All results.
How often people identify AI designs, and which designs they prefer.
- Period
- Since Oct 12, 2026
- Last refreshed
- Oct 15, 2026, 07:55 UTC
AI identification rate
Rounds in which the AI design was correctly identified compared to the real product.
AI design preference rate
Rounds in which the AI design was preferred to the real product.
Identifying AI design.
How often people identified the AI design correctly.
Everyone
- Identified 58%
- Missed 42%
By role
RoleIdentified AI
Design58%
Engineering59%
Product60%
Marketing53%
Founder57%
Other57%
By model
ModelIdentified AI
Claude Opus 5.5 (medium)59%
GPT-6.1 Sol (medium)52%
Models are compared on the pairs every model has answers for.
Design preference.
Which design people preferred.
Everyone
- Real product 33%
- About the same 10%
- AI design 56%
By role
RoleReal productAbout the sameAI design
Design34%12%54%
Engineering31%9%60%
Product34%10%56%
Marketing33%14%53%
Founder32%9%59%
Other35%11%54%
By model
ModelReal productAbout the sameAI design
Claude Opus 5.5 (medium)32%11%57%
GPT-6.1 Sol (medium)33%11%57%
Preference & identification.
How identification accuracy varies with design preference.
PreferenceIdentified AI
Preferred real product73%
About the same63%
Preferred AI design49%
Confidence & accuracy.
How often were confident answers correct?
Share of rounds
- Not sure 28%
- Pretty sure 56%
- Certain 16%
Accuracy by confidence
ConfidenceIdentified AI
Not sure52%
Pretty sure60%
Certain63%
Product familiarity.
Do people identify AI designs more often when they use the product?
Share of rounds
- Uses the product <1%
- Not reported >99%
Accuracy by familiarity
FamiliarityIdentified AI
Uses the product58%
Not reported58%
Reported cues.
What gave AI design away versus what made it hard to tell?
AI design identified
- Layout & spacing45%
- Component styling27%
- Missing or extra UI19%
- Gut feeling18%
- Icons14%
- Alignment14%
- Color14%
- Typography11%
AI design missed
- Layout & spacing43%
- Component styling18%
- Missing or extra UI17%
- Gut feeling13%
- Alignment13%
- Icons9%
- Color9%
- Too polished8%
Note
Only the first submitted response per round is counted. Roles are self-reported; rounds without a role are included in the overall results. How results are counted