How to Research a Guest Before You Invite Them

Most teams research a guest by reading their website and their LinkedIn bio, which are the two sources that tell you nothing except what that person wants said about them. We run outbound for 50+ B2B companies and have shipped over 8 million cold emails this year, and the research that earns a reply never starts in either place. Below, the 5 fields a research pass has to return, where each one actually lives, the time budget per name, and the rule that takes a name off the list instead of writing a softer invite around the gap.

How Do You Research a Podcast Guest Before You Invite Them?

Research on a guest is a data job, not a reading job. That sounds academic until you watch two people do it. One opens 6 tabs, skims for 10 minutes, feels informed, and writes an invite that could have gone to anyone in the industry. The other fills 5 fields, stops, and writes an invite only one person could have received.

The second version is faster, and it is the only one that survives volume. Reading produces a feeling. Fields produce a row, and a row can be checked, reused in the follow up, and handed to someone else without the knowledge walking out with whoever did the reading.

Research Pass

A fixed set of fields filled for one name before any copy gets written, with a named source for each field and a defined action when a field comes back empty. The pass is not a profile of the person. It is the minimum evidence needed to decide whether to invite them and what to say in the first sentence.

Own Words

Something the guest said or wrote where they were the one talking. An interview, a panel, a long post, a conference talk, a comment where they pushed back on someone. Marketing copy about them does not count, their own homepage included, because that was written to position rather than to argue.

Fit and copy come out of the same pass, which is why it is worth running once and running it properly. The fit decision itself is its own subject, covered in how to qualify podcast guests before you invite them and in the ICP gate that runs before any invite. This post is about the looking.

Why Does Reading Their Website Produce Nothing Usable?

A company website is an edited document. Every sentence on it was approved, so it says what the business wants believed about itself and carefully nothing that would complicate a sale. A LinkedIn bio works the same way. Both carry claims, and a claim is the one thing an invite cannot be built on, because the recipient already knows the claim. They wrote it.

Run the substitution test on whatever you pulled. Swap in a competitor's name. If the sentence still reads true, you did not research a person, you summarized a category. "Acme helps mid market manufacturers cut procurement spend" holds up for 40 other firms, which is why it lands as a merge field even when a human typed it by hand.

The reader's behavior makes this worse than it sounds. Nielsen Norman Group's work on scanning behavior found that people do not read text on a screen, they scan for a reason to keep going. A recycled description gives them nothing to catch on, so the email is gone before the ask arrives. Hunter's state of cold email report keeps landing on the same split across millions of sends, which is that depth of personalization rather than the presence of a merge field is what separates the top reply rates from the median.

There is a second reason to skip the homepage, and it is about the ask rather than the copy. Gartner's research on the B2B buying journey puts the share of a buyer's purchase time spent with any potential supplier at about 17%, spread across all of them. These are people with very little patience for a stranger who did not do the work, and a summary of their own positioning is the clearest possible signal that nobody did.

Which 5 Fields Should the Research Pass Capture?

Five fields, and the pass is done when all 5 are filled or one of them has forced a drop. Nothing else gets captured, because anything that does not change the decision or the first sentence is reading for its own sake.

Three of those 5 fields come off the row you already bought. That is the part teams skip, and it is the cheapest research in the stack, which is the argument made at length in how to personalize cold emails at scale. The expensive field is the artifact, and it is the only one worth paying attention for.

The fifth field is not research in the normal sense and it still belongs in the pass, because a perfect artifact sent to a dead address is a wasted hour and a bounce that hurts the next 1,000 sends. Both passes are covered in verifying emails before podcast invites.

Where Do Those Fields Actually Live?

Ranked by what each source returns per minute spent, cheapest first. The order matters more than the list, because most of a research pass should be finished before anyone opens a browser tab.

1. The row from the list pull. Title, headcount, industry, company description. This covers role, authority and fit for most names, and it costs nothing extra because you already paid for it. Teams that feel behind on research are usually sitting on 4 unread fields per lead. Tighten what gets pulled in the first place, per how to define an ICP for cold email.
2. Their past interviews. The single richest source for the artifact, and the one almost nobody checks. Someone who has been a guest before has already said something arguable out loud, and a podcast search tool like Listen Notes will find the appearances by name in seconds. Ten minutes of one episode beats all 6 of them.
3. Their long form posts. Not the reposts and not the milestone pleasantries, the posts where they argued with a standard practice in their field. A stance hands you the episode angle in the same breath as the first line, which is why it ranks above everything else in first line personalization on podcast invites.
4. Company announcements. A new service line, a new market, a senior hire, a product launch. Weaker than a stance because it carries no topic on its own, so the bridge into the episode has to work harder. Strongest while the change is recent enough that they are still explaining it to people.
5. Their own published numbers. A result in their own case study, a milestone they announced. Safe to use because they put it in public, and specific enough that it cannot be recycled onto anyone else. Never state a number about their business that you inferred rather than read.

Then the sources worth skipping, which is the half of this nobody writes down. Funding databases on a 12 person consultancy return nothing, because there is nothing to return. News alerts on a company that has never been covered produce noise. Awards pages, press release archives and conference speaker bios read like the homepage, because the same person wrote them.

The deliberately slow version of this exists too, and it is correct for a short list of names you genuinely cannot afford to miss. That is a different motion with different economics, laid out in account based podcast invites and in how to get high profile podcast guests. Everything in this post is the version that runs at 10,000 names a month.

How Long Should Research Take Per Name?

About 6 minutes by hand once the fields are defined, and 15 to 20 when they are not. The defined version wins on speed for a boring reason: you stop at 5 answers instead of reading until you feel informed. Feeling informed has no end condition.

Now the arithmetic that decides the whole design. At 10,000 invites a month, 6 minutes a name is roughly 1,000 hours of reading, which is 6 full time people doing nothing else. At our averages it takes about 95 invites to produce one completed recording, so that is 95 research passes per episode on the books. The full math is in how many invites it takes to book one recording, and the definitions behind those rates are in what a positive reply rate is.

So the research step gets built once rather than performed per person. One model pass per lead, reading the text already captured on the row, composing the artifact sentence from facts rather than hunting for them. No second lookup, no per lead web search, no model asked to be a researcher. A model handed a company name and a job title will invent a flattering sentence, and a model handed a real description and a real stance will write a line that only fits one business. Woodpecker's analysis across more than 20 million cold emails puts the lift from genuinely personalized copy as high as 142%, and the programs getting that lift are not the small ones. They are the ones where the research step is built rather than improvised.

Humans still read, just not 10,000 times. Twenty rows sampled before every launch, which catches the batch where enrichment degraded and every artifact came back about a conference nobody attended. That failure is quiet, it is common, and it is first on the list in common podcast acquisition failure modes.

When Does the Research Say Do Not Invite This Person?

Every field has a drop rule, and the drop rule is the part that makes the pass worth running. Research that can only ever say yes is not research, it is preparation for an invite that was already going out.

• No artifact in their own words. The name comes off. This is the rule that gets broken first, because the temptation is to write a warmer, vaguer line around the gap. That is exactly where invented compliments come from, and an invented compliment is worse than no invite.
• Fit fails the buyer gate. A fascinating guest who will never buy anything is a content decision, not an acquisition decision. Hold the line here or the show slowly fills with peers, which is covered in how to pick your first 100 podcast guests.
• No desk access. If the role does not sit at a computer as part of the job, the invite is not reaching them no matter how good it is. This is the gate that replaced industry filtering for us, and it is more predictive than either headcount or revenue.
• The address will not verify. Two passes, both have to agree. A guess costs more than the name is worth once bounce rates start moving placement, per where invites are landing.
• Already touched recently. A second invite from a new angle 3 weeks after the first one reads as a sequence, not an invitation. The follow up sequence owns that window, and research does not get to reopen it.

A healthy pass drops names constantly, and a list that loses 20% to 30% at this stage is working as designed. The volume comes from the guest list being wide enough to absorb the drops, not from lowering the bar on the names you have. Reply benchmarks by segment are in podcast invite reply rate benchmarks.

Dropping names also protects something people underrate. Google's bulk sender guidelines ask senders to hold spam complaints under 0.1% and warn that 0.3% causes real problems, and an invite that arrives with no evidence anyone looked is one of the faster ways to collect complaints. The research rule that protects the reader is the same rule that protects the domain.

Frequently Asked Questions

How do you research a podcast guest before inviting them?
Decide the fields first, then look. Role, fit, one artifact in their own words, the episode angle that artifact creates, and a verified address. Each field has one source and one rule for when it comes back empty.

How long should it take to research one guest?
About 6 minutes by hand with the fields defined, 15 to 20 without them. At 10,000 invites a month that is roughly 1,000 hours, which is why the step gets built once instead of performed per name.

Why is their website a poor source?
Because every sentence on it was approved. It carries claims the recipient already knows, having written them. Swap a competitor's name in, and if it still reads true, you summarized a category.

What should you look for in a guest's past interviews?
A position they argued, and the topics nobody asked them about. The first gives you the opening line, the second gives them a reason to accept your invite instead of repeating an answer they have given 5 times.

Should you mention their revenue or headcount?
Only numbers they published themselves, quoted the way they published them. An inferred figure about someone's own business loses the reader who knows the real one.

Can AI do the research?
A model composes well and hunts badly. One pass per lead over research already on the row produces a specific line reliably, which is the real answer to personalization at scale.

Does better research mean you can send fewer invites?
It lifts the positive reply rate on the same list, and a wrong list stays wrong. Segment first, then research, then write, per what to say when inviting a podcast guest.

What the Research Pass Is Actually Protecting

It is not the reply rate, though the reply rate moves. What the pass protects is the honesty of the invitation. You are asking a stranger for 40 minutes of their thinking, and the only thing that makes the ask reasonable is that you can name why it has to be them. A filled artifact field is that reason, written down, where someone else can check it.

Which is why the empty field has to end in a drop rather than a softer sentence. The moment research is allowed to return nothing and the invite goes anyway, the whole engine starts producing polite mail to people it knows nothing about, and no amount of volume fixes that.

It is also the discipline underneath a commitment like 30 recorded conversations with your ideal buyers in 90 days, or your money back. Invites go out by email only, every episode comes back edited and published, and the client owns the recording on their own show. None of that holds up without a research step that drops the names it cannot serve, because the engine starts with one person reading one sentence and believing it was written for them.