THE PROOF
Guests who fit the show drew more views.
Before we pulled any data, we wrote the test down and committed it in public. Then we checked 755 episodes from eight of the biggest podcasts on YouTube. The link is small, it passed the test we set in advance, and it was positive in all eight shows, though barely in one.
Method committed 19 Oct 2026 / Data pulled 21 Oct 2026 / 755 episodes / 8 shows
It does not show cause and effect, audio downloads, or how smaller shows behave. Those limits are spelled out at the end.
ONLY THE TOP AND BOTTOM FIFTHS WERE NAMED IN ADVANCE. THE MIDDLE THREE ARE SHOWN FOR CONTEXT.
755 episodes in five groups by Qloo fit. Median views for their age, against the bottom group: 1.00×, 1.18×, 1.35×, 1.21×, 1.55×.
Each dot is one episode.
755 episodes from eight of the biggest podcasts on YouTube, published from Oct 2024 to Jul 2026, each at least 20 minutes long with one guest named in the title.
Sorted by how well the guest fit.
Qloo scores how closely a guest's fans match a show's listeners. We split each show's episodes into fifths by that score, from the weakest fit to the strongest, then put the shows' fifths together.
Then we looked at views.
Each ring is a fifth's median views, adjusted for how long its episodes had been up, against the bottom fifth's.
The best-fitting fifth drew 1.55× the bottom fifth's views.
The middle fifths don't climb in a straight line, so read it as a tilt, not a promise. A better fit raises the odds of a bigger episode; it doesn't decide one.
THE TWO TESTS
Both passed the rule we set in advance.
The rule, written before the pull: a test passes when its correlation is above zero and the low end of its 95% interval is above zero too. The intervals come from 10,000 resamples.
Fit and views
Views adjusted for how long the episode had been up.
Fit and views, fame held equal
The same test after taking out how famous each guest is in general.
CHECKED AFTER THE FACT
Without the misread episodes (683)
- Fit and views
- 0.149 (0.070 to 0.226)
- Fame held equal
- 0.100 (0.017 to 0.184)
Resampling guests, not episodes (603 guests, 2,000 resamples)
- Fit and views
- 0.065 to 0.211
- Fame held equal
- 0.020 to 0.167
One row per guest
- Fit and views
- 0.153
- Fame held equal
- 0.111
We added these after the result, at a reviewer's request. They are not part of the pre-registered test, and neither changed a verdict.
SHOW BY SHOW
Positive in all eight.
Each show's own correlation. The filled dot is fit and views; the hollow one holds fame equal. Shows with more episodes give steadier numbers.
Armchair Expert with Dax Shepard
196 episodes
0.157 / 0.140
The Joe Rogan Experience
164 episodes
0.132 / 0.049
Call Her Daddy
127 episodes
0.026 / 0.033
This Past Weekend w/ Theo Von
73 episodes
0.125 / 0.105
Conan O'Brien Needs A Friend
62 episodes
0.222 / 0.056
Club Shay Shay
53 episodesCUT SHORT
0.126 / 0.119
The Diary Of A CEO
43 episodes
0.104 / 0.102
Impaulsive
37 episodes
0.410 / 0.296
Cut short: its channel's 2,000 most recent uploads only reach part of the way back.
WHAT WE LEFT OUT
From 11,927 uploads to 755 episodes.
Every rule was written before the pull. Each bar shows what one rule took out.
- Uploads read11,927
- From another show on a shared channel−922
- Published outside Oct 2024 to Jul 2026−5,181
- Shorter than 20 minutes, such as clips and shorts−4,215
- Title doesn't name exactly one guest−403
- Guest not found in Qloo under the same name−257
- Qloo gave no fit score−194
- Kept755
SHOWS LEFT OUT BY THE RULES
- Hot OnesQloo's podcast search found no podcast with this name.
- Good Hang with Amy PoehlerQloo's podcast search found no podcast with this name.
- FlagrantTwo Qloo podcasts matched, and the rule drops the show rather than pick one.
- SmartLessIt has no official full-episode YouTube channel.
THE NAMES
44 of 50 names were right.
A program reads each guest's name from the episode title. We checked 50 of them by hand, drawn at random with a fixed seed, against the title and the show's format. Most misses were Armchair Expert's "Fact Check" episodes, which name an earlier guest who isn't on them.
WHAT THIS DOES NOT SHOW
- Cause and effect.
- Qloo learns from what audiences enjoy together, so a hit episode may itself strengthen the link between a guest and a show. Holding fame equal doesn't remove that.
- Downloads.
- YouTube views are not podcast downloads, and many listeners never open YouTube.
- Smaller shows.
- These are some of the biggest podcasts there are. Mid-size shows, the hosts Vasal is built for, may behave differently.
- Every guest.
- Episodes whose guest isn't named in the title, or isn't in Qloo, are missing. The bars above show how many.
CHECK OUR WORK
The test was public before the data existed.
19 Oct 2026, 05:41 UTC
The method and the list of shows were committed, before we had a Qloo key.
52e208f19 Oct, 10:25 UTC
Each show's YouTube channel was pinned.
4a2a8ca21 Oct, 10:16 UTC
Three fixes to the title reader went into the change log, found by reading 12 titles and nothing else.
40547e021 Oct, 10:52 UTC
The pull behind every number on this page started.
dataset 20261021T105230Z
21 Oct, 15:14 UTC
The results were published as they came out.
f513835
The dataset holds Qloo and YouTube data, so it stays private. The code is public, and with your own keys one command reruns the whole pull:
npm run backtest -w packages/agent -- --dev-db
The pull behind this page used 915 Qloo calls. View counts keep moving, so a fresh pull comes out slightly different.