Viral reels
Design short-form videos the way outlier videos are actually built: as a chain of three psychological beats, each doing one job.
Stop the scroll -> hold attention -> pay off.
Every breakdown of a video that massively outperformed its creator's normal reach comes down to one deliberate trigger per beat (fear, curiosity, surprise; pattern interrupt, curiosity, payoff; story, surprise, product). Videos that flop usually miss a beat: a strong idea with no reason to stop, a great hook with nothing holding the middle, or a payoff that does not answer the question the hook raised. So the job is not "write a catchy script". The job is to choose the three triggers on purpose and then build everything else around them.
No method guarantees virality, and you should never promise view counts. What this method does is remove the avoidable reasons a reel fails. If the user expects a guarantee, say this once, plainly, and move on.
Pick the mode
Work out which of these the user needs. If it is unclear, the default is Create.
| Mode | The user has | The user gets |
|---|---|---|
| Create | A topic, product, offer or vague wish for content | A shoot-ready reel plan |
| Breakdown | A reel (link, description, transcript or screenshots) | Why it worked, beat by beat, and how to reuse the structure |
| Fix | A script or published reel that underperformed | A diagnosis by beat and a re-cut |
| Research | A niche and no ideas | A method (or, with browsing tools, a shortlist) of outlier videos worth modelling |
| Batch | A need for several reels | A set of plans that vary triggers and formats |
Before anything: who is this for
Read references/brand-profile.md. If it is filled in, use it and do not re-ask what it already answers.
If it still says it is not filled in, take what the request gives you. Three things change the plan, so if they are missing and cannot be inferred, ask for them in one short message (and if the user wants you to just proceed, state your assumptions and go):
- Niche and audience: what they sell or talk about, and to whom.
- Goal of this reel: reach, follows, comments, or sales/leads. The goal decides the payoff (see step 2).
- What they can actually film: themselves talking, a product, a place, a screen recording, or nothing (text and stock only). A plan the user cannot shoot is worthless, however clever.
Create mode
1. One idea, one goal
State the single idea the reel carries, in one sentence. If the user's topic contains three ideas, pick one and say the others are separate reels. Viewers decide in seconds; a reel that tries to say several things says none of them.
2. Choose the chain
Read references/triggers.md. Pick one trigger for each beat. The goal narrows the pay-off beat:
| Goal | Pay-offs that serve it |
|---|---|
| Reach and shares | Unexpected twist, emotional misdirection |
| Rewatches | Frame break, twist that changes the meaning of the opening |
| Comments | Unclosed loop, a side to take |
| Follows | Missing payoff that continues on the profile, a repeatable series |
| Sales or leads | Product as the punchline, attention transfer, delayed product reveal |
Offer three candidate chains that differ in their stop trigger, since the first seconds matter most and the user should see real alternatives. Name each chain in three words the way a good breakdown would ("tension, curiosity, reveal"). Recommend one and say why in a sentence, taking into account what the user can film.
A trigger is a mechanism, not a label. For each one, say what the viewer will see and what question it puts in their head. "Pattern interrupt" means nothing until it is "a full-size chess clock sitting in a frying pan".
3. Build the first three seconds
The hook has three layers and they must agree with each other:
- See: the opening image or action. This carries most of the weight. Many outlier hooks have no words at all.
- Read: on-screen text, seven words or fewer, readable with the sound off.
- Hear: the first spoken line or sound.
Write the recommended hook, then three alternates that keep the same chain but change the execution. These are what the user will A/B test.
Check the hook against the beat it sets up: it must open the question that the hold beat sustains and the payoff answers. A hook that promises one thing and delivers another gets the stop and loses the watch time, and it teaches the audience not to trust the next one.
4. Write the beat sheet
Read references/formats.md and pick the format that fits the idea and what the user can film. Then write the reel second by second using the template in assets/reel-plan-template.md.
Guidelines, with the reasons:
- Length: default 20 to 40 seconds. Go shorter for a pure visual gag, longer only if every stretch earns its place. Longer videos need more reasons to stay.
- Label the trigger at work on every row. If a row has no job, cut it. Dead seconds are where viewers leave.
- Slow down just before the payoff. Rushing the punchline is one of the most common differences between a flop and a hit cut from the same footage.
- Fewer words on screen. Every unnecessary word is friction. If a label can be understood without speaking the language, it travels further.
- Give the eyes something to follow during any stretch of talking: hands doing something, a process, a change on screen.
- Leave imperfection in when it is real. A mistake reads as human and breaks the "this is an ad" pattern; a flawless take is predictable.
5. Payoff and call to action
The payoff must answer the question the hook opened, or deliberately refuse to (unclosed loop, missing payoff) for a reason tied to the goal. Then one call to action that matches the goal: a keyword to comment for a resource, a prompt to pick a side, a pointer to the profile for the result, or the offer itself. One action, not three.
6. Caption and cover
Write a caption of one or two lines that adds to the reel instead of repeating it: it can extend the joke, plant the comment prompt, or carry the keyword. Add three to five relevant hashtags only if the platform and niche use them. Give cover text of two to five words for the profile grid.
7. Check it
Read references/checklist.md and run the plan through it. Fix what fails. Report the result honestly at the end of the plan: what passes, and the one or two real risks that remain (for example "depends on the reveal being visible in the first frame").
8. Deliver
Use the structure in assets/reel-plan-template.md exactly, so plans are comparable from one reel to the next. Keep commentary outside the plan to a few lines: the recommendation, the main risk, and what to test first.
Keep the whole reply tight. Give each candidate chain two or three lines, do not restate the plan in prose, and put a detail in one place only. The user has to film this; a plan too long to read on a phone between takes does not get used.
Businesses that serve one area
For a clinic, shop, restaurant or any business whose customers are local, reach only counts where the customers are: a few thousand views in town are worth more than millions elsewhere. Add local signals (the place name in the caption and text, recognisable locations, the location tag), prefer formats that show the real place and staff, and judge the reel by enquiries and bookings that mention it, not by the view count.
Breakdown mode
When the user brings a reel, find out what happens in it. If you have a link and browsing tools, watch it or read what you can; captions rarely explain a video, so ask the user to describe the first three seconds and the ending if you cannot see them. Do not guess at the content of a video you have not seen.
Then answer in this shape:
- What happens, in two or three sentences.
- The chain: stop trigger, hold trigger, pay-off trigger, each with the mechanism (what is shown and the question it raises), plus a three-word recap.
- Multipliers present (timing, word economy, imperfection, format reuse). See
references/triggers.md. - Outlier check, if numbers are available: views against the creator's follower count and usual views. A video is only evidence of a good format if it travelled beyond the existing audience.
- What to take: the structure and psychology, restated without the original's content.
- Versions for the user's niche: two or three adaptations in a few lines each, ranging from the closest structural match to a freer one that fits the niche better, each with its own twist. Recommend one and write only that one as a full plan. Copying the content is both unoriginal and ineffective; the structure is the reusable part, and the closest copy is rarely the best fit.
If the breakdown surfaces a trigger that is not in the library, name it, describe its mechanism, and tell the user it is worth adding to references/triggers.md.
references/case-library.md has worked breakdowns to compare against.
Fix mode
Diagnose by beat before rewriting anything. For each beat, say whether it exists, which trigger it uses, and whether it works:
- No stop: the opening explains instead of showing, starts with a greeting or logo, or looks like every other video in the niche.
- No hold: the question is answered immediately, or no question was opened, so there is no reason to stay.
- No payoff: the ending does not resolve the opening, the punchline is rushed, or the product arrives as an interruption instead of the answer.
Also check the common drags: too many words on screen, more than one idea, dead seconds, nothing moving while someone talks.
Then give the re-cut as a full plan in the standard template, and list what changed and why. Keep whatever was already working; a flop and a hit are often the same footage with a different first three seconds and a slower build to the punchline.
If the user has performance numbers (views, average watch time, where viewers drop), use them: an early drop points at the stop beat, a mid drop at the hold, a late drop at a payoff people saw coming.
Research mode
Read references/outlier-research.md. If you have browsing tools and the user wants you to look, follow the method and return a shortlist with a breakdown of each. If not, give the user the method as short steps they can do in a few minutes, and offer to break down whatever they bring back.
Batch mode
For several reels, do not write the same chain five times. Vary the stop trigger and the format across the set so the user learns what their audience responds to. Put the plans in a short table first (idea, chain, format, goal), get agreement if the user is present, then write each plan in the template.
Add the rule that makes a batch pay off over time: when one reel clearly outperforms, make it again. Same opening, same format, new subject. Creators who break out usually found one opening that worked and repeated it; they did not keep inventing.
Lines not to cross
Some triggers work by briefly misleading the viewer. That is fine when the reel resolves it and nobody is worse off; it is not fine when the deception is the product.
- Do not fake safety, health or emergency warnings, and do not present staged events as real news. These get reported and suppressed, and they cost the trust the account depends on.
- Do not invent numbers, results or testimonials for hooks.
- The payoff has to honour the hook. Misdirection that ends in a satisfying reveal earns shares; bait that never delivers earns one view and a lost follower.
- Do not copy another creator's video. Take the structure, write new content.
- Keep the stunt safe to film. No plan should require anything dangerous to get the opening shot.
If the user asks for something over these lines, say what the risk is and offer the nearest version that keeps the trigger without the harm.
Producing the video
This skill ends at a plan that can be shot or edited. If the user wants the video itself made and a video-production skill or tool is available in the session, hand the finished plan over as the brief: the beat sheet maps directly to scenes, and the on-screen text column is the caption track.
Reference files
references/triggers.md: the trigger library by beat, with mechanism, ways to stage it and how it fails. Read for every Create, Fix and Breakdown.references/formats.md: reel formats with timing skeletons. Read when writing a beat sheet.references/checklist.md: the pre-publish check. Read before delivering any plan.references/outlier-research.md: how to find and vet videos worth modelling. Read in Research mode.references/case-library.md: worked breakdowns of real outlier videos. Read when you want a precedent for a chain or the user asks for examples.references/brand-profile.md: the user's niche, audience, voice and constraints. Read first, every time.assets/reel-plan-template.md: the output structure for every plan.