Evidence, not empathy.
Simulated personas are a fast, repeatable way to find where a funnel breaks for different kinds of people. They are not those people, and we will never market them as if they were.
What the research says
Simulators are too polite: in CMU’s Sim2Real study, simulated users made polite turns 49% of the time vs 15% for humans, inflating success from 63.6% to 77.8%. Lost in Simulation found older and non-native personas are the least valid.
What we claim
A persona is a set of enforced parameters (patience, reading level, dexterity, input mode…). Findings are observed behaviour. Precision per persona is published with n and confidence intervals.
Who confirms it
Paid testers from the populations we model (older adults, non-native readers, people who use assistive technology) re-run Confirmed findings. “Nothing about us without us.”
Join the Eval Army →Words we use and avoid
| We say | We avoid |
|---|---|
| Accessibility-tree navigation probe | Screen-reader simulator / simulated blind user |
| Older adult, low-dexterity persona | Elderly user |
| Non-native reader (CEFR B1) persona | Foreign user |
| Human-confirmed by a paid tester who uses assistive technology | Accessibility-certified |
Where simulation stops
W3C’s Web Accessibility Initiative is clear that tools can only assist in evaluating accessibility, not determine it. For depth, we route work to specialist human-testing partners rather than compete with them. If a persona’s precision drops below 50% for two consecutive months, we stop selling it and keep it only for internal calibration.