Know what works before you ship it.
AI users test your prototype, website, or app like real users. See where they drop off, compare potential fixes, and decide what to test with real users.
A real run on Morsel, a food-diary prototype we made for this sample. Watch the whole simulation →
Test your product with AI users.
From a first prototype or a live website to the app in your repo.
- 1It learns your product and how it makes money
- 2It reads every screen like a real user
- 3It simulates a week of real use
Not personas. People.
Each one has a life, a temperament and a breaking point — and tests your product like a real user.
set in their ways · price-sensitive
“That number looks impossible, I'm going to be starving by lunch.”
Right now — used MyFitnessPal or a similar app a while back, or winging it with less snacking.
Account creation or payment screen appears before they've logged a single meal.
Know what to fix first.
Get a prioritized fix based on the simulation. Compare the results with the same AI users before testing the change with real users.
Ask for “Create account” after “Today — first meal logged”, not before
36.7%→51.3%
logged a first meal, rerun on the same people
Run A/B tests.
The same AI users try each one and tell you which one wins.
Which version gets more people to log a first meal?
The same 300 AI users tried both — so the gap is your change. A/B tests are part of Pro.
Why not just ask your AI?
We tested them side by side.
No evidence. Just guesses.
35%right when the answer wasn’t obvious — and it sounded as sure as ever.
Evidence behind every answer.
AI users modeled on your target audience try it first. You see who got through, where the rest quit, and why.
Changes its answer when questioned.
88%of the time, one “Are you sure?” made it swap a right answer for a wrong one.
Can’t be talked out of it.
Are you sure?Version B wins
Its answer comes from what people did, not from your chat.
Has never met your users.
0people tried itIt reads your screens and imagines the rest.
Watches hundreds of them try it.
300AI users try itEach one like your users, each trying it on their own.
What makers ask first.
Is this as good as real users?
It is the cheap first pass. It finds where people quit and which fix moves the number — so the real users you pay for test the right thing.
What can I test?
Anything your assistant can read: a Figma prototype, screenshots, a live site or the app you are building.
Which AI assistants does it work with?
Claude, ChatGPT and other AI assistants that connect to tools. Add one address — no account needed.
How long, and what does it cost?
One page is free to test here: paste a link and 300 AI users try it in under two minutes — no account, no card. Testing the whole flow — sign-up, first value, payment — runs in your AI assistant on a plan: Starter is $5 a month, Pro $14. A new product takes 3–5 minutes, and the same people again after a change 1–2.
How lifelike are the AI users?
Each person has a job, a past, a temperament and dealbreakers, and thinks through every screen as they would. It is still a simulation: trust the direction and the size of a gap, then confirm the winner with real users.
Can I trust the numbers?
Every number is counted from what the simulated people did — no model writes one. Each fix is rerun on the same people, and the report says how sure it is.
What happens to my screens?
Only what the simulation needs is sent. Images are deleted after 30 days and never used to train AI models.