Where do web products concentrate their experiments, and what does that mean for my roadmap?
- By Ali Abouelatta
- n = 365
- Published
- Updated
- 2 min read
- experiments · web · landing-page · pricing · design
Short answer
Across 880 web experiment annotations, testing concentrates on the hero (272), navigation (144), CTA (91), value props (82), and announcement bar (63), with pricing tables (47) and forms (37) close behind source[1]. Notably, web is far more heavily annotated per experiment than mobile — 2.4 annotations per web experiment vs 0.29 per mobile source[2]. Web testing is a messaging-and-structure game, not a single paywall button.
Web experiments carry 2.4 area annotations each vs 0.29 on mobile — Lazyweb Research, July 2026.
Bar chart · Annotations
Annotations — Finding: hero, nav, and pricing structure dominate web tests
One block is about 9.7 Annotations
- HERO
- 272
- NAV
- 144
- CTA
- 91
- VALUE PROPS / FEATURES
- 82
- ANNOUNCEMENT BAR
- 63
- PRICING TABLE
- 47
- FORM
- 37
- SOCIAL PROOF
- 34
- SECONDARY ACTIONS
- 33
- LEGAL/TRUST
- 19
Finding: hero, nav, and pricing structure dominate web tests
Web experiment annotations concentrate on the top of the page and its structure:
| Web area | Annotations |
|---|---|
| HERO | 272 |
| NAV | 144 |
| CTA | 91 |
| VALUE PROPS / FEATURES | 82 |
| ANNOUNCEMENT BAR | 63 |
| PRICING TABLE | 47 |
| FORM | 37 |
| SOCIAL PROOF | 34 |
| SECONDARY ACTIONS | 33 |
| LEGAL/TRUST | 19 |
Web also gets richer annotation coverage: 880 annotations from 365 experiments (2.4 each) versus 1,280 from 4,449 mobile experiments (0.29 each) source[1]source[2].
How to apply this
For a web product, weight your test backlog toward hero messaging, navigation, and value-prop clarity first — that is where the corpus concentrates and where a category leader like Calendly reframed its own positioning ('More than a scheduling link') source[3]. Pricing-table restructuring is a real but lower-volume lever. Announcement bars and forms are web-native test areas with no mobile-paywall equivalent, so borrow those ideas from web peers, not from app paywalls.
Caveats
Web experiments (n=365) are ~12x fewer than mobile; these are absolute annotation counts, not rates implying parity source[2]. All web experiments share one placeholder canonical screen, so you cannot compute where on the page each test sat source[4]. The Calendly hero example is a detected before/after change with inferred rationale, not a measured lift source[3].
The numbers
| # | Stat | Computed from |
|---|---|---|
| [1] | Web area annotations: HERO 272, NAV 144, CTA 91, VALUE PROPS 82, ANNOUNCEMENT BAR 63, PRICING TABLE 47, FORM 37 | experiment_area_mix_by_platform (web) |
| [2] | 880 web annotations from 365 experiments (2.4/exp) vs 1,280 mobile from 4,449 (0.29/exp) | annotation_platform_split + experiment_platform_split |
| [3] | Calendly hero reframed from 'Book meetings that work for you' to 'More than a scheduling link' | qualitative: web HERO Calendly reframe |
| [4] | 365 web experiments all share placeholder canonical screen id 23588 | web_experiments_not_screen_anchored |
Methodology
Universe: 365 web experiments with 880 area annotations, July 2026. Method: count annotations by area; compare annotation density across platforms. Caveat: absolute counts with thin web n; detected changes with inferred rationale, not measured lift.
Sources
- [1]
Lazyweb Research analysis of 880 web experiment annotations (web corpus, 365 experiments), July 2026
Area mix for web experiments.
- [2]
Lazyweb Research analysis of 4,814 detected experiments and 2,160 annotations (mobile + web corpora), July 2026
Web is annotated 2.4x per experiment vs 0.29 on mobile.
- [3]
Lazyweb Research analysis of web hero experiments (web corpus), July 2026
Calendly hero reframe; detected change with inferred rationale.
- [4]
Lazyweb Research analysis of 365 web detected experiments (web corpus), July 2026
All web experiments share one placeholder canonical screen; not screen-anchored.
Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.