How do you test a website or prototype before it launches?
A worked example from a real run on a landing page, with the full sample report to explore.
Give simulated visitor personas one specific task on the page and have them read it as a visitor would, then record where they misread a claim, hesitate or lose trust. The report ranks the friction so you know what to fix first. Confirm the most serious problems with real users before you act on them.
A page can read well to the team that wrote it and still confuse a first-time visitor. Website evaluation with simulated personas checks a page or prototype against a defined task before real traffic arrives.
When does website evaluation fit?
Use it for landing pages, product sites and prototypes that are about to change, where you want to know whether the value and the next step are clear. It helps most when a page serves several audiences, because each persona reports on the same page separately.
What does a sample website evaluation report look like?
The example below is a real run on the landing page of an invented company, Northline Automation. Four simulated personas, executives and senior operations and engineering leads, each read the page top to bottom across eight observed steps and answered questions twice: what the company does, what they pictured when they read the headline, whether the page is for a company like theirs, and whether they would book the free assessment.
Northline landing page: comprehension and the word “robot”
Does the landing page make automation buyers understand the company as an end-to-end integration partner, or does its “robot” language mislead them?
Proceed to a controlled pilot with explicit success measures and risk controls.
- Personas4
- Runs each2
- Simulations8
- Page steps observed8
- Task completedYes
Northline is a Canadian robotics systems integrator for warehouses and factories. It does not appear to manufacture one particular robot. It assesses an operation, selects equipment from partners, installs it, connects the software, trains employees and provides ongoing maintenance and support.
Not immediately. A free on-site assessment lasting two to three days could be valuable, and the initial ROI view is attractive. However, that is a meaningful time commitment and the page gives me no customer results, performance metrics, project examples or price guidance.
A simulated example for a fictional company, run on synthetic personas from a stated catalogue. It estimates how a defined audience might respond to the material shown; it does not measure real people, and company names are invented.
The overall signal is strong, 72.1 out of 100, and the task itself succeeded: task completion was 99.6 and navigation 80. The weak spot was presentation, at 57.5. The personas understood that the company was an integrator rather than a robot manufacturer, but “See the robots” set up an expectation of photos, specifications and demonstrations that the page did not deliver. In the booking answers quoted in the report, the personas wanted customer results, performance figures and price guidance before committing two to three days of staff time.
One bar needs explaining. Latency tolerance sits at a neutral 50 because no page timing was supplied, and the report says plainly that it cannot be judged. A good evaluation does not invent evidence it was not given.
How do you read the report?
Every persona-research report follows the same order. These six steps walk through the example above.
Start with the decision brief
The ring is the overall signal: 72.1 out of 100, labelled “Strong signal”. It combines the dimension scores with equal weight, so read it as a direction, not a verdict. Beside it sits the line the evidence supports acting on.
In the report: Decision briefFind the dimension pulling the score
Task completion is the strongest dimension at 99.6; Latency tolerance is the weakest at 50.0. An average can hide one weak dimension, and that is usually where the next round of work belongs.
In the report: Performance across dimensionsFollow the task flow
The report replays each observed step on the page and marks where the personas hesitated, misread a claim or lost trust, so you can see which part of the page to fix first.
In the report: Prototype task flow and frictionRead findings with their prevalence
Each finding states how many of the 8 simulations it appeared in, next to a quoted simulated answer. A point raised in 8 of 8 runs carries more weight than one raised once.
In the report: Reported findings, prevalence and simulated voiceCheck who was simulated and why
The panel is 4 personas drawn from a catalogue with stated filters, each answering 2 times. The report lists every profile and why it was selected, so you can judge whether the right people were in the room.
In the report: Who was simulated—and whyFinish with the limits and the next step
Every report ends with what the evidence cannot show and the real-world checks to run next. The simulated answers are hypotheses to test with real people, not findings about them.
In the report: Limitations and appropriate use · Where to go next
Can you explore the whole report yourself?
Yes. The frame below holds the complete sample report, including the observed page steps and every persona’s answers.
What would you do after a result like this?
The report ranks fixes in plain language. For this page they were to:
- show real equipment in operation, with capabilities and suitable applications
- add case studies from comparable customers, with outcomes and payback
- say exactly what the free assessment includes, who takes part and what the visitor receives
- expand terms such as “AMRs” and turn “orchestration software” into plain business language
Fix the page, run the same evaluation again to see whether presentation moves, and then check the changes with real visitors.
What can a simulated website evaluation not tell you?
It cannot prove that a live page is broken, measure real conversion, or show how real people behave on slow connections or small screens. Usability testing with real participants is built on observing realistic users doing realistic tasks, and Nielsen Norman Group recommends about five participants for a typical qualitative study (usability testing 101); five users find roughly 85% of problems on its model (Nielsen). Its guidance on AI-generated users is that they should supplement real research, never replace it (Nielsen Norman Group). Use the evaluation to arrive at those sessions with a short list of problems.
A short sequence
- Write one task and the audience for each run.
- Agree the success criteria before you run it.
- Read the friction points by severity and by who they affect.
- Fix the top problems and run the evaluation again.
- Confirm the critical fixes with real visitors before and after launch.
Frequently asked questions
Can simulated visitors tell me that a live site has a bug?
No. They report how a page reads and where it creates friction. A defect needs an observed session, a log or a reproducible test.
What task should I give the personas?
One clear task per run, such as understanding what the company does or deciding whether to book a call. The example on this page asked personas to read the page, say what the company does, and say whether they would book. Vague goals produce vague findings.
Does this replace a usability test?
No. It helps you find likely problems before a usability test, so the sessions with real people spend their time on the hardest parts of the page.
Sources
Sources for the facts on this page, last checked October 17, 2026.
- Nielsen Norman Group, Usability (User) Testing 101 checked October 17, 2026
- Nielsen Norman Group, Why You Only Need to Test with 5 Users checked October 17, 2026
- Nielsen Norman Group, Synthetic Users: If, When, and How to Use AI-Generated “Research” (2024) checked October 17, 2026
Run research end to end. Keep the knowledge working.
One AI-native operating system for market research and insight professionals. Study design, evidence generation, agents, institutional knowledge, delivery and action, all in one place.
