Most advice about product demo videos is written from opinion. We had a chance to check it against a corpus, because InstaDemo films real websites automatically — you give it a URL and it crawls the site, writes a script, records the pages and narrates them.
As of 19 August 2026 we have generated 1,026 demos across 810 distinct hosts. Below is what that corpus says about the web those demos were made from. Every number here is aggregate. No customer, URL or host is identified.
How many pages does a typical product site actually have?
Two. Across 965 demos where we recorded what the crawler found, the median number of discoverable pages was 2, and the maximum was 20.
That was the most surprising result in the set, and the most useful. Demo tooling — ours included — is largely built on the assumption that a product has a tour in it: a landing page, a features page, pricing, a dashboard, a settings screen. The median site we are asked to film has a homepage and one other page.
If you are writing a demo script and struggling to fill eight scenes, the problem may not be your script. Most products do not have eight scenes.
How long is a finished product demo?
33 seconds at the median. Across 972 completed demos with a recorded duration: median 33s, 90th percentile 46s, longest 188s.
Nine in ten finished demos come in under 46 seconds. The two-minute product walkthrough is not what people are actually making, even when nothing stops them from making one.
Are product demos still landscape?
Not always — 37.5% are vertical. Of 890 demos with a recorded output format, 556 were landscape and 334 were portrait.
More than a third of product demos are being made in a 9:16 shape. That is not a rounding error, and it is not what most demo tooling assumes. It suggests these videos are going somewhere other than a landing page hero — into feeds, into stories, into places a widescreen file cannot follow.
Where are new products actually hosted?
Roughly seven in ten sites are on their own domain. The remaining 30% cluster on a small number of platforms:
| Host | Demos | Share |
|---|---|---|
| Own domain | 723 | 70.5% |
| Vercel | 110 | 10.7% |
| Lovable | 69 | 6.7% |
| Netlify | 47 | 4.6% |
| Base44 | 27 | 2.6% |
| GitHub Pages | 13 | 1.3% |
| Railway | 9 | 0.9% |
| Replit | 9 | 0.9% |
| Render | 8 | 0.8% |
| Cloudflare Pages | 5 | 0.5% |
| Bolt | 4 | 0.4% |
| Shopify | 2 | 0.2% |
Grouping the AI app builders together — Lovable, Base44, Replit and Bolt — gives 109 sites, or 10.6% of everything we have filmed. One in ten products we are asked to demo was assembled on a platform that did not meaningfully exist three years ago.
Has .app caught up with .com?
Nearly. Counting the top-level domain of all 1,026 demo URLs:
| TLD | Count |
|---|---|
| .com | 356 |
| .app | 327 |
| .dev | 33 |
| .io | 28 |
| .net | 20 |
| .in | 19 |
| .ai | 18 |
| .br | 13 |
.com still leads, but only just, and the gap is almost entirely explained by deploy-platform subdomains. A large share of the products being built right now do not have a .com at the point where they first need a demo — they have a preview URL.
What does this mean if you are making a demo?
Three things follow from the numbers above, and none of them is what conventional advice says.
Write for two pages, not eight. If your product genuinely has one screen worth showing, a 30-second demo of that screen beats a padded tour. The corpus says short is normal, not lazy.
Decide the shape before the script. A third of these videos are vertical, and a vertical demo is not a landscape demo cropped. What fits on screen changes, which changes what a scene can show.
A preview URL is a real product. If one in ten sites being demoed is on Lovable or Base44 or Replit, then "wait until we have a proper domain" is advice that half a cohort is already ignoring.
Methodology and limits
These are all the demos in our production database on 19 August 2026, not a sample. Counts are exact; percentages are rounded to one decimal place.
Some honest caveats:
- This is our customer base, not the web. People who ask an AI tool to film their site are self-selecting. The 10.6% AI-builder share says something real about who buys automated demo tooling; it does not say 10.6% of the internet runs on Lovable.
- Format was recorded for 890 of 1,026 demos. The portrait share is of those 890, not of everything.
- Page counts are what our crawler discovered, which is a lower bound. A page behind a login, or one reachable only from a menu the crawler could not open, is not counted.
- We are not publishing failure-rate data, even though it would be interesting. Failed demos in our system get retried, and a successful retry overwrites the original error category — so the corpus has overwritten its own failure history and any number we gave you would be wrong.
If you want to check any of this against your own product, the fastest way is to point InstaDemo at your URL and see what it finds.