says so and scores with low confidence. A guess about a guest's job becomes a wrong claim in the email you send that guest.
Say what the score rests on. One line, citing the actual evidence. You should be able to push back on any score the agent gives you, and that's only possible if it shows its reasoning.
On matching contacts to companies, be deliberately unclever. Exact name match, or one name is the other plus more words, with no similarity percentages. A near-match that is wrong is worse than no match at all. The entire value of the list is that you trust the names enough to email them.
What does "automated" actually mean here?
The ranking, the angle, and the draft are automated. The send is not.
Put "draft, never send" in the agent's instructions and connect it to your email. Drafts land in your drafts folder. You read them, you fix the one line that is off, you send. That review step is what protects the relationship, and the relationship is the reason the person says yes at all.
Keep the trigger on manual for the first several runs. Move it to a schedule, weekly is plenty, only after you have read enough output to know how it fails. That progression, manual until tested, is one of the eight operating habits BNEDai's AI Readiness Index scores teams on, and it applies at a scale of one just as well.
What does it cost to run?
It costs something, and the honest version matters here.
You're trading a per-seat booking platform for a build platform running instructions you wrote. The instructions are yours, they move between tools, and the contact data stays in a file you hold. That is the ownership argument, and it is a real one.
The platform still costs money. Gumloop, the platform used in Jacqueline's live builds, runs on a 14-day trial and then a paid plan. Model usage costs whatever your platform charges for it. What you're buying is real: logic you own and data you keep.
If you would rather have the done-for-you version, that is exactly what Pod Green Room is, built by a solo host for solo hosts. Building your own makes sense when you want the scoring rules to be yours and to keep working the same way in two years.
What breaks a matchmaker agent?
Four things, in the order they happen most often.
A dirty contact file. Half the rows have no company, or the company field says "Freelance." The reader step exists to handle this, and skipping it is why most first attempts return nonsense.
A vague show brief. "Business and leadership" is too broad to score against. If the brief cannot distinguish two plausible guests, neither will the scores.
Prompting the ghostwriter without voice samples. The model copies whatever voice you give it. Give it nothing and you get corporate filler, and the guest can tell.
Trusting the first run. Read twenty scored contacts before you send anything. You'll find one obvious error in the first ten, and finding it is the point of the exercise.
Where to start this week
Export your contacts and your episode list. Write the four-line show brief. Build the matcher first. Run it on your own list without connecting email at all. The only output is a ranked file you read.
If the five-move build sequence is new to you, start with How to Build Your First AI Agent. For the wider question of which parts of a podcast operation belong to you, see AI for Podcast Hosts. Once the booking pipeline is running, the sponsorship conversation is the next one, and L.A.T.T.E. for sponsor negotiation covers it.
Frequently asked questions
What is a matchmaker agent for podcast booking?
An AI agent that ranks people you already know against a written profile of your show and audience, then produces a specific conversation angle and a draft invite for each top match. It sources no new names. Its job is telling you what is already in your contact list, in an order you can act on.
What data do I need to build one?
Three inputs. A contact export with first name, last name, company, position, and connection date. A record of your published episodes, one row each with guest, topic, and date, so the agent does not propose a conversation you already ran. And a four-line show brief naming your audience, your topic, your angle, and this season's themes.
Should the agent send the invites automatically?
No. Put "draft, never send" in its instructions. Drafts land in your drafts folder for review. Keep the trigger on manual until you have read enough output to know how the agent fails, then move to a weekly schedule. The review step is what protects the relationship that makes a guest say yes.
How do I keep the guest scores honest?
Score audience fit and strategic value separately from how warm the relationship is, so the ranking stays useful when it disagrees with you. Require the agent to name the evidence each score rests on in one line, and to report low confidence rather than guessing when it does not know what someone currently does.
Is building my own matchmaker agent actually free?
No. You swap a per-seat booking platform for a build platform running instructions you wrote and own, with your contact data in a file you keep. The build platform still charges: Gumloop runs on a 14-day trial and then a paid plan, plus whatever model usage costs.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an agent that is already running.