Method, evidence and limits
AI user research: methods and benchmarks
User insights in minutes. Sunzu simulates human behaviour with AI so product, design and marketing teams can test ideas, designs and ads. This guide explains the evidence, methods and benchmarks behind the research.
What is Sunzu?
Sunzu is AI user research software for product, design and marketing teams. It simulates human behaviour using AI-powered synthetic humans, helping teams test ideas, designs and ads and get user insights in minutes.
Shape products people love. Test ideas, designs and ads 500× faster than human research. Cut research costs. Let every team test and learn. Teams can use the web app or a compatible AI assistant to run research.
What are synthetic humans and audiences?
Sunzu’s synthetic humans are AI simulations of human behaviour. An audience is a model of a customer group. Individual personas simulate how members of that group might respond to a research question, concept, design or advert. Their responses are AI-generated simulations, not observations from recruited participants.
Sunzu grounds each audience in a company’s customer evidence and enriches it with a behavioural corpus. The aim is to produce findings that teams can trace and review, then use to sharpen the questions they take to real people.
What can teams test with Sunzu?
Sunzu’s web app supports four research workflows:
- Depth interviews to explore customer needs, motivations and objections.
- Concept testing to compare ideas and propositions before building.
- Usability testing to identify likely friction in designs and user journeys.
- Ad testing to explore how an audience responds to creative and messaging.
These workflows help teams reduce research costs and test more directions before committing time and budget. Explore workflows for your team.
What data grounds the research?
Customer evidence can include interview transcripts, surveys, support themes, CRM data, product analytics and previous research. That evidence anchors the audience; Sunzu’s behavioural corpus provides additional context. Broader evidence can suggest what to investigate when customer data is sparse, while gaps in support must remain visible.
The corpus described in the methodology paper contains 1M+ profiles, 34M+ attributes and 57M+ data points, according to Sunzu’s internal research. These describe the evidence base, not the number of participants in any benchmark. The underlying proprietary data is not publicly disclosed.
How are the models validated?
Sunzu’s composable AI combines 20+ layers, including algorithms and small models. The methodology describes data enrichment and behavioural modelling. Coherence and alignment checks assess personas before use; internal back-testing compares findings with human research. Source references, confidence ratings and explicit evidence gaps support review. These checks and references do not, by themselves, establish predictive accuracy.
What do the 96% and 500× claims mean?
The Sunzu homepage reports “96% of the insights humans can uncover” and testing ideas, designs and ads “500× faster than human research”. The first claim concerns insight coverage; the second concerns research speed.
Both are Sunzu-reported figures from its own research on proprietary data. The underlying datasets and detailed evaluation protocols are not published. The study benchmarks below are separate comparisons against defined human references. They should not be combined into an overall accuracy or average speed score, or treated as a guarantee for every study.
Internal back-testing
How does Sunzu compare with human findings?
Each percentage is relative to a human reference set to 100% for that study. The results measure themes covered, insights found or defects detected. They are separate comparisons, not an overall accuracy score.
| Study and measure | Sunzu in minutes | Human expert with AI in days | Human reference in months |
|---|---|---|---|
| Depth interviewsThemes covered | 76% | 54% | 100%Human insights, retested |
| Concept testingInsights found | 85% | 58% | 100%Research, 500+ humans |
| UX testingDefects detected | 75% | 34% | 100%Research, 500+ humans |
| Ad testingInsights found | 69% | 55% | 100%Research, 500+ humans |
What does “Human insights, retested” mean?
For depth interviews, the reference is the insights retrievable from the same humans when retested within a two-week interval. Sunzu covers 76% of the reference themes; a human expert with AI covers 54%.
Concept, UX and ad testing use research with 500+ humans as their reference. The time labels describe the compared research workflows; the two-week interval describes the depth-interview retest.
How should I interpret the sample-size comparisons?
- Depth interviews: equivalent to sample size of 10+ humans.
- Concept testing: equivalent to sample size of 30+ humans.
- UX testing: equivalent to sample size of 20+ humans.
- Ad testing: equivalent to sample size of 20+ humans.
These are Sunzu’s reported, study-specific equivalences. They do not mean audience characters are recruited humans, or that the same result is guaranteed for another audience or study.
Source: Sunzu product benchmarks. The methodology paper explains the approach and its limits. These figures come from Sunzu’s own research on proprietary data. The underlying datasets and detailed evaluation protocols are not published. See what the benchmarks establish for coverage, sample equivalence and timing definitions.
Does Sunzu replace research with real people?
No. Audience research supports a faster first pass: explore directions, identify likely friction and focus the next human study. Lived emotion, cultural nuance and unexpected behaviour still require real people. Teams should review the evidence and use human research to validate consequential decisions.
The methodology paper describes these boundaries alongside the engineering and published literature.
Can I use Sunzu through MCP?
Sunzu supports depth interviews, concept tests and UX tests through the Model Context Protocol, using Claude, ChatGPT or another compatible MCP client. The web app supports all four workflows, including ad testing.
Who provides Sunzu?
Sunzu is a service and brand of Sunzu, Inc., a Delaware corporation. Sunzu, Inc. serves customers outside the UK, Switzerland and EEA. Its wholly owned subsidiary, Alpha Base OS Ltd, serves the UK, Switzerland and EEA. Company registration and contact details are available on the legal entity page.