What Has ChatGPT Changed On Its Paywall And Signup This Year?
- By Ali Abouelatta
- n = 53
- Published
- Updated
- 3 min read
- paywall · pricing · signup · experiments · saas · monetization
Short answer
Lazyweb Research detected 53 distinct UI experiments at ChatGPT (July 2026), of which 12 were detected in calendar 2026 — one of only two companies in the 276-company corpus with a defensible "this year" volume. source[1] The signals cluster on signup (at least 18 experiments) and paywall/pricing (at least 8), and the 2026 changes rename generic plan rows into concrete capability labels and add named premium features like Codex and deep research. source[1] These are observed before/after variations with inferred rationale, not company-confirmed A/B tests.
Lazyweb Research detected 53 distinct ChatGPT experiments (July 2026), 12 of them in 2026 — the second-largest current-year experiment volume of any app tracked.
The finding
Lazyweb Research detected 53 distinct experiments at ChatGPT across the tracked window, 12 of them in calendar 2026. source[1] That puts ChatGPT second only to Finch (30 in 2026) for current-year experiment volume, so "what did ChatGPT change this year" is one of the few brand queries this corpus can answer honestly rather than reaching back to older diffs. source[1] The surface mix skews to signup (at least 18 experiments) with paywall/pricing at least 8. source[1]
What actually changed on the paywall
Two dated 2026 diffs show the direction of travel. On 2026-05-24, the plan comparison table renamed generic rows — "basic models," "smarter models" — into concrete capability labels: core model, advanced models, advanced image creation with Thinking, expanded memory. source[2] On 2026-05-28, the feature table added Codex and deep research as a Plus-exclusive row and set the CTA price language to USD 19.99. source[3] The pattern: the paid tier's differentiator moved from vague "better model" language to named, scannable capabilities.
| Detected change | Date | Inferred rationale |
|---|---|---|
| Generic model rows renamed to capability labels | 2026-05-24 | Concrete names make paid value easier to scan source[2] |
| Codex + deep research added as Plus-exclusive row; CTA reads $19.99 | 2026-05-28 | Named premium capabilities give a reason to upgrade beyond model access source[3] |
How to apply it
If your paywall leans on abstract tier language ("pro features," "advanced access"), ChatGPT's 2026 diffs are a working example of swapping abstraction for named capabilities. The move is copy-only — no price change was detected alongside the capability relabeling source[2] — which makes it a low-risk test to replicate on a mature paywall. Cross-reference against your own signup surface, since ChatGPT concentrated most of its detected 2026 activity there (at least 18 signup experiments). source[1]
Caveats
All 53 experiments are observed UI variations with LLM-inferred rationale, not company-confirmed A/B tests — no lift is measured or implied. source[1] Surface splits are lower bounds because screen category is unlabeled on 1,425 of 4,814 corpus experiments. source[4] The 12-in-2026 count is the honest current-year figure; the remaining 41 predate 2026. source[1]
The numbers
| # | Stat | Computed from |
|---|---|---|
| [1] | 53 distinct experiments (12 in 2026; at least 18 signup, at least 8 paywall) | company_total:chatgpt (value 53; splits signup 18, paywall 8, in-2026 12) |
| [2] | Capability-label rename detected 2026-05-24 | qualitative[] chatgpt 2026-05-24 entry |
| [3] | Codex + deep research row added, CTA $19.99, detected 2026-05-28 | qualitative[] chatgpt 2026-05-28 entry |
| [4] | 1,425 of 4,814 experiments have no screen category | screen_category_null_on_experiments (1425/4814) |
Methodology
Universe: 53 distinct ChatGPT experiments (COUNT(DISTINCT experiment_id)) within 4,814 detected before/after UI diffs across 276 companies, July 2026. Extraction: LLM-inferred rationale on observed screenshot variations. Caveat: these are detected variations, never company-confirmed A/B tests; no lift is measured.
Sources
- [1]
Lazyweb Research analysis of 53 detected experiments (ChatGPT, ~800-app mobile corpus), July 2026
COUNT(DISTINCT experiment_id) on before/after screenshot diffs joined to company_name; surface splits from is_paywall + screen_category.
- [2]
Lazyweb Research analysis of 53 detected experiments (ChatGPT), July 2026
Dated before/after diff, 2026-05-24; rationale is LLM-inferred, not company-confirmed.
- [3]
Lazyweb Research analysis of 53 detected experiments (ChatGPT), July 2026
Dated before/after diff, 2026-05-28; rationale is LLM-inferred, not company-confirmed.
- [4]
Lazyweb Research analysis of 4,814 detected experiments (276 companies, ~800-app mobile corpus), July 2026
screen_category is NULL on 1,425 experiments, so all surface splits are lower bounds.
Source: Lazyweb Research — proprietary analysis of real, in-market app screens. Cite as Lazyweb Research, 2026-07-07.