Usability testing with AI personas
See what's confusing people in your product, today.
Luupo sends AI personas through a task on your product, each with its own identity, habits and tolerances. You find out where your product breaks down and what the persona was thinking. Then change it and run again.
replay on the left, transcript on the right, current line highlighted
Paste a link to any page we can reach: a live site, a web app, a Claude artifact or a Figma prototype.
Who it's for
For product designers
Test a flow at any fidelity. A Figma prototype, a Figma Make app, a staging build. Discover what tripped people up, change it, and run it again.
For developers
Your tests prove the code works. They don't prove anyone can use it. Point Luupo at a branch preview. No SDK, no instrumentation, no test IDs to maintain.
For startup founders
No researcher, no design team, no budget for either. Find out where your product loses people before it costs you a customer.
How a run works
paste a link, describe the task, read the findings
Paste your link
Point Luupo at any page we can reach. We check the link first and say straight away if we can't test it.
Describe the task
Set the scenario, then say what the user is trying to do and define your demographics. Luupo writes the test and casts the personas.
Luupo walks it
Each run happens in a real browser. You can watch it back and read what the persona was thinking at every step.
On realism
Why our personas behave badly
Everyone else
Most AI testing tools role-play an ideal user. Infinite patience, perfect reading, and a habit of clicking the right button first time. They finish the task, report success.
Luupo
We build them to be human. Each AI persona is cast for the role with its own identity, habits and tolerances, and is fallible by design.
Straight through. Nothing discovered.
Back a step, tried again, stopped. Each mark is a finding.
What comes back
Every run returns findings ordered by severity.
Illustrative example. Not real session data.
quote highlighted
The full transcript
What each persona said, word for word.
suggested change beneath
A suggested change for each issue
So you finish reading the report knowing what to try next.
Be one of the first to try it
Leave your email and we'll be in touch. Tell us what you're building and we'll choose the best matches for early access.
Thanks, you're on the list.
We'll email when there's something worth showing you.
Register someone elseQuestions we get asked
Does this replace testing with real users?+
No. The personas are synthetic, and nothing about them models who your real users are. What Luupo does is catch the problems you should not be spending a real participant's hour on, so the sessions you do run can go deeper.
What can't you test?+
Products behind SSO, behind a VPN, or behind bot protection. Where a page sits behind a bot wall we detect it and tell you. That detection has its own limit: it only recognises walls that announce themselves in a form we know.
How do I know a finding is real?+
You check the evidence. Anything a run claims, it can show you: the persona's own words, quoted from the transcript. We publish no accuracy figures. If the evidence does not convince you, throw the finding out.
Can a coding agent use this?+
Not directly yet. There is no API or MCP server, so a person still has to pass the findings across. They come back ranked and specific enough for an agent to act on once they do.
What does a run cost?+
Pricing is not public yet. Register your interest and we will tell you before we ask you for anything.