Heuristic Evaluation & UX Audit
Overview
A UX audit is one of the most directly billable engagements in UX consulting. Without AI, a comprehensive heuristic evaluation takes 2-3 days. With AI, the same quality of output is achievable in 4-6 hours. This is not about doing less work - it is about compressing the mechanical parts (checklist generation, severity rating, report structuring) so you spend more time on the high-judgment parts: prioritisation, stakeholder framing, and recommendation quality.
AI workflow
10 steps
- 01
Context-Specific Heuristic Checklist
Claude
Do not use generic Nielsen heuristics. Prompt: "I am auditing a [product type] used by [user segment] for [primary task]. Generate a 40-point heuristic checklist specific to this context, organised by severity risk. Include mobile-specific heuristics if applicable." Generic checklists miss context-specific failure modes.
- 02
Technical Crawl
Screaming Frog
Run the free version (under 500 URLs) to identify broken links, redirect chains, missing meta descriptions, duplicate page titles, and accessibility flags. Export to CSV. This data feeds directly into the technical section of your audit report.
- 03
Accessibility Audit
WAVE + Axe DevTools
WAVE browser extension and Axe DevTools free tier both identify WCAG violations automatically. Run both because they catch different issues. Prompt Claude: "Here are the accessibility violations found. Prioritise by user impact for [disability type]. What are the legal risk implications for each Critical violation?"
- 04
Heuristic Evaluation Walk-through
Manual + Claude
Walk through the product manually against your custom checklist. Document raw observations as bullet points. Do not structure as findings yet - raw capture is faster and less biasing than structured capture.
- 05
Pattern Benchmark
Claude Code / Codex + Mobbin MCP + Figma MCP
Benchmark the audited flow against shipped products in minutes instead of half a day of screenshots. With Mobbin MCP and Figma MCP connected, prompt: "Read the [flow] frames from this Figma file. Search Mobbin for 10 comparable [flow type] flows from established [industry] products. For each of my raw observations, show how those products handle the same moment, with screen links." Attach the references to each observation as evidence, then ask for two or three variations written back into Figma, each stating which patterns it adopts, adapts, or rejects and why. Mobbin MCP requires a paid plan.
- 06
Visual Quality Pass
Manual - the Eye Test
Usability heuristics miss why a screen feels cheap or confusing. Run a second pass per key screen with seven questions: Hierarchy (what is allowed to be loud?), Spacing (does distance match relationship?), Type (how many sizes and weights, and why?), Alignment (do edges actually line up?), Density (does it fit the user and the task?), Colour (is colour carrying meaning?), Consistency (do components behave like a system?). Log each failure as a raw observation with a screenshot, so it goes through the same severity rating as everything else.
- 07
Severity Rating and Structuring
Claude
Paste raw observations into Claude: "Rate each observation by UX severity: Critical (blocks task completion), Major (significantly impairs experience), Minor (annoyance), Enhancement (improvement opportunity). For each Critical and Major finding, write a one-sentence evidence statement and a one-sentence design recommendation." Review every rating before finalising.
- 08
Findings Categorisation
Claude
Group findings by heuristic category: Navigation, Content, Interaction, Visual Design, Accessibility, Performance. Generate a severity matrix showing finding count by category and severity. This becomes the executive visual in your report.
- 09
Executive Summary
Notion AI
Prompt Notion AI: "Write a 200-word executive summary for a UX audit with these key findings. Audience: non-design leadership. Include: top 3 business impact risks, most critical finding, and recommended immediate action." Use this as the opening page of the report.
- 10
Recommendation Roadmap
Claude
Generate a prioritised roadmap: Quick wins (under 1 day), short-term (1 sprint), medium-term (1-3 months). Prompt: "Organise these audit recommendations into a delivery roadmap by effort vs. impact. Prioritise by highest impact / lowest effort for the quick wins quadrant."
Tools
| Tool | Use | Free tier | Client-safe |
|---|---|---|---|
| Claude | Custom heuristics, severity rating, structuring, recommendations | Free tier available | Yes |
| Screaming Frog | Technical crawl, broken links, accessibility flags | Free under 500 URLs | Yes |
| Mobbin | Shipped-product pattern evidence for findings and variations, via MCP | Paid plan required for MCP | Yes |
| Notion AI | Executive summary, report formatting | Free tier available | Yes |
| Microsoft Clarity | Behavioural data, rage clicks, session evidence | Completely free | Yes |
| WAVE | WCAG accessibility violation detection | Free browser extension | Yes |
Constraints
4 constraints
- Honest Assessment: AI severity ratings are a starting point, not a final judgment. Always apply your own professional judgment to severity - AI has no context for business risk, legal exposure, or client priorities.
- Consultant impact: AI compresses a 3-day audit to 4-6 hours. This directly improves fixed-bid economics and allows more thorough coverage in the same budget.
- Screaming Frog free tier caps at 500 URLs. Most small-to-medium sites are under this limit. For enterprise sites, either the paid version ($259/year) or manual sampling is required.
- Accessibility audits expose legal liability. For any WCAG Critical violations, frame the risk in terms of ADA/WCAG compliance and legal exposure, not just user experience.