How Mirror Works
Mirror is an AI-powered usability testing platform. You provide a URL and a set of tasks, and Mirror sends AI personas to navigate your live site in real browsers. Each persona thinks aloud, flags UX issues, and scores your experience. You get a full insight report in minutes instead of weeks.
The 4-Step Flow
1. Paste Your URL
Enter the URL of the site or page you want to test. Mirror will use this as the starting point for every persona session.
2. Define Tasks
Describe what you want tested. Tasks are written in plain language, the same way you would brief a real participant.
Examples:
- "Find the pricing page and compare the Pro and Enterprise plans."
- "Add a blue t-shirt in size Medium to your cart and begin checkout."
- "Sign up for a free trial using an email address."
You can define one task or several. Each persona will attempt every task independently.
3. Pick Personas
Choose from a library of 20+ pre-built personas, or describe a custom one in your own words and let Mirror generate it.
Pre-built personas cover a range of demographics, technical skill levels, accessibility needs, and behavioral tendencies. Each persona has a distinct profile that shapes how it navigates: a first-time smartphone user will behave very differently from a power user with a screen reader.
4. Get Results
Once the study finishes, you receive:
- UX scores for each persona and an overall score for the site.
- Issues list with severity ratings (critical, major, minor, enhancement) and recommendations.
- Session replays showing every step each persona took, with think-aloud narration.
- Heatmaps showing where personas clicked and scrolled on each page.
- Downloadable reports in PDF and Markdown formats.
What Happens Under the Hood
When you launch a study, Mirror runs through three phases automatically.
Navigation
Mirror opens a real headless browser for each persona. These are full Chromium instances, not simulated or mocked environments. Each persona navigates your live site exactly as a real user would: clicking links, filling out forms, scrolling, and reading content.
At every step, the system captures a screenshot and the page's accessibility tree. An AI model (Claude) looks at the screenshot through the persona's eyes, generates a think-aloud narration ("I'm looking for the sign-up button but I only see a login link..."), decides what action to take next, and flags any UX issues it encounters along the way.
All personas run in parallel, so a five-persona study takes roughly the same time as a single-persona study.
Analysis
After navigation completes, a deeper analysis pass examines every screenshot in detail. This second pass catches issues that real-time navigation might miss, such as visual hierarchy problems, inconsistent styling, or accessibility gaps.
The system then synthesizes findings across all personas to identify patterns: issues that tripped up everyone, problems unique to certain user types, and comparative differences in how personas experienced the same flows.
Report Generation
All findings are compiled into a structured report with an executive summary, ranked issues, actionable recommendations, and supporting evidence from session data.
Study Lifecycle
Every study moves through four stages:
| Stage | What happens | Duration |
|---|---|---|
| Setup | You configure the URL, tasks, and personas. | Manual (you control this) |
| Running | Personas navigate your site in parallel. Real-time progress is streamed to your dashboard. | 1-3 minutes typically |
| Analyzing | Screenshots are analyzed in depth. Cross-persona insights are synthesized. | 1-2 minutes typically |
| Complete | Scores, issues, replays, heatmaps, and the full report are available. | Instant once analysis finishes |
You can watch the study in real time during the Running phase. Each persona's current step, think-aloud narration, and emotional state are streamed live to the progress view.
What You Get
UX Scores
A numeric usability score (0-100) for the overall site and per persona. Scores are based on task completion, navigation efficiency, issue severity, and accessibility.
Issues
Every issue includes:
| Field | Description |
|---|---|
| Description | What the problem is and where it occurs |
| Severity | Critical, Major, Minor, or Enhancement |
| Heuristic | Which usability principle is violated (based on Nielsen's heuristics) |
| WCAG criterion | Relevant accessibility guideline, if applicable |
| Recommendation | A specific, actionable fix |
| Page URL | Exact page where the issue was found |
Session Replays
Step-by-step playback of each persona's session. Every step shows the screenshot the persona saw, the think-aloud narration explaining their thought process, the action they took, and their confidence and emotional state.
Heatmaps
Aggregate click and scroll data across all personas, overlaid on each page. Heatmaps reveal where attention concentrates and where users miss important elements.
Reports
Export the full analysis as a PDF or Markdown document. Reports include the executive summary, all issues ranked by severity, recommendations prioritized by impact and effort, and evidence from session data.
Mirror vs. Traditional Usability Testing
| Traditional testing | Mirror | |
|---|---|---|
| Cost | $12,000+ per study | A fraction of the cost |
| Time to results | 2-4 weeks (recruiting, scheduling, sessions, analysis) | Under 5 minutes |
| Participants | 5-8 typical (limited by budget) | As many personas as you need |
| Coverage | Limited to recruited demographics | Broad demographic and behavioral range |
| Consistency | Varies by moderator and participant | Reproducible and consistent |
| Availability | Business hours, scheduling required | On demand, any time |
| Session recordings | Video files requiring manual review | Structured step-by-step replays with annotated narration |
| Analysis | Manual synthesis by researchers | Automated cross-persona synthesis |
Mirror is not a replacement for talking to real humans. It is a way to catch 80% of usability issues before you invest in a full research study, and to test iteratively as you ship changes.