The exercise anyone can repeat.
Written for hosts of interview-style B2B podcasts, who book their own guests and are trying to reach people more senior than their last few episodes.
The prompt was not mine. I asked a search engine what hosts actually type into models when they want senior guests, then fed that back to another model word for word. One prompt, five seconds, no originality required. You can run it yourself before you read any further, and I would rather you did.
Seven sections, several thousand words, with citations. I read it twice looking for something to disagree with and mostly could not find anything. The advice is sound, and that is the whole issue.
Why sound advice fails.
Every host who asks gets the same play at roughly the same standard, and they all arrive in the same inbox saying structurally identical things. The guest you actually want lives above the floor, receives all of it, and deletes all of it.
The model told me to build a pipeline of 50 to 100 targets and keep 15 to 20 in active outreach at any time. Those two figures are quotations from the generated answer rather than advice from us, and they are worth reading as a description of what everyone else is now doing rather than as a target.
Reference their previous appearances to show you did your research. Good instruction. It is also the one the tool cannot help you carry out, because carrying it out requires reading things it cannot read and having an opinion it cannot have.
It cannot reach the audio
Paste the links into any model and it cannot pull the transcripts. It offers you the gist, and the gist is what everybody else is working from.
It sends you to the famous clip
The keynote, the big show, whatever ranks first on a search. Search your own name if you host anything. Whatever comes up first is what every pitch in your inbox opens with.
It has no opinion about your show
The playbook optimises the ask. Subject lines, word counts, touch cadence, calendar links. All of it assumes you already know what the episode is.
So the output is accurate, unremarkable, produced at volume, and aimed at somebody whose only defence against volume is deleting things quickly. What follows is the list of specific marks that get it deleted.
How many people got that same answer.
Two rules govern this table. Each figure sits on the same line as the population it was measured against and the index it came from. Where we looked for a number and found nothing that traces to a primary source, the row stays empty rather than getting filled with something plausible.
| Figure | What it counts, against what, from where |
|---|---|
| 3.8 million | Podcasts that exist in the Listen Notes index, counted by Listen Notes against its own index. |
| 519,708 | Shows active in 2026, the highest active count since 2021, counted by Listen Notes in June against the same Listen Notes index as the row above. |
| 156,000 | Shows that published in the week reviewed, out of roughly 720,000 that have passed ten episodes, from the Podnews study of the Podcast Index, whose base is about two million shows. This is a different index from the two rows above and the two counts do not stack. |
| two thirds | Share of shows that never reach ten episodes, from the same Podnews study of the Podcast Index and the same base of about two million shows. |
| no figure | Share of shows that are B2B. We looked. Every number in circulation traces back to a marketing post citing another marketing post, and none of them reaches a primary study, so this row stays empty. |
Two indexes appear above and they are not interchangeable. Listen Notes counts one population and the Podcast Index counts a smaller one, so dividing a figure from one by a figure from the other produces a number neither source supports. The two-million base also appears to be older than the 2026 Listen Notes counts, and we could not date it precisely, which is a second reason to keep them apart.
The empty row costs us a statistic. A page telling you to check where your quote came from has to survive the same question about its own numbers.
What gets it deleted.
Every specimen on this page is labelled. Where a line is verbatim model output it says so. Where a line was constructed to illustrate a shape, it says constructed. There are no unlabelled examples anywhere on this page, and no real pitch sent to us by a real person appears here at all.
Each tell says whether it is checkable or judgement. Checkable means you can settle it with a search, a word count or a transcript, and the answer is the same for everyone. Judgement means it depends on the guest and you have to decide. Six of each. Saying which is which costs us the appearance of rigour on half the list, and it is the only reason to believe the other half.
This page is set on an off-white background, which some catalogues of machine-writing tells list as a tell in itself. That rule is aimed at a warm beige reached for by reflex. This one is the colour the rest of the site has used since before the rule existed, and naming the objection here costs less than letting a reader find it.
Constructed, to show the shapeHi, I host a podcast about enterprise technology. We are in our second season, we release weekly, and previous guests have included leaders from several large companies. I would love to have you on.
Count the words before the guest is mentioned. Most pitches spend the opening paragraph on the show, the format, the download numbers and the guest list. The decision does not turn on who you are, so every sentence introducing yourself spends attention on a variable that is not in their decision.
The checkCount the words before you first mention something the guest said. If it is more than about fifteen, the opener is about you.
Credentials at the top do something worse than waste space. They read as pre-emptive defence, which tells the reader you expected to be doubted. One line covers it. Credibility arrives through specificity instead, which is a harder proof than anything you can claim about yourself.
The checkIs there a sentence defending the show against an objection nobody raised.
They run finance, the show is about finance, that is a fit. It is an overlap in job title. There are several hundred people with that title and they have all been approached this month. A category match produces somebody who can competently discuss your topic. A deep match is somebody whose current preoccupation sits on top of a question your show keeps circling without landing on.
The judgementWould this pitch still make sense if you sent it to a different person with the same title. If yes, you matched on the title.
There are two quotes worth opening with. One is under three weeks old, because the thought is still live and they have not finished having it. The other is over twelve months old and buried, from an appearance with almost no audience that nobody followed up on, because finding it took work a search engine cannot do. Everything between those two windows is dead ground. It is genuinely theirs and genuinely relevant, and it still reads as generic, because it is the first thing anybody finds and it has already been used on them this month.
The checkWhere did the quote come from. If a search engine handed it over in under a minute, it will read like a search result no matter how well you write around it.
Constructed, to show the shapeYou have spoken before about how important culture is during a merger, and I would love to explore that further.
Paraphrasing somebody's position back at them works in conversation and fails in a first message, because the single thing you are trying to prove is that you read the source. A paraphrase proves you read a summary. The reader cannot tell a paraphrase from a guess, and at four seconds they will not give you the benefit of the doubt.
The checkIs the quote verbatim, and can you produce the link and the timestamp. No link, no line.
Constructed, to show the shapeI really enjoyed your episode on AI in enterprise sales.
That names a real episode, so it passes as specific. It stays generic, because everyone else who watched that episode could have written the same sentence. Specificity gets measured against every other message in their inbox that week rather than against the guest's life.
The checkCould another host have written this exact sentence without doing anything you did.
You can get everything right and lose the guest at question four. Read each question as a person with an employer, a board, competitors and a reputation that took twenty years to build. Does answering it incriminate them. Does it invite them to criticise a competitor, their own company, or a predecessor's decision. Hard questions are fine. Ask the hard one, and tell them it is coming. The ambush is what costs you.
The judgementWould they be comfortable if this question, and their answer, were the clip.
The angle is real but sits so far off their main territory that showing up looks like a detour. They would be answering questions about something they think about occasionally, in a format where they get introduced as an expert in it.
The judgementIs this a subject they would want attached to their name in eighteen months.
The framing is built to stop the scroll, so the guest reads it and pictures the clip. Whatever you put in that opening line is the headline attached to them. Worth asking whether it is one they would forward to their own board.
The judgementWrite the episode title and read it as though you were them. If it is something they would hope their own communications lead never sees, better wording will not solve it.
The opposite failure. There is no episode in there, just a subject area, and the guest has to do the work of imagining what you would actually talk about. Busy people do not do that work for you.
The judgementCan you write the episode title. Not conversation with, an actual title telling a stranger what they would learn. Stalling here means you have an interesting person and no episode.
Something has to have changed. A piece they published, a role they moved into, a position they have taken that contradicts one from last year. Without it you are asking them to find time for a conversation that would have made equal sense at any point in the last three years, which is a polite way of saying next quarter.
The checkName the thing that changed, with a date. If you cannot, the pitch is not ready.
A pitch can fail because it was weak. It can also fail because nobody was home, and those look identical from your side, which is why hosts keep rewriting messages that were never opened. Plenty of senior people have a profile the way they have a fax number. Somebody set it up years ago and nothing arrives there.
The judgementBefore spending two hours on research, spend two minutes on whether a human being logs in. Wherever they are currently active is also wherever they are currently thinking out loud, so the hygiene check doubles as research.
Before and after, for prose.
Design comparisons work because a layout differs at a glance. Two paragraphs do not, and a contrast a reader has to work for is a contrast they will not feel. So the thing to put side by side is where each element of the writing came from, because an empty provenance slot is visible from across the room and a weak sentence is not.
Both columns below are constructed and labelled as such. The right column deliberately shows named slots rather than a finished sentence, because writing a persuasive fake and presenting it as the good example would be the exact behaviour this page exists to name. The slots are the argument, and filling them is the work.
Where each element came from
Five empty slots. The sentence is grammatical, warm and completely unsupported. It trips tell 06, tell 05 and tell 11 at once.
Where each element came from
Same five slots, each with something in it. Every one of them takes reading that a search result cannot give you, which is the entire cost of the good version and the entire reason it works.
Hold the guest constant when you try this on your own material. Take one person, write the version you would have sent, then draw the five slots underneath and fill in what you can actually source. The count of empty slots is the finding.
Fifteen minutes, on your own material.
The six judgement calls need the guest in front of you. These six do not, and you can settle all of them before lunch.
Most drafts fail at least one. The failure rate is the point: a pitch that clears all six is expensive to produce, and it is expensive in exactly the way the person receiving it can detect.
Nothing on this page is withheld. The method is published in full because a reader who follows it and books a guest without paying us has still proved the argument.
Route one · free
Read everything the guest has said in public. Find the thread nobody pulled. Work out what changed recently. Write it so the work shows in the first two lines. De-risk the questions so the yes is cheap.
It takes an afternoon per guest, and it works. The principles it rests on are written up in the booking playbook, and the matching method is on the golden thread page.
Route two · from $49 a month
The same sequence, with the reading automated. Every appearance transcribed and read, the candidate threads surfaced, the quote verified against its source before you see it.
The deciding does not move. You keep what is good, kill what is not, and send it yourself. Ninety seconds of research against an afternoon, which is the only argument we are making.
Hadef is in open beta. Small cohorts every few weeks, applications reviewed within a few days.
If you want the parts of a pitch set out one by one before you try either route, the anatomy of a pitch is built but not yet published, because its proof chain is not filled, and until it is, the golden thread page covers the matching logic.