Open on your customer’s problem in eight words or fewer. Show your real product doing one job. End on the finished result and one next step. Fifteen to twenty seconds is usually plenty, and thirty is the most we’d start with. A free screen recorder is enough. Make three versions with different openings and see which one people watch.
Why a plain demo is worth doing
Makers often assume they need something beautiful. The evidence we could find points the other way. Motion’s 2026 report, covering 578,750 ads from 6,015 advertisers on Meta, puts demos among its biggest formats: about 12.6% of ads tested, 12.9% of spend, and a hit rate of about 8.1%. Offer-first banners hit at about 8.6%, and some rarer formats scored higher. In its table of asset types, pure animation hit at 4.57%, against 11.60% for plain text. Motion sells to ad buyers, its site wouldn’t load for us so we read search excerpts, and these are big spenders, not small accounts. It shows what tends to go with success, not what causes it.
We also looked at 39 ads ourselves in Meta’s Ad Library: active US video ads, sorted by impressions, across a few searches. Under “try it free”, 10 of 12 showed the app’s real output. Under a home repair search, 7 of 12 showed job footage with text on it. Pure motion graphics: 0 of 39. That’s a small count from poster frames only, and it leans toward AI photo apps and health products, but it matches. Of 124 builder complaints picked using marketing words, none were about looks. But when we searched for complaints about making content, 6 of 17 were about looks, mostly about one demo or launch video, so clarity comes first and looks still matter here.
The plan
| Beat | Seconds | What’s on screen | Words on screen |
|---|---|---|---|
| Opening | 1.5 to 3 | The problem, or the result. Not your logo | Up to 8 |
| First step | 2.5 to 6 | The first thing a person really does, cropped close | Up to 6 |
| Second step | 2.5 to 6 | What the product does next | Up to 6 |
| Result | 2 to 5 | The finished thing, on a real screen | Up to 6 |
| Proof, if you have it | 1.5 to 3.5 | One true number, or one customer’s real words | Up to 8 |
| Next step | 2 to 4 | One thing to do | Up to 6 |
A typical run of these comes to about 16 seconds. These are our own working numbers, not research, and the first post you learn from should overrule them. Words on screen do the talking, because plenty of people will have the sound off.
Six ways to open
| Pattern | Shape | Example |
|---|---|---|
| Flat pain | You did the work. Bad result. | You built it. Nobody came. |
| The old way | Still doing the painful thing by hand? | Still copying invoices by hand? |
| Watch this | Watch what happens when I do this. | Watch what happens when I paste a link. |
| Wrong belief | This belief is why you have the problem. | More features is why nobody signs up. |
| Result first | Open on the end result, then say how few steps it took | This took 2 clicks. Only if it did. |
| Real number | A number, then what it means. | 4 minutes. That’s the whole setup. Only if it is. |
One maker, who now sells a tool for this, reports that TikTok slideshows did nothing until the customer’s pain came first and the product came later. That’s one seller’s account of slideshows, not video, but it’s why the product doesn’t open here.
A worked example
Take a made-up tool that reads invoices. Everything on screen here is the real product, and any number in it has to be true.
- 0 to 2 seconds. A crowded inbox of invoice emails. “Still copying invoices by hand?”
- 2 to 6 seconds. Close crop: a PDF dragged into the app. “Drop one in.”
- 6 to 11 seconds. Close crop: the fields fill themselves. “It reads the numbers.”
- 11 to 14 seconds. The finished row in the spreadsheet. “Already in your sheet.”
- 14 to 17 seconds. The name and one instruction. “Free to try. Link in bio.”
How to make it, for free
- Pick the one moment that makes someone say “oh”. If your product has ten features, show one. Pick the screen with the result on it, not a settings page or an empty dashboard.
- Fill it with realistic sample data first. An empty account makes any product look dead. Don’t use a real customer’s data.
- Record it. On a Mac, QuickTime does this. Windows has its Snipping Tool and Game Bar, and iPhones and Android phones have screen recording built in. Record more than you need, moving slowly.
- Cut it to the plan in whatever editor you have. Crop in hard, so someone on a phone can read it.
- Add the words as text on screen, large, and keep them away from where the app’s buttons and captions sit.
- Export vertical, 1080 by 1920. Our sizes guide has the one file that fits nearly everywhere.
- Add sound last, if at all, and be careful which. It has to make sense with the sound off.
- Make two more versions with only the opening changed.
We’d guess an hour or two for the first one, and less after that. If you want it smoother, people in this thread compare tools. One builder pays for Screen Studio, at $30, only in the month of a release, and another makes demos with a coding agent and a free plugin they wrote, for $2 to $3 each in API costs. Both are their own reports. Before you post, run it through our free checker.
What usually goes wrong
- A whole page shrunk until nothing can be read. We’ve seen this in our own test videos. Crop to one part of the screen.
- A screen that sits still for more than about 2 seconds.
- Opening on the logo or the product’s name. Nobody knows you yet.
- Listing features instead of showing one thing happen.
- Words that describe what’s on screen, instead of what it means.
- A user interface redrawn to look nicer than the real one. If people sign up and it looks different, you’ve lost them.
How long, and where
Our starting points are about 18 seconds for TikTok and Reels, 20 for Shorts, and 25 on LinkedIn. We have no good data for small accounts on this. Nobody we read does. So post it, and watch how many people get past the opening and whether anyone replies. Views alone tell you little. If you’re new to posting, read the rules for each place first, and say you built it.
This is the recipe behind the first video in Lou by Luven AI, the tool we’re building, so we’ve had to think hard about it. Lou is meant to do this from your link, using your real screens. It isn’t open yet, and it hasn’t been tested on real customers. Doing it by hand once will teach you what your own opening should say, and that’s worth having even if you never use a tool.
How we write these: with AI help, from the sources we link, with dates. Counts marked as ours were made by an AI reader with no second check. If something’s wrong, email everest@luven.ai and we’ll fix it and say so.
