You don't have a product to show yet. Good. You have something better for this stage: a problem that real people complain about.
Short clips about the problem do two jobs at once. They're research, because the comments tell you whether the pain is real and how people describe it. And they're audience building, because everyone who says "this is me" is a future customer you can reach again. This guide covers hooks, a 15-second structure, an example two-week plan, how to read comments as data, and a Claude skill that turns your complaint log into scripts.
Why film the problem, not the product
A product demo asks people to care about you. A clip about the problem asks them to recognize themselves. The second is easier, and it works before you've written a line of code.
It also keeps you honest. If ten clips about a problem get a shrug, a slick demo of the solution won't save it. If one clip gets a pile of "how do I fix this?" comments, you've found your opening.
Hook formulas
The hook is the first line on screen and the first thing you say. It has one job: make the right person stop scrolling. Fill these in with phrases from your complaint log (see Find app ideas in Reddit complaints).
- "If you [do the job], you know [the pain]." If you run an Etsy shop, you know the custom-order spreadsheet.
- "Is it just me, or [the pain]?" Invites replies, which is what you want.
- "Nobody warns you about [hidden cost]." Good for problems people only discover after starting.
- "POV: it's [time] and you're still [doing the task]." Shows the pain instead of describing it.
- "Stop [the workaround]." Only if you can show a better way in the clip, even a manual one.
- "People keep asking Reddit for an app that [does X]." Only if your log really shows that. Don't invent a count.
Write three hooks for every clip and pick the one you'd stop for. Test different formulas across the first week so you learn which one your audience responds to.
A 15-second structure
Fifteen seconds forces you to cut everything except the problem. Here's a structure that fits:
| Seconds | Beat | What's on screen |
|---|---|---|
| 0 to 2 | Hook | Text on screen plus the spoken line |
| 2 to 7 | Show the pain | The real thing: the messy spreadsheet, the 14 tabs, the sticky notes |
| 7 to 12 | The cost or the twist | What it costs ("that's my Sunday gone") or what makes it worse |
| 12 to 15 | Question | Ask how they handle it. End on the question, not a logo |
Screen recordings and your hands on a keyboard are fine. You don't need to be on camera, and you don't need a product.
The same vertical clip can go to TikTok, Instagram Reels and YouTube Shorts. YouTube's help center says vertical or square videos up to three minutes uploaded after October 15, 2024 are classed as Shorts, so 15 seconds is well inside every limit you'll meet. If you want to stretch one recording into several posts, Turn one video into a week of posts covers that workflow.
A two-week posting plan
Here's an example cadence, not a rule. Pick something you can keep up without dreading it.
| Days | What to post | Why |
|---|---|---|
| 1 to 5 | One clip a day, each using a different hook formula on your top complaint cluster | Find which hook works |
| 6 | No new clip. Read every comment and log it | Research day |
| 7 to 10 | One clip a day on the strongest hook, aimed at the second and third clusters | Test which problem pulls hardest |
| 11 to 12 | Reply to the best comments with video answers | Turn commenters into conversations |
| 13 | One clip asking directly: "If this existed, would you use it?" with your waitlist link where allowed | First ask |
| 14 | Review. Log, count, decide | Pick your problem |
On TikTok you can reply to a comment with a video. In TikTok's own tutorial, you tap the comment, tap the camera button beside the comment box, and record (the app may have moved things since). TikTok can show the original comment as a sticker so viewers know what you're answering, and the reply posts as a new video on your profile. Comment replies often make your best clips, because someone else wrote the hook.
Read the comments as research
Views tell you the hook worked. Comments tell you whether the problem is real. Log them in the same complaint log you already have, with the clip as the source. Look for:
- "Me too" with detail. "Same, I have a whole Notion board for this" is worth ten heart emojis. Specifics are signal.
- Named workarounds and tools. These are your competition, and the reasons people give for disliking them are your positioning.
- Price talk. "I'd pay for that," "I already pay X for Y," "too expensive." Write it down word for word.
- Disagreement. "Just use a calendar" is useful. It tells you who doesn't have the problem, or which objection you'll need to answer.
- "Where can I get it?" Your best signal. Reply with the waitlist link (where the platform allows links) and log the person's situation, not their handle.
Reply to good comments with one follow-up question: "How do you handle it now?" or "What's the worst part?" You'll get answers people would never give in a survey. When the replies start sounding the same, you've found your language. Then point people at a waitlist and a refundable reservation (see Validate before you build).
Turn your complaint log into scripts with a skill
A skill is a saved set of instructions Claude Code can run on demand. You write it once as a SKILL.md file, then call it with a slash command. (For the bigger picture, read Skills are the new SOPs.)
Create this file in your project:
---
name: clip-scripts
description: Turn the complaint log into 15-second clip scripts about the problem.
argument-hint: "[number of scripts]"
disable-model-invocation: true
---
Read research/complaints.csv. Write $ARGUMENTS short video scripts
about the problems in it. Do not mention or pitch a product.
For each script:
1. Pick one complaint cluster. Name it and list the rows you used.
2. Write three hook options using these formulas:
- "If you [do the job], you know [the pain]."
- "Is it just me, or [the pain]?"
- "POV: it's [time] and you're still [doing the task]."
Use the customer's own words from the quote column.
3. Write the script in four beats with timings:
0-2s hook, 2-7s show the pain, 7-12s the cost, 12-15s a question.
4. Add a shot list: what to film or screen-record for each beat.
5. Add one comment question to pin.
Rules:
- Only use quotes and details that are in the file. Never invent
numbers, quotes or experiences.
- Mark any claim that needs checking with [CHECK].
- Plain, spoken English. No hype words.
Write the scripts to scripts/clips.md, newest at the top.
The disable-model-invocation: true line means Claude only runs this skill when you type the command, never on its own initiative. That's what you want for a skill that writes files. $ARGUMENTS is replaced by whatever you type after the command, so this asks for five scripts:
/clip-scripts 5
Read every script out loud before you film it. If a line sounds like an ad, cut it. If Claude marked something [CHECK], check it or delete it. Then, after each review day, add the new comment-quotes to the log and run the skill again. The scripts get better as the log gets more specific, which is the whole point: your audience is writing your content for you.
What you should have after two weeks
A clear favorite problem, a handful of hooks that work, a log full of the words people use, and a small group of people who've told you they want a fix. That's a better starting point than most launches get. Build for those people.