The honest division of labour
| Part of the video | Generated | Recorded |
|---|---|---|
| Opening / hook | Yes — scenario, reaction, presenter | Also possible |
| Problem context | Yes, if it stays generic | Optional |
| Product interface | No — can invent screens | Required |
| The action | No | Required |
| The result | No | Required |
| Caption and CTA | Rendered text overlays | Recorded or overlaid |
The moment a viewer needs to believe the product does something, the footage must be real. A generated dashboard can look convincing and still be a liability.
Build the demo around one task
Choose the viewer question
Start from a question a real person asked: “Where do I find the next step from a client call?” The answer becomes the task. If you cannot name the question, the video has no job yet. Thecustomer-question guide turns a support thread into a sequence.
Script the words beside the screens
Write each line next to the screen state that proves it. Use theSaaS demo script worksheet; it includes a 30-second example and a 15-second cut.
Capture readable proof
Record the starting screen, the action and the result. Keep the control and the outcome legible at phone size. If a wide interface loses a required control, plan a close-up instead of cropping it away. Theapp recording guide covers framing and privacy.
Use AI for the opening — then hand off
A generated creator or text-led opening can make the problem feel immediate. Then cut to the recording. Keep the handoff clean: the opening names the task, the recording performs it, and the ending shows the result and the next step.
Vidmake supports this combination through its AI UGC video generator and the talking-head plus demo workflow, with a review step before download. Theproduct demo editing guide covers the assembly.
Review rubric before publishing
- Does the first three seconds name the situation the viewer recognises?
- Is the starting screen understandable without prior knowledge?
- Is the exact control used visible and readable?
- Does the result stay on screen long enough to inspect?
- Is any generated footage clearly not presented as product proof?
- Are sensitive records, names and notifications removed?
- Do captions match the audio and the words on screen?
- Does the next step work for the intended viewer?
Common failure modes
- Generating the UI: invented screens and labels.
- Speeding up the result: implying the product is faster than it is.
- No hold on the outcome: the viewer never sees what changed.
- Unreadable crop: the button that matters disappears outside the frame.
- Leaked data: a real customer name or email left in the recording.
FAQ
Can an AI generate a product demo video?
It can generate an opening, a scenario or a presenter. It cannot be trusted to show your real interface doing a real action. For the demonstration itself, record the actual product.
What should a product demo video show?
One task: the starting screen, the action, and the visible result. Keep the control you use and the result readable on a phone. Remove anything that does not help the viewer follow that task.
How long should a SaaS demo video be?
As short as the task allows. A single action can be clear in 15–30 seconds; a multi-step workflow may need longer. Length is a comprehension decision, not a target to hit.
Do I need a script for a demo video?
A short one. Pair every line of narration with a screen state the viewer can see. If a sentence has no visible evidence behind it, rewrite it or remove it. The SaaS demo script guide includes a worksheet.
Can generated footage replace a screen recording?
No. Generated footage can invent buttons, labels and behaviour. It may introduce the problem, but the proof of the product should be a genuine recording you reviewed.