Usability testing with synthetic users
Synthetic users test your site and app, show you where people get stuck, and hand you the fix.
Any public page works. For apps behind a login, you verify your domain and add a test account.
Each one says what they expect before they click. We record every session: video, screenshots and their reasoning.
Not a 60-item audit. The problems that cost you most, the smallest change for each, and a prompt your coding agent can run.
What you get
Every finding links to the moment in the video where it happened, so you can judge it yourself.
What people think your site is for in the first five seconds, and what they would click first.
Ranked by how many sessions they blocked and how badly. Each with video, screenshots and quotes.
Everything else we saw, sorted by impact and effort, so the backlog stays honest.
A quick picture of the smallest fix, so design and development agree on what changes.
Markdown ready for Claude Code, Codex, Cursor or a human developer: evidence, change, acceptance checks.
A short guide to confirm the findings with three to five real users before you commit to bigger work.
Plain Markdown. Paste it into your coding agent or a ticket. It says what to change, why, how to check it, and what to leave alone.
Ship the fix and run the same profiles and tasks again. The history shows which problems are gone, which remain and what the change broke.
# Fix: "Continue to payment" is below the fold on mobile
## Evidence (synthetic sessions)
- 4 of 6 mobile sessions stopped after the shipping form.
- Session 3, 00:41-01:12: "I filled everything in... now what?"
## Smallest change
1. Move the primary "Continue to payment" button directly under the last shipping field (today it sits after the upsell carousel).
2. Keep one primary button per step. Turn "Keep shopping" into a text link.
## Acceptance checks
- At 390x844 the button is visible without scrolling once "Postcode" is filled.
- Keyboard: Tab from "Postcode" lands on the button.
## Do not touch
Prices, legal notices, tracking scripts.Human variability
Real visitors squint at small grey text, miss buttons with their thumb and leave when they are in a rush. We simulate that on purpose.
Sees blurred, low-contrast screenshots. Pale text and tiny icons stop being readable.
Taps land a few pixels off. Small, crowded targets get missed or hit the wrong thing.
Skims, skips paragraphs and gives up after a few wrong turns.
Unsure about icons without labels, hidden menus and jargon.
Arrives from a search result knowing nothing about you or your product.
Expected: “Prices will be under Bikes.”
What happened: 5 of 6 profiles opened Bikes, scrolled and left. Prices live in the footer link “Rates”.
Smallest fix: show “from €12/day” on each bike card and rename “Rates” to “Prices”.
Pricing
$0 free
See what the method finds on your own site before you pay anything.
$149 one-time
One site, one price. No subscription.
$199 per month
For teams shipping every week.
No. They are AI agents that browse your site with a goal, a pace and simulated abilities. They are good at finding the obvious problems quickly. Treat their findings as hypotheses backed by video, and confirm the important ones with real people.
On public sites the users only read and navigate. They do not submit forms that create orders, accounts or messages, and they never submit payments. Testing inside an app requires a verified domain and test accounts you provide.
Any of them. It is plain Markdown: Claude Code, Codex, Cursor, or a developer reading it as a ticket.
Only if you own it or have permission from the owner. Agencies test client sites with the client’s authorization.
No. One site, one price, 90 days of access to the report. Nothing renews.
In the EU, in Madrid. Reports are kept for 90 days and you can ask us to delete them sooner.
Synthetic users are not your customers. They are quick, cheap and tireless, and they find the problems a first-time visitor would hit. They do not know your market, your prices or your competitors the way people do. Every report includes a short script so you can check the findings with three to five real people.
Paste your URL. In a few minutes you get the 3 worst problems and one session video.