How Many Episodes Should You Check Before Pitching?

The same shows, read twice

Both reads ask one question of one thing: does this episode's own title credit a guest. Same test, same shows, same day. The only difference is how far back it goes.

The second row is not the truth either, and this is the one page here that can put a number on its own ceiling rather than a caveat. A show whose last guest sat fifty episodes back is invisible to both rows. So we read every show the 40-episode pass placed nowhere out to 120 episodes, and 60 of them credit a guest past 40. That makes the second row 76.7% of the shows in this corpus that demonstrably book guests rather than all of them, and every number on this page is written on that side: the real miss rate is worse than 54.8%, not better.

One number worth stating because it must be true: the count of shows that show a guest in the short window and none in the long one is 0. The short window is the newest 5 episodes and the long window is the newest 40 of the same feed, so a guest visible in the first has to be visible in the second. If that count were not zero, this comparison would be measuring two different days rather than two different depths.

Why a short scan fails, and on which shows

The failure is not random. It falls almost entirely on the shows that seat a guest occasionally, which is most of the shows that seat one at all.

Read the first row, because it is the market. Of the 197 shows that book guests, 105 do it in one episode in ten or fewer, and a 5-episode scan finds 20% of them. That is close to what chance alone would deliver, and it is the row a researcher spends most of their time in.

The ladder is the mechanism, stated plainly: 20%, then 46.9%, then 75%, then 100%. A quick scan is not a weak version of a real check. It is a check that works perfectly on the shows you did not need it for, and barely at all on everything else. The interview show whose every episode is a guest is the one show you could have identified from its name.

And it is worth being explicit about what this does NOT say. It does not say the shows a short scan misses are eager for pitches, or that they are good bookings. It says only that rejecting them for having no guest on the first screen is rejecting them for something that is not true.

So how many episodes is enough?

Because we recorded the position of each show's first guest credit rather than just a count, the question has an answer at every depth rather than only at 5 and 40. Here is what each extra episode buys.

The shape is the finding. The first episode alone finds 45 of the 197, the shows that put a guest in every episode and that you could have identified from the show's name alone, and then the returns decay slowly and never stop. Getting from half to three quarters costs 9 more episodes, and getting from three quarters to nine in ten costs 14 more. There is no knee in this curve where the scrolling stops paying.

Which makes the practical answer a budget rather than a number. If you are cutting a long list down, read to about 7 episodes and accept that you are keeping roughly half of the real targets; the cut is cheap and you know what it costs. If you are about to spend a week writing pitches to twelve shows, read all 40 for those twelve, because at 36 episodes you are at 95% and the remaining doubt is smaller than the doubt in everything else about the pitch.

And the honest version is in what each row ADDS rather than in the percentage: past the first handful of episodes, every row above is still worth a few more shows on a list of this size. That is a rate that stays flat much longer than intuition says it should, and it is why "I scrolled a bit" is not a check.

And 40 episodes is not the bottom

Everything above is a share of the shows a 40-episode read can see, which is the honest way to write it and also an admission. A show whose last guest sat sixty episodes back is not counted as un-found in that table. It is not in the table at all. So we went and read them.

The population is exactly the 161 shows the 40-episode read placed nowhere (a show already placed at episode seven cannot move), and every one of them was re-read to 120 episodes with the same walk and the same title test. 23 of those feeds turned out to have no more episodes to read, so they are un-found rather than merely un-reached. Of the 138 that did carry more, 60 credit a guest past episode 40: 43.5% of them.

Read the chance per episode, because it is the finding. Between episode 41 and episode 120 the chance that one more episode is the one that answers the question stays between 0.43% and 1.27%. It does not decay toward zero, and the deepest band we read, episodes 101 to 120 on shows already scanned a hundred deep, still turns up 9 more. The curve above has no knee inside 40 episodes, and this is the same curve continuing to have no knee three times further out.

Which means the honest total is a floor with a direction attached. This corpus holds at least 257 shows that credit a guest somewhere in their recent history; a 40-episode read finds 197 of them, or 76.7%, and the 5-episode scan a feed gives you for free finds 89. We are not claiming 257 is the real number. 58 of the shows we re-read hit our own new ceiling with nothing showing, and the rate above says a deeper read would place some of them too.

None of which is an argument for reading a hundred episodes. It is an argument about what a null result means. "No guests" never graduates into evidence by scrolling a bit further; it just gets slightly less wrong, at a rate you can now see. The decision that actually changes is the one on the other side: when a show does credit a guest, you have proof, and proof arrives at a depth you choose.

What to do instead

Four changes, in the order they are worth making:

• Scroll, and know how far. Reading 40 titles instead of 5 costs about twenty seconds in any podcast app and recovers 108 shows in a list this size. If twenty seconds a show is too much, 7 is the depth where you have found half of them and 30 is where you have found nine in ten. Nothing else on this page has that ratio.
• Do not treat "no guests on the first screen" as a rejection. Treat it as no information. The check has an answer only when it finds something; when it finds nothing it has told you how far back you read, and that is all.
• Check the show's own description too, not just its titles. A show that says it interviews people and credits nobody in a recent title is a different case from a show that says nothing, and the two checks agree far less often than people assume.
• Spend the depth where the decision is. Reading 40 titles for every podcast you have ever heard of is not the point; reading them for the twelve you are about to spend a week pitching is. A shallow pass is fine for cutting a hundred candidates down to twenty, as long as you know the cut is throwing away real targets and you go deep before you write anything.

The wider version of this is the one worth keeping. Every check a guest runs on a podcast is a read of a window, and the honest form of every one of them is a floor rather than a verdict. "Publishes no contact route" means we did not find one. "No new episode in six months" means the feed has not moved in six months. "No guests" means not in the episodes you read. The number on this page is what that distinction costs on one check, on one list, on one day.

How we measured it, and what would break it

Two reads of the public RSS feeds behind our own directory, joined show by show. The short read is the episode snapshot the directory stores for each show and renders on its pages: the 5 most recent episodes. The long read is a separate pass that walked up to 40 recent episodes per feed and counted, among other things, how many titles credit a guest. Both use the same strong-tier title test, imported from one module rather than written twice, and the pass that produced the long read stores counts only.

4 feeds are excluded from every figure above rather than counted as zeroes: some published too few readable episodes for the long read, and some had fewer than 5 episodes stored, which means nobody ever ran the 5-episode scan on them. 3 of the 358 that remain published no more episodes than the directory already stored, so for those the two reads are the same read; they stay in the denominator, and they pull the measured miss rate DOWN.

The title test is deliberately conservative. It counts a title as crediting a guest only when it does so unambiguously, so a show that names its guest after a pipe or in the notes alone counts as having no guest at BOTH depths. That under-counts bookers in the long window, and the bookers it drops are the ones with the least visible guest activity, which are exactly the shows a short scan is least likely to catch. A better test would find more shows to miss, not fewer.

The depth itself comes from the long read: as it walked each feed in order it recorded the position of the first title that credited a guest, using the same call that produced the count, so the curve and the totals are one measurement rather than two. That gives an accidental cross-check worth printing, because the two files answer the SAME question at the shallow end by two routes. The stored snapshot's 5 episodes and the long read's own first 5 agree on 344 of the 358 shows. They disagree on 14: 8 shows have seated a guest since the snapshot was taken, and 6 have published enough new episodes to push their guest out of the newest 5. That is the size of the elapsed-time error in this comparison, measured rather than asserted, and it runs in both directions.

The third read is the one that bounds the second. Because a show the 40-episode pass already placed cannot move, the only shows that can be hiding a guest deeper are the ones it placed nowhere, and there are 161 of those inside this page's population. All 161 were re-read to 120 episodes by the same code, on the same day, and the results live in their own file with their own window: nothing in the tables above moved, and nothing was estimated or scaled up from a sample, because the whole subpopulation was read. One number is worth stating for the same reason the inverted count is: 0 of the re-read shows credited a guest inside the newest 40 episodes. That has to be zero, since the 40-episode pass walked those same titles and placed nothing, and a non-zero would mean the two reads happened far enough apart to be measuring the calendar.

Three things would change the answer, and the first one now has a measured size. A longer window than 40 episodes raises the true number of bookers and widens the gap: at 120 episodes it is 60 shows wider. A test that read episode notes as well as titles would find guests both reads currently miss. And if the two files were ever generated weeks apart, the comparison would silently start measuring the elapsed time instead: the 0 inverted shows above is the number that says they were not.

The corpus is the same one behind every data page here: 384 shows, 362 with a feed we could read, gathered between July 16, 2026 and August 26, 2026. The roll-up is a single module with its own test suite, and no figure on this page is typed by hand.

Sources and benchmark date

Every figure above is derived at build time from reads of the public RSS feeds behind 384 podcasts, taken on August 27, 2026: a 5-episode snapshot, a 40-episode pass over all of them, and a 120-episode pass over the 161 the second one placed nowhere; the corpus itself was gathered between July 16, 2026 and August 26, 2026. Method, limitations and the raw data are public:

• Which Podcasts Actually Book Guests? (our study)

  The full analysis behind the corpus: which shows publish visible evidence they book guests, how to reach them, and the whole dataset under CC BY 4.0.

• Does This Podcast Interview Guests?

  The other half of the same check: what a show's own description claims about guests, how rarely that agrees with its episode titles, and why running one of the two checks finds about half the shows.

• How Often Do Podcasts Actually Have Guests?

  The volume behind the bands above: how many guest seats the still-publishing shows visibly fill in a year, and how few shows hold most of them.

• Is This Podcast Still Active?

  The same window problem on the other check a guest runs, where the honest answer is also a statement about how far back you read rather than about the show.

• How Many Podcasts Should You Pitch?

  What a list looks like after all four checks, and the two honest list sizes that come out of the same feeds.

Frequently asked questions