Do Companies Keep Hero Image and Video Swaps After Testing Them?

Across Lazyweb's 4,814-experiment corpus, 307 detected experiments (6%) swapped hero media — an image, video, illustration, or animation — spanning 122 companies.[1][2] Among classifiable changes, additions of hero media outnumber removals about 3.7-to-1 (52 added vs 14 removed).[3] Hero is also the single most-annotated change area in the corpus, at 405 experiments.[4]

307 detected experiments swapped hero media across 122 companies, and hero is the most-annotated change area at 405 experiments — Lazyweb Research, July 2026.

Lazyweb Research · n=307 · Published 2026-07-07

landing-pagedesignexperimentsux-patternsweb

The finding: the hero is the most-worked area

The hero draws more experimentation than any other page area: 405 experiments are annotated as touching the hero, ahead of offer (281), pricing (229), and CTA (225).[4] Narrowing to media swaps specifically — hero co-occurring with image/video/illustration/media/visual/animation — there are 307 such experiments across 122 companies.[1] So a hero media swap is a mainstream tactic, not an edge case.

Direction of change

Among hero-media experiments Lazyweb could classify by direction:

DirectionExperiments
Added52
Removed14
Mixed14
Changed / unclassified227

Source: direction heuristic.[3] Additions lead removals ~3.7:1 among classifiable rows, but 227 of 307 are swaps that replaced one media asset with another rather than adding or removing — which is what 'media swap' usually means in practice.

A named example

Epidemic Sound reworked its hero from 'You bring the idea. We bring the sound.' to 'Bring your stories to life. Just add sound.' with a '2.5 billion times per day' scale claim, and changed the above-fold brand case-study image grid.[5] Inferred rationale: front-loading the licensing category's two core objections — legitimacy and legal safety — before the CTA. The media change carried a positioning change with it.

Caveats

These are detected UI diffs with inferred rationale, not measured results.[6] Unlike discounts or social proof, the stat pack has no persistence/retention proxy for hero media, so this page reports prevalence and direction only — do not infer 'swaps get kept' from these numbers. The 307 count is keyword-matched ('hero' + a media word).

The numbers

StatComputed from
307 hero-media experiments across 122 companiestactic_mentions__hero_media_swaps
4,814 total detected experiments (hero media = 6%)total_detected_experiments; 307/4814
52 added / 14 removed / 14 mixed / 227 unclassifieddirection_split__hero_media
Hero 405 / offer 281 / pricing 229 / cta 225 annotated experimentsannotation_area_breakdown
Epidemic Sound reworked hero copy + case-study image grid with a '2.5 billion/day' claimqualitative: hero media swaps, epidemic-sound, 2026-07-04
0 measured outcomes (learnings are LLM-inferred)family universe note
Methodology. Universe: 4,814 detected experiments (362 companies) in Lazyweb's ~800-app mobile corpus. Hero-media tagged by keyword co-occurrence; direction by add/remove heuristic (227 of 307 unclassified). No retention proxy exists for this tactic. Detected diffs with inferred rationale, not measured results. July 2026 pull.

Sources & citations

  1. [1] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. 307 experiments mention hero + (image/video/illustration/media/visual/animation) across 122 companies.
  2. [2] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. 307 of 4,814 total detected experiments (6%).
  3. [3] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. Direction heuristic: 52 added / 14 removed / 14 mixed / 227 unclassified.
  4. [4] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. Annotated change-area breakdown: hero 405, offer 281, pricing 229, cta 225.
  5. [5] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. Epidemic Sound (2026-07-04): hero copy and case-study image grid changed.
  6. [6] Lazyweb Research analysis of 4,814 detected experiments (362 companies, mobile-app corpus), July 2026. Learnings are LLM-inferred hypotheses, never measured A/B outcomes.

Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.

Related questions

Explore the underlying screens, flows, and A/B tests inside Lazyweb. More research