Usability studies with simulated participants
Watch simulated visitors get stuck on your live site, and read in their own words why.
We open your website in a real browser for hundreds of simulated people, drawn to match the customers you describe. Nobody tells them what to do. The report says where they stalled, who left, what they said about it, and what to change.
The first Quick read on a site you control is free. Participants are simulated, not real people, and we say so in every report.
The gap
Analytics says how many left. It never says why.
A funnel chart can tell you that 60% of visitors left at step three. It cannot tell you that the button to continue was below the fold on a phone, that the free page answered the question your paid one was built for, or that the people who left had already decided the product was not for them two screens earlier.
Those are the answers a moderated usability session gets you, one participant at a time, over a week. We run the same kind of observation on hundreds of simulated participants in a couple of hours, and keep only what shows up twice.
How a study runs
Four steps, and nobody is given a script
You describe your customers
In plain English: "practising physicians in the US", "small-business owners who do their own books". We resolve that against a pool of 10 million simulated personas and add a control group of people with no use for your product.
They arrive with no instructions
Each participant gets a fresh real browser on your live site and is told only that the product exists. They read, scroll, click, type into forms, or leave, as they decide.
It runs twice, on different people
Two runs, drawn separately. One run is an anecdote with a sample size; a finding has to appear in both to be reported as holding.
You get findings you can act on
Each one says the screen it happened on, how many raised it, how many left there, how that compares with the control, what to change, and how to check the change worked.
One real result
Same appetite on a phone. A fifth as many decided it was for them.
A physician salary-benchmark site, studied in September 2026. We sent 120 simulated physicians through it on a 390×844 phone, 120 on a desktop, and 120 control participants with no reason to want it. Each group ran twice, 60 different people each time.
Where it happened
Both groups reached the site's main form about as often (89 and 98 of 120). The difference was on it. The form's Next button sat at the top of each step, disabled until fields below the fold were filled. On the phone, 36 participants said some version of
no Next or Continue button visible
and all 36 ended their visit there. On the desktop, 33 raised it and 18 left.
What we told them to change
Put the Next control within thumb reach after the last field, stop the specialty picker covering the form, and make tap targets 44 pixels. How to check: run the same people again on the changed form.
How to read these numbers
They are simulated rates. They do not say what share of the site's real phone visitors leave. What held is the comparison: a 34-point gap between phone and desktop on the same product, three times the spread between the two runs on that measure.
For coding agents
Order a study from your agent
ProductTex is an MCP server. Add it once and your coding agent can register your site, describe the audience, submit a study and read the report — for example, a Quick read after every change that touches a flow people pay through.
The report comes back with the sentences that are safe to repeat as written, and a list of what the study cannot support, so the agent passes on the caveats along with the findings.
claude mcp add --transport http producttex \
https://178-156-204-0.sslip.io/mcp \
--header "Authorization: Bearer ptx_…"Then ask: "Run a ProductTex Quick read on our pricing page for small agency owners." Refusals come back as plain instructions — verify the domain, name three guesses — that the agent can carry out or forward to you.
What you can buy
Seven studies, one instrument
| Study | What it answers | Price |
|---|---|---|
| Quick read | First impressions and themes from your audience and a control, 30 each. No rates. About 40 minutes. First one free. | $49 |
| Friction study | Up to two audiences and a control, 60 each, run twice. The full report with rated findings and fixes. | $299 |
| Phone vs desktop | Your audience split by device, run twice. The study above. | $349 |
| Accessibility read beta | Low vision at 200% zoom, colour blindness, and screen-reader users beside your usual audience. Vision only; not a WCAG audit. | $249 |
| Language read | Your audience in their first language beside the same audience reading English as a second language. Where wording, not design, stops people. | $299 |
| Fix check | The same people again on your changed site: what went away, what stayed, what is new. | $149 |
| Second visit beta | The same people return, remembering their first visit. | $199 |
Read this before you buy
What a study can tell you, and what it cannot
It can tell you
- Where on your live site people stall, on which screen, next to which control.
- In their own words, why — the question they could not answer, the button they could not find.
- Whether a problem is about your audience or would stop anyone, by comparison with the control group.
- Which of two versions did better with the same people, and whether that held on a second run.
- Whether your own three guesses about where people struggle were right. You write them down before the study runs.
It cannot tell you
- Your conversion rate, or how much a change will lift it. Simulated participants click far more than real visitors; the direction of a difference carries over, the size does not.
- How fast your site is. Participants get a settled page; load time and layout shift never reach them.
- What returning customers do. Everyone is a first-time visitor, except in a Second visit study.
- What any real population thinks. We do not sell survey answers or population numbers.
- Anything about a control we cannot drive, such as a canvas or drag handle. That is reported as our blind spot, not your bug.
On ourselves first
We run it on our own work
When we measured our own website's contrast with the same audit the participants receive, it found five contrast failures, including text we had shipped that day that could not be seen. We fixed them. The rules in the method — two runs, a control group, no rates below 60 people per run — each exist because leaving one out once gave us a confident wrong answer.
Start with a Quick read on your own site
Sign up with an email, verify that you control the site with a file, a meta tag or a DNS record, and describe who your customers are. The first Quick read is free.