A product video does not have one right form. It has a right form per placement. Your site hero can open on the product; paid social has to earn the next second before it earns anything else. Design every cut to work on mute with burned-in captions, master it in the aspect ratio the channel actually serves, and put proof inside the frame instead of in a caption nobody reads. Then measure the placement, not the video.
An AI product video is a video that shows what your product does, with AI handling the script, the voice, or the edit rather than a crew. The tooling question is now the easy one — dozens of generators will hand you a finished file.
The hard question is the one almost nobody asks before hitting render: where is this going to play?
A video that converts on your homepage will die in a paid social feed. A cut that stops the scroll on Reels will look unserious embedded above your pricing table.
Same product, same footage, different job.
This guide is about that job. Where a product video lives, how the first three seconds change by placement, why sound-off design is the default and not the fallback, what aspect ratio and length each channel tolerates, how to put proof inside the frame, and how to tell whether any of it worked.
For what goes inside a software walkthrough — structure, what to cut, how to script narration — read the AI product demo playbook. This post starts after that decision is made.
Where an AI product video actually lives
Five placements pay back reliably. Everything else is a variant of one of them.
| Placement | Primary job | Typical length | Aspect ratio | Sound |
|---|---|---|---|---|
| Site hero | Replace the paragraph nobody reads | 30–60 sec | 16:9 | Muted autoplay |
| Product / feature page | Prove one workflow works | 60–90 sec | 16:9 | Click to play |
| Paid social | Buy the next second | 6–15 sec | 9:16, 4:5 | Off by default |
| Email (outbound + lifecycle) | Earn a click on a still frame | 60–90 sec | 16:9 | Off until clicked |
| App store / marketplace | Convert a listing visitor to an install | 15–30 sec (iOS), 30–120 (Play) | 9:16 or 16:9 | Off by default |
The placements are not equally easy. Wistia’s 2026 State of Video Report, built on 900+ surveyed professionals plus 13 million videos and 79 million hours of viewing data, found homepage, video gallery, and product page videos draw the highest play rates of any placement.
That makes sense. Someone on your product page already chose to be there.
The same report found 81% of teams distribute video on LinkedIn and 76% on YouTube, which is where the difficulty lives — you are interrupting, not answering.
The practical move is to pick one placement, cut specifically for it, and expand only after it works. A single master file pushed to five surfaces underperforms on all five.
The first three seconds change completely by placement
There is no universal hook. There is a hook per level of viewer intent, and the levels are far apart.
On your own site, attention is already granted. Open on the product doing the thing — a cursor mid-action, a result appearing, a state changing.
No logo card, no title slide, no ambient b-roll.
In a paid feed, attention has to be bought back frame by frame. Nielsen research commissioned by Facebook found ad recall lifted 47%, brand awareness 32%, and purchase intent 44% among people who viewed for three seconds or fewer.
Read that carefully, because it cuts both ways. Impressions form almost instantly, which means whatever you put in second one is what most of your audience will ever see of you.
So the paid cut leads with the problem state, not the product. The messy spreadsheet before the clean dashboard, the eleven-tab workflow before the one screen.
For an app store listing the constraint is even tighter. Average watch time on preview videos runs only a handful of seconds across categories, so the value proposition has to be legible before anyone decides to keep watching.
A useful test: freeze your video at 0:03 and show the still to someone who has never heard of your product. If they cannot name the problem you solve, the opening is decoration.
Design for sound off, then reward sound on
Assume nobody hears the narration. Build the video so it still lands, then treat audio as an upgrade rather than a dependency.
The data on this is unambiguous. A study by Verizon Media and Publicis Media, reported by 3Play Media, found 92% of US consumers watch mobile video with the sound off at least some of the time, and 80% are more likely to watch a video to the end when captions are available.
The same research recorded a 13% lift in brand linkage, a 10% lift in ad memory quality, and an 8% lift in ad recall on captioned ads.
What that means in production terms:
- Burn captions into the frame for anything in a social feed. Player-level caption tracks are off by default in most feeds, so a sidecar file is a caption most viewers never see.
- Keep captions inside the safe area. On 9:16, the bottom ~20% is covered by platform UI and the top ~15% by the account header. Anything you place there is gone.
- Carry the argument in on-screen text, not only narration. One short line per beat, in your own type, is what a muted viewer reads.
- Use motion as punctuation. A highlight box, a zoom into the field that changed, a cursor that pauses before it clicks. These replace the emphasis your voice would have carried.
- Then make sound worth turning on. Clean narration, a light bed, and audio that syncs to the cut. The 37% who turn sound on because the captions caught them should get something better, not the same information twice.
On your own site, keep a real caption toggle as well as the burned-in text — that is the accessibility baseline, and it costs nothing once the transcript exists.
Aspect ratio per channel, and why one master cut fails
Aspect ratio is not a resize. It is a re-composition, and software demos are the hardest case because a desktop UI is natively wide.
| Surface | Ratio | Note |
|---|---|---|
| Site hero, YouTube, Google Play | 16:9 | Google Play promo videos run through YouTube, so 16:9 at 1920×1080 or better |
| Reels, TikTok, Shorts, Stories | 9:16 | Punch into one panel; never letterbox a full desktop screen |
| Feed (Meta, LinkedIn) | 4:5 or 1:1 | Taller frame buys more screen height in a scrolling feed |
| App Store previews | Device-native portrait or landscape | Apple’s app preview specifications require device-specific dimensions and up to three previews per localization |
The failure mode with software is predictable. You export a 1920×1080 demo, drop it into a 9:16 frame, and ship a postage stamp with two black bars — legible on a 27-inch monitor, unreadable on the phone where it will actually play.
The fix starts at the recording stage. Record at a browser width that keeps your key panel narrow enough to crop to vertical, hide the sidebar, and raise the app’s font size before you hit record.
Then cut vertical as its own edit: one panel, punched in, following the action. Not a shrunken copy of the wide version.
If you are choosing tools for this, the AI product video generator comparison covers which ones export multiple ratios natively versus which make you re-render per channel.
How long each placement tolerates
Length is a function of where the viewer is, and the honest range spans a factor of thirty.
Vidyard’s business video benchmarks put full-completion rates at roughly 65% for videos under 60 seconds. That number is the reason short cuts carry most placements — not because attention spans collapsed, but because a one-minute video is one someone can finish before their next interruption.
| Placement | Target length | Hard ceiling |
|---|---|---|
| Paid social | 6–15 sec | 30 sec before cost per outcome climbs |
| Site hero | 30–60 sec | 90 sec |
| Product page | 60–90 sec | 3 min |
| Email demo | 60–90 sec | 2 min |
| Evaluation walkthrough | 3–5 min | Whatever they asked for |
| App Store preview | 15–30 sec | Enforced by Apple |
| Google Play promo | 30–120 sec | Enforced by Google |
Two of these are decided for you. Apple requires each App Store preview to run 15 to 30 seconds; Google Play’s promotional content rules put the promo video between 30 and 120 seconds.
The Wistia data adds a wrinkle worth knowing: shorter videos win on engagement rate, but longer ones accumulate more total watch time and drive more clicked actions.
A 45-minute webinar is not competing with your 15-second Reel. They are different instruments.
For the placements above, short wins. Save length for people who asked for it.
Put the proof inside the frame
Most product videos keep their evidence outside the video — in the landing page copy, the logo strip, the caption. Then they wonder why the video does not carry weight on its own.
It has to travel. Once a cut is in a feed or embedded in someone else’s Slack, the page around it is gone.
Four kinds of proof that fit inside a product video:
- Real, real-shaped data. Plausible company names, realistic volumes, a record with a missing field. Acme Corp and John Doe tell the viewer they are watching a sales asset.
- A number on screen at the moment it is earned. Not a claim card at the end — the count appearing as the job finishes, the timer visible while the export runs.
- A customer’s own words, cut short. Ten seconds of a real user finishing your sentence beats sixty seconds of a testimonial arc.
- The unedited part. A load state you did not cut, a real second of latency. Perfect footage reads as a mockup, and buyers evaluating without a rep are looking for the seam.
The one thing to avoid is stacking logos over the product. It moves the eye away from the workflow at exactly the moment the viewer is deciding whether the workflow is for them.
App store and marketplace listings are their own format
If you ship a mobile app or list in a marketplace, the listing video is not a shortened version of your website video. It is a different asset with different economics.
A/B testing data from app-store optimization firms, including SplitMetrics and StoreMaven, has generally put the lift from a well-made preview video in the 20–30% range on install conversion. The result varies sharply by category.
The counterexample matters more than the average. Apple’s own product page optimization documentation describes a test where the page without an app preview video outperformed the version with one, against the developer’s hypothesis.
So treat the listing video as a hypothesis, not a given. Both stores ship native A/B tooling — Product Page Optimization on the App Store, store listing experiments on Google Play — and both are free to use.
Three format rules that hold either way. Open on the app in use rather than a splash screen, keep every caption clear of the install button and the platform chrome, and assume the video plays muted in a scrolling gallery.
Measuring whether the AI product video worked
Measure the placement, not the file. “Views” is not an outcome, and a video with a beautiful play rate can sit on a page that converts worse than the version without it.
| Placement | The number that decides it | Supporting signals |
|---|---|---|
| Site hero | Page conversion rate, video vs no video, A/B tested | Play rate, drop-off curve |
| Product page | Demo requests or sign-ups from that page | Average watch time, replays |
| Paid social | Cost per outcome | 3-second views to thruplay ratio, hook rate |
| Click-through on the thumbnail | Watch time after the click | |
| App store | Store conversion rate in native A/B tooling | Preview watch time |
Two diagnostics are worth building the habit around.
The drop-off curve tells you where the video breaks. A cliff in the first five seconds is an opening problem.
A steady slope that flattens at 60% is a length problem, and a dip in the middle is usually a section that explains rather than shows.
The hook rate — three-second views divided by impressions — isolates your opening from everything else. Change only the first three seconds, hold the rest constant, and you can attribute the difference cleanly.
One warning on the hero test. If your homepage video moves conversion by a point or two, that is inside the noise band for most traffic volumes, so run the test long enough to reach significance instead of calling it in a week.
Where AI shortens the loop
The reason placement-specific cuts were rare is that they used to cost five times as much. Five versions meant five scripts, five voice sessions, five edits.
That is the constraint AI actually removed. Script, voice, and render collapse to minutes, so cutting for the channel becomes a scheduling decision instead of a budget one.
For software, Vidwip’s AI Demo Polisher works from the recording you already have. It reads what is on screen, writes a scene-by-scene script from what it sees, and voices it.
You steer that script before the render with a written brief, a voice and a language. It can’t be edited line by line afterwards, so a wording fix means a new render, not a re-record.
A finished minute costs 60 credits, and 1 credit is $0.01, so $0.60 per minute of output. Cutting a 90-second product page video down to a 15-second paid variant and a 30-second hero costs cents, not another production cycle.
Vidwip is free to start with no credit card, which is enough to run one short placement end to end and judge the output. Free-plan renders carry a small Vidwip badge, and paid plans have no watermark.
Free accounts reach the same 190+ voices, the same 74 languages, and the same models as paid ones. What a paid plan adds is credits, badge-free exports, batches of up to 5 demos (10 on Pro) and more renders at once, 4 on Starter and 10 on Pro against 1 on Free; see the full pricing.
Two things that matter for distribution specifically. Captions render burned into the export, which is the format social feeds need, and dubbing into up to 94 languages means a localized cut for a new market is a re-voice rather than a reshoot — the mechanics of that are in the AI video dubbing guide.
What none of this decides is which workflow to record or where the cut should stop. That judgment is still yours, and it is still what separates a video that sells from a video that renders.
Start with one placement
Pick the placement where you already have traffic and no video. For most software companies that is the product page, not the homepage — it has intent, it has fewer competing elements, and it is easier to A/B test cleanly.
Cut 60 to 90 seconds showing one workflow end to end. Burn in captions, master it 16:9, and put a real number on screen at the moment it is earned.
Then cut the 15-second vertical variant from the same recording, open it on the problem state instead of the product, and put it into paid social to find out which opening earns the second second.
Measure the page, not the play count. Once the first placement pays back, the second one costs you a render.
If you want the structure of what goes inside the cut, the demo playbook covers it. If you are still choosing tools, compare AI product demo video makers for general software walkthroughs or the SaaS demo video generators if you are shipping weekly and need to re-cut often.