Find usability issues before real users do
Simulated users from Alyosha run the tasks on your website, app or prototype before launch. You get task success, where they stall or quit, and each issue ranked by severity. Then you check the big ones with real people.
- Task success
- 52%103 of 200 sessions
- Median time on task
- 4:10target under 3:00
- Ease rating (SEQ)
- 4.2out of 7
- 1Landing200
- 2Choose plan186
- 3Create account149
- 4Delivery and payment121
- 5Confirm103
| Severity | Issue | Step | Hit, then quit | What the user said |
|---|---|---|---|---|
| Blocker | Weekly price read as monthly | 2Choose plan | 54 hit, 29 quit | “$79 a month for three meals? No way.” |
| Major | Password rules appear only after an error | 3Create account | 41 hit, 18 quit | “It keeps saying invalid. Invalid how?” |
| Major | Delivery fee first shown at payment | 4Delivery and payment | 33 hit, 15 quit | “Where did $9.99 come from?” |
| Minor | Promo code clears the card form | 4Delivery and payment | 14 hit, 3 quit | “I typed my card twice. Annoying.” |
| Minor | Google sign-in below the fold on small phones | 3Create account | 22 hit, 4 quit | “I don't want another password.” |
Severity: Blocker stops the task for most who hit it, Major for some, Minor slows people down. Most hit by first-time users on phones and users over 55.
Run it with
Methods
Task-based usability test
Each simulated user gets a task, like signing up or finding a return policy, and you see who finished, how long it took and where they quit.
Severity and frequency ranking
Every issue is rated blocker, major or minor and counted by sessions hit, so the fix list starts with what costs you the most users.
Think-aloud at every step
Each simulated user says what it expected and what confused it, so an issue comes with the reason, not just a drop in the funnel.
Post-task ease rating (SEQ)
After each task the user rates how easy it was on the 7-point Single Ease Question, which you can compare build to build.
Follow-up interviews with real users
Jean's AI interviewer asks real people about the issues that matter, so you confirm them overnight before you rebuild.
Users who act like people
Alyosha is trained on what people say, what they do and why. Its users skim, misread and give up the way people do. A prompted LLM tends to finish every task it is handed.
Enough sessions to count
A lab study watches five to eight people. Here each user type runs the task many times, so you can tell an issue that hits a third of users from one that hits a few.
Every build, then real people
No recruiting or scheduling, so you can run the same tasks on each release and compare scores. When an issue matters, check it with real users through Jean's AI interviewer.
Questions
What is a simulated user in usability testing?
An AI that uses your product as a person with a given profile would: age, device, experience, goals and patience. It clicks, reads and decides on its own, and says why it did what it did.
Can simulated users replace testing with real people?
No. Use them to catch the obvious issues early and on every build. Then test the issues that matter with real users, which Jean runs through AI-moderated interviews.
What usability metrics do I get?
Task success rate, time on task, drop-off by step and a post-task ease rating (SEQ) for each task. Each issue comes with a severity rating, the number of sessions it hit and quotes from the users who hit it.
How are usability issues rated for severity?
We use a blocker, major and minor scale, close to the common Nielsen severity ratings. A blocker stops most users who hit it, a major one stops some, and a minor one slows people down. The list is ranked by severity, then by how many sessions hit it.
What can I test?
Live websites, web apps and clickable prototypes. Tell us the tasks and the audience, and we set up the first runs with you. This product is early, so we work with each team directly.
How is this different from QA or automated testing?
A QA script checks that each button works. A simulated user checks whether a person would understand what to press, and where they would give up.