Product demo videos for startups, in one afternoon

A startup demo video is a ninety-second screen recording that shows one workflow end to end: the problem in the first ten seconds, your product doing the thing, the outcome, and one instruction. You can write it in twenty minutes, record it in an hour, and cut it before dinner. It needs a screen recorder, a microphone, and a demo account with believable data. It does not need a motion designer, a script doctor, or a budget.

The reason most founders do not have one is that the phrase demo video summons the wrong reference. What comes to mind is the polished brand film with the swelling piano and the animated wordmark, which costs thousands and converts nothing, because it never actually shows the product. The version that earns its place is closer to handing someone your laptop and saying watch this. Unglamorous, specific, and finishable today.

What follows is the whole production: the five placements where a demo changes an outcome, a fill-in beat sheet with real timestamps, the recording setup that reads as professional without costing anything, the ordering rule that decides your watch-through, and the five surfaces one recording can cover.

Where does a demo video actually convert?

In five places, and each one imposes a different constraint on the file. Build the ninety-second narrated master first, then cut down for the placement rather than shooting five separate videos. Knowing the placements before you write is what stops you producing an asset that only fits one slot.

The hero placement deserves one technical note, because it is where founders most often ship something broken. Browsers block autoplay with sound unless the visitor has already interacted with the page, and the exact rules differ per browser. The autoplay guide in the MDN Web Docs spells out what will and will not start on its own. In practice: mute it, loop it, keep it under twenty seconds, and put the narrated cut behind a button. What the page says around that loop matters just as much, and the rewrite is covered in landing page copy for founders.

The 90-second script skeleton

Write the script before you touch a recorder. Every expensive problem in amateur video production, the rambling, the retakes, the forty-second setup before anything happens, is a symptom of recording without one. The structure below is five beats with fixed timestamps. Fill in the blanks, read it aloud once, and cut whatever made you run long.

Two rules govern the fill-in lines. Write the way you talk, including the contractions, because a script that reads well silently often sounds like a press release out loud. And keep every blank concrete: a named job, a real number from your own product, an actual unit of time. Vague blanks are how a ninety-second script becomes a three-minute recording.

Notice what the skeleton has no room for. There is no logo animation, no founder introduction, no feature list, no pricing, and no tour of the settings screen. Each of those is a thing founders add because a competitor's video has one, and each one costs seconds from the only part that persuades anybody.

How do you record a demo that looks professional?

Professional, in this context, means a tidy screen, a slow cursor, clean audio, and no dead air. Every one of those is free. The gear question is almost irrelevant next to the preparation question, which is why an afternoon is genuinely enough.

• Build a demo account with believable data. This is the single biggest tell. A screen full of test, asdf, and Untitled Project undermines the product more than any production value rescues it. Use real-looking names, plausible numbers, and content that matches the industry you are selling into.
• Clean the browser. Hide the bookmarks bar, close every other tab, turn off notifications, and use a fresh profile so no personal account leaks into frame. Record a window at 1280 by 720 rather than your full display, and zoom the page in a step so the text is readable on a phone.
• Record the audio separately. Voice over first or last, never simultaneously. A stumble then costs you one line instead of the entire take, and you stop narrating at the speed of your own clicking, which is the most common reason demo voiceovers sound anxious.
• Record in short takes, one per beat. Five takes of fifteen seconds is a far easier afternoon than eleven attempts at a perfect ninety-second run, and the cuts between beats are invisible once a continuous voiceover runs across them.
• Move the cursor deliberately. Slow down, pause before each click, and never move while you are explaining. Frantic mouse movement is the thing that makes an otherwise good demo feel amateur, and it costs nothing to fix.
• Cut every second where nothing happens. Loading spinners, page transitions, typing a long input. Speed those up or cut them out. A ninety-second demo is usually a three-minute recording with the boring half removed.
• Add captions. A large share of views happen with the sound off, on phones, in public. Auto-generated captions, corrected for your product names, take ten minutes and materially change how much of the video gets consumed.

The magic-moment rule

Put your best ten seconds as early as the story allows, because the viewers you lose, you lose at the start. Attention on a demo drains fastest in the opening, which means the beginning is the only stretch you can be confident anyone watched. Anything placed there is worth several times the same footage placed at the sixty-second mark.

The magic moment is the specific instant where your product does the thing nothing else does. Not the signup, not the dashboard, not the integrations. The moment. A file becomes a formatted report. A paragraph becomes a working page. Two hours of manual reconciliation resolve in a click. If you cannot name that instant in one sentence, the video is not the problem, and the fix lives upstream in the first-100-users guide, where the same clarity gap shows up as a distribution problem.

Two practical consequences follow. First, delete the intro animation rather than shortening it, because a three-second wordmark is still three seconds of your most valuable footage spent on decoration nobody came for. Second, if your product needs setup before the magic moment can happen, do the setup off camera and open on the result. Nobody has ever chosen a product because the onboarding was faithfully depicted.

Where to host it, and what to measure

Host the narrated master wherever you get analytics, and self-host the silent hero loop so you control the poster frame and the load time. Those two decisions are separate, and treating them as one is why so many landing pages embed a full player where a fifteen-second muted loop belongs.

Whatever you pick, the page around the player is what search engines and answer engines actually read. Google's own video best practices for Search are blunt about the requirements: a stable page for each video, a descriptive title and description in the visible HTML, a real thumbnail, and structured data. That structured data is the VideoObject type on schema.org, and filling in name, description, thumbnailUrl and uploadDate is a ten-minute job most startup pages skip entirely.

Publish the transcript underneath the player as well. It costs one copy and paste from your caption file, and it turns ninety seconds of speech into several hundred words of relevant page copy. That matters more than it used to, because the engines that increasingly sit between your product and a searcher lean on text they can parse and attribute. A video with no transcript is a blank space to every one of them.

Measure watch-through, not views. The number that tells you something is the share of viewers still watching at twenty-five, fifty and one hundred percent. A steep drop before ten seconds means your opening beat is wrong. A drop in the middle means the workflow is too long. A high completion rate with no signups means the video works and the instruction at the end does not.

One demo, five surfaces

The master cut is raw material for four more assets, each of which takes minutes rather than another afternoon. This is the part that turns a one-day project into something that keeps paying, and it is most valuable during a launch, when you need a different asset every day for a week.

• The silent hero loop. Ten to twenty seconds of the magic moment, no audio, no captions, looping cleanly. This is the only version that belongs above the fold.
• The vertical clip. The same moment cropped to nine by sixteen with burned-in captions, for short-form feeds and stories. Fifteen seconds is plenty.
• The animated GIF. Five seconds, small, for your README, your changelog entries, and direct messages where a video link would not get clicked.
• Stills for the gallery. Three or four frames from the recording, exported at full resolution, cover every directory listing that wants screenshots.
• The transcript. Page copy, launch-post body text, and the outline of your onboarding email, all already written by you, out loud, in your own voice.

Sequencing all of that across a launch week is its own exercise, and the day-by-day version lives in the launch week playbook. The short version: the hero loop goes up before launch day, the master cut anchors the listing and the first comment, and the vertical clip and the GIF fill the days after, when most founders have run out of things to post.

Get it torn apart before you ship it

Show the cut to three founders before it goes on your landing page, and ask them one question: at what second did you stop caring? That is the only feedback that improves a demo, and it is the one thing you structurally cannot generate yourself, because you know what the product does and cannot un-know it while watching.

The usual advice here is to post it somewhere and hope. The problem with hoping is that watching someone else's ninety-second demo carefully enough to timestamp your own boredom is real work, and there is no reason a stranger should do it for free. That asymmetry is why Favors.dev exists. It is a founder marketing co-op with a points economy: you earn points by doing verified marketing favors for other founders, including detailed feedback on assets exactly like this one, and you spend those points asking for the same back. The feedback is graded and attached to a real profile, so what comes back is a timestamped critique rather than a thumbs up.

The same loop covers the launch itself. Founders on the launch calendar are shipping the same week you are, which makes them the readiest audience your demo will ever have and the only one that will tell you the truth about it.

Frequently asked questions