Borrow formats from niches that aren't yours

"Familiar but unexpected" is the phrase Paddy Galloway uses for a format that works, and most people hear the second word. They brainstorm a shape nobody has seen, ship two episodes, and call the miss proof that the audience is tired. The first word is doing the work. Familiar means the viewer already knows how to watch it. Unexpected means they have not seen this payload inside that shape. You get both by transplanting a proven structure from a niche your audience does not follow, not by inventing one.

Invention is expensive early because a new skeleton has no baseline. A borrowed skeleton has a run of working examples you can strip to beats, then one test to see whether those beats survive the payload swap.

Familiar-but-unexpected is a transplant

A format is a fixed structure plus one variable. The cold open, beat count, end card and caption style stay put. The object in the slot changes. That is the definition used on Versely's repeatable series formats, and it is why a transplant is cheaper than a blank page: you copy the fixed part from a run that already survived contact with an audience.

What you are not copying is the topic, the talent, the in-jokes or the niche's face. Copy those and you have a costume. The test is whether a viewer who has never seen the source niche still understands the game by second three.

Two failure modes sit on either side.

Too close. Borrowing "day in the life" from another creator in your own vertical is not unexpected. It is the same show with a different logo. Audience fatigue is usually the format repeating, not the model that rendered it.

Too loose. Keeping only "it is a list" or "someone talks to camera" is not a structure. A structure specifies the first frame, the number of turns, where the verdict lands, and what the last second withholds. If you cannot storyboard the empty version, you have borrowed a vibe.

This or That is a split-frame verdict with a locked label style. Speed Run is a visible process against a clock. Those are skeletons. They became skeletons because some other niche ran them until the beats were obvious.

A sourcing method that does not start in your niche

Do this as a sitting, not as a mood board.

Write five niches your audience does not subscribe to. Not adjacent. Unrelated. A dental clinic does not need more dental clinics. It needs courtroom shows, restoration channels, speedrunning, radio call-ins. If you can name a creator in that niche that your viewers watch, throw the niche out.
In each niche, collect fifteen to twenty videos that share a shape. Same cold open, same number of turns, same end card. If you cannot find a cluster, you found a one-off. One-offs do not transplant. You are looking for a run, not a hit.
Write the skeleton with the nouns removed. Frame 1 is the object on a table. Seconds 2–8 are three labelled options. Seconds 9–14 are the test. Last two seconds are the withheld score. No product names, no niche slang. If the card still makes sense, you have a format.
Check the demand column. Some skeletons require source material you do not have. Restoration needs an archive. Rate-my-X needs inbound submissions. A weekly changelog needs a weekly change. The series-format selection layer is built around this constraint on purpose. A beautiful skeleton you cannot feed is not a candidate.
Keep three skeletons. Not ten. Three is enough to test and few enough that you will actually test them.

A shortcut that looks like research and is not: grabbing the highest-viewed item from a "formats brands can copy" round-up. Those lists are usually format costumes sitting on a promotion goal nobody named. Steal structure from a niche that has been running the shape for months. Ignore the view count on any single file.

Worked examples, as payload swaps, not as claimed case studies:

Source niche	Skeleton you strip	Payload in a different vertical
Speedrunning	Clock on screen, one process, no talking over the finish	Kitchen reset, warehouse pick path, studio lighting setup
Restoration	Then / now held on the same frame	Product version 1 vs version 4, office before the move
Courtroom shows	Charge, evidence, verdict card	"Is this landing page legal," "is this claim on the tin"
Radio call-in	One submitted question, one answer, one withheld follow-up	Customer question of the week for a studio that already has a support inbox
Fashion "rate my fit"	Submission, three criteria, score	Rate my homepage, rate my packaging, rate my onboarding email

The last row is how Rate My X exists at all. The source is not "brand content." The source is a comments-section sport from a different internet.

Strip the payload, keep the skeleton

The transplant fails in production when people "adapt" the skeleton. Adaptation is how the first frame becomes a logo sting, the verdict card becomes a paragraph, and the runtime doubles. You no longer have the thing you borrowed, so you cannot read the test.

Rules for the swap:

First frame stays a first frame. If the source opens on the object, you open on the object. A presenter intro is not a localisation; it is a different format. OpusClip's April 2026 analysis of 2,847 clips using visual hooks found Visual Demonstration the largest subcategory (1,316 clips). Do not "improve" a demo open into a talking head because that feels more on-brand.
Beat count stays. Three turns stay three. Adding a fourth "for context" is how the unexpected part dies.
End card stays a card. A spoken CTA over a fade is not the same game as a labelled verdict.
Caption preset stays one preset. Recolouring every episode is a weekly decision the format was supposed to remove.

You can still change the variable: new object, new guest, new question. If you change skeleton and payload in the same week, you will not know which one failed. A meme format is a narrower case: the skeleton is already public; the work is launching a visual variant that can survive a week of copies. That is a launch problem, not a sourcing problem. Do not mix the two sheets.

The three-episode transplant test

Do not commit a season to a skeleton you have not run. Do not run it once and call the feed lottery a verdict.

Pre-commit, in writing:

Sample. Three episodes of the transplant, at your normal slot, at the source skeleton's length band. Not five different skeletons once each.
Control. Your current core format, posting in the same weeks, so the comparison is against your median rather than against a screenshot from the source niche.
Metric. The same one you would use to kill a series: Viewed vs. Swiped Away on Shorts, completion or average watch on TikTok, sends per reach on Reels. Pick it before you look.
Stop rule. Continue to a season only if the three-episode rolling median of the transplant sits at or above your core format's median for the same length band. One pop does not count. Three consecutive misses on the median does.

Posting the three episodes as simultaneous A/B variants on the same day cannibalises the pool. The off-platform protocol that actually holds is one variable, consecutive days, same clock time. Here the variable is the skeleton. Body quality, caption style and hashtags stay put.

Three episodes is a screen, not a season. If the transplant clears the bar, lock the skeleton for a real pilot and stop tinkering with the first frame. If it fails, log the skeleton and the median, then run the next candidate. Do not "fix" a failed transplant by adding a presenter. That is a new format, and it resets the counter.

Milk the one that clears. Galloway's other half of the line is to run a working format while it still delivers rather than retiring it because a copycat showed up. Copycats are evidence the skeleton is readable. They are not a stop rule.

FAQ

Is it theft to lift a skeleton from another niche?

A beat structure is not a clip. Do not lift their footage, catchphrase, end-card illustration or audio. Do not clone a living creator's face or room. You are taking the game: three labelled options, a clock, a then-and-now hold. If you cannot describe the skeleton without naming the source show, you are still too close.

What if my audience will not "get" a format from a distant niche?

That is the screen the three-episode test exists to run. Some skeletons need in-group knowledge (a scoring system, a slang caption, a nested joke). Those fail the "nouns removed" step and should have been discarded in sourcing. If the empty storyboard is readable, the audience does not need the source niche. They need the first frame to teach the game, which is the skeleton's job.

Can I transplant two skeletons in the same month?

You can source two. You should not test two in overlapping slots. Each test needs your normal slot and a control. Run three episodes of A, then three of B. The calendar feels slow. It is faster than a season of a skeleton you never isolated.

Do I need a different model to make a borrowed format work?

Almost never. The miss is usually that the first frame no longer matches the source game, or that you cannot feed the demand column (no archive, no submissions, no weekly change). Swapping generators to chase fidelity is how format problems get misread as model problems. Hold the skeleton still. Change the payload. Read the median.