# What % of landing pages use a product screenshot vs video vs abstract art in the hero?

Source: Lazyweb Research
Published: 2026-07-07
Sample size: n=488
Tags: landing-page, experiments, design, web, ux-patterns
HTML: https://www.lazyweb.com/research/can-i-measure-hero-media-mix-from-this-data
Markdown: https://www.lazyweb.com/research/can-i-measure-hero-media-mix-from-this-data.md

**Answer.** This is not computable as a per-page census: the web landing corpus (1,238 pages) and the mobile screenshot corpus have no working join, so hero media type per landing page cannot be extracted.[1] What the data does support is media incidence within tested experiments — video/cinematic in 68 of 488 hero experiments (14%) and product screenshots in 42 (9%).[2] Those are change descriptions, not a media-mix census.

> Hero media mix is not a computable census; video appears in 14% of hero tests, screenshots in 9% — July 2026.

## Finding: the honest limit

sites_screen_labels (web landing pages) and the screenshots table (mobile) do not share a working path or id join — a path match returns zero rows.[1] That means vision-derived hero anatomy (exact media type, CTA-button count, social-proof presence) cannot be computed for the 1,238 landing pages. Anyone claiming 'X% of landing pages show a screenshot hero' from this data is overreaching.

## Breakdown: what you can say instead

Media incidence within the 488 tested hero experiments (non-exclusive keyword incidence over experiment text):[2]

| Hero media theme | Count | Share of 488 |
|---|---|---|
| Video / cinematic / full-bleed | 68 | 14% |
| Product screenshot / UI mockup | 42 | 9% |

This describes what changed in tests, not the resting media mix of all landing pages. Media is a minor share of hero tests overall — headline work (57%) dominates.

## How to apply

Do not benchmark your hero media against a fabricated 'industry mix.' Use the incidence data to see that media swaps are a relatively rare test lever (14% and 9%), and prioritize headline before media accordingly. If you need a true media census, it requires a data source that joins page identity to vision JSON — which this corpus does not provide.[1]

## The numbers

| Stat | Computed from |
| --- | --- |
| No working join between web landing corpus (1,238 pages) and mobile screenshots corpus; per-page hero media not computable | smallSampleWarnings[web-mobile join] |
| Video/cinematic 68 (14%), product screenshot 42 (9%) of 488 hero experiments | hero_change_themes |

## Methodology

Universe: 1,238 web landing pages and 488 hero experiments, July 2026. Method: attempted census (blocked by missing join) plus keyword incidence on experiment text. Caveat: per-page media mix is not computable; incidence figures describe tested changes only.

## Sources & citations

- [1] Lazyweb Research analysis of 1,238 landing pages (web landing census), July 2026. sites_screen_labels and screenshots have no working path/id join; per-page hero anatomy not computable.
- [2] Lazyweb Research analysis of 488 hero experiments (detected-experiment corpus), July 2026. Keyword incidence over summary+learning text; describes tested changes, not a media census.

## Related questions

- [Should I test the hero headline or the hero media first?](https://www.lazyweb.com/research/should-hero-headline-or-media-be-tested-first)
- [How many CTAs should I put in the hero?](https://www.lazyweb.com/research/how-many-cta-buttons-should-the-hero-have)
- [How big is the landing-page benchmark I'm reading these stats against?](https://www.lazyweb.com/research/how-many-landing-pages-and-companies-in-benchmark)