Evidence your team can act on
Trace the finding back to the behavior, artifact or number behind it
Evidence groundedUserApproved3.20
GPT2.57
Claude2.23
Published by our team as UserApproved.ai · August 2026
Better decisions
before you experiment
UserApproved outscored Claude and GPT on every quality dimension in a blind review by three independent CRO experts
See the resultsOne ecommerce store · One run per system
85/100
The highest-rated
growth analysis in this run
UserApproved85
Purpose-built agentsGPT Sol 5.668
HighClaude Opus 565.2
X-highThree independent reviewers · Scores out of 100
Fewer findings
More that matter
Three of UserApproved’s five findings were rated critical. Claude and GPT produced twelve findings combined, with one rated critical.
UserApproved60% critical
3 of 5 findings rated critical
GPT Sol 5.60% critical
0 of 5 findings rated critical
Claude Opus 514% critical
1 of 7 findings rated critical
What makes a finding
worth acting on
Each finding was rated on a four-point scale across evidence, reasoning, usefulness and the next action
UserApprovedGPT Sol 5.6Claude Opus 5Reviewer scores · 1–4
The why behind the behavior
Connect observations into an explanation that shapes the next experiment
Strategic synthesisUserApproved3.50
GPT2.93
Claude2.85
Work worth prioritizing
Focus team attention on findings that can change a growth decision
Impact and usefulnessUserApproved3.20
GPT2.63
Claude2.35
A clear next experiment
Know what to change, where to change it and what to measure
Action specificityUserApproved3.70
GPT2.76
Claude3.00
Same brief
Independent judgment
On August 4, 2026, all three systems investigated the same live US ecommerce store with the same available data, APIs, MCP integrations and browser tools. Three external CRO professionals scored the outputs without knowing which system produced them.
The brief
Act as an ecommerce growth expert and identify actionable revenue opportunities
Scoring, coverage and cost
Each finding was scored from 1 to 4 on each dimension, for a maximum of 16 points, then normalized to 100. Published scores are the means across the three reviewers.
- 4
- One minor flaw a reader would not act on differently
- 3
- Passes, with a real and citable flaw
- 2
- Several real flaws, or one that materially weakens the finding
- 1
- Barely passes; a reader would distrust the pillar
| System | Coverage | Elapsed | Quality / $ |
|---|---|---|---|
| UserApproved | Desktop + mobile · 4 persona paths | 46:52 | 3.3 |
| GPT Sol 5.6 · High | Desktop · 1 primary path | 40:00 | 3.7 |
| Claude Opus 5 · X-high | Desktop · 1 primary path | 39:00 | 2.0 |
GPT had the highest reported quality-per-dollar index. UserApproved had the highest quality score, with broader coverage and the longest elapsed time. Provider accounting can differ, so the cost index applies to this run. The store is available to prospects under NDA.
This benchmark measures CRO analysis quality in one ecommerce run. It does not measure patient enrollment lift or live experiment outcomes.
Put better intelligence
behind your growth
See how we bring patient understanding and experimentation together to turn more interest into enrollments
See the Eden storyFree setup and integration