How to conduct an expert review: a practical guide with AI prompts

What is an expert review?

Expert Review is a usability inspection method in which an experienced UX professional evaluates a digital product by walking through its interface and applying their accumulated knowledge of usability principles, design patterns, user behavior, and domain context. Unlike the more structured Heuristic Evaluation (which follows a predefined set of rules like Nielsen’s 10 heuristics), Expert Review gives the evaluator freedom to draw on multiple frameworks, personal experience, and contextual understanding of the product’s users and business goals. MeasuringU research shows that most practitioners who say they conduct a heuristic evaluation are actually performing an expert review — a broader, more flexible assessment that combines formal principles with professional judgment.

What question does it answer?

What usability problems exist in this interface that users are likely to encounter?
Which design patterns violate established UX principles, and how severely?
Where does the interface create unnecessary friction, confusion, or cognitive load?
How does this product compare to current UX standards and conventions in its category?
What are the highest-priority fixes that would improve the user experience with the least effort?

When to use

Early in the design process (wireframes, prototypes) to catch obvious usability flaws before investing in development.
Before usability testing, to eliminate surface-level problems so that testing sessions focus on deeper behavioral questions.
When the budget or timeline does not allow for full usability testing with recruited participants — an expert review delivers 60-80% of the value at a fraction of the cost and time.
After a major redesign or feature launch, to verify that changes did not introduce new usability regressions.
As part of a regular UX audit cadence (quarterly or biannually) to maintain product quality.

Not the right method when the team needs to understand how real users behave, think, and feel — expert reviewers predict usability problems based on principles, but users routinely surprise experts with unexpected workarounds and mental models. Expert Review is most powerful when paired with usability testing — the expert catches the obvious issues first, freeing the test to reveal the non-obvious ones.

What you get (deliverables)

Issue log: a structured list of every identified usability problem, each documented with location, description, violated principle, severity rating, and annotated screenshot.
Severity-prioritized recommendations: a ranked list of suggested fixes organized from critical to cosmetic, with estimated implementation effort.
Executive summary: a 1-3 page overview of the overall usability state, top findings, and recommended next steps.
Annotated screenshots: visual documentation of problem areas with callouts explaining what is wrong and why.
Strengths report: a list of what the product does well — patterns the team should preserve.

Participants and duration

Participants: none — the evaluator is the instrument.
Evaluators needed: 1-3 experienced UX professionals. A single expert catches roughly 35-50% of issues; 3 evaluators working independently cover approximately 60-75%.
Review time: 4-16 hours per evaluator. A focused review of 3-5 key flows takes 4-8 hours; a full product review takes 1-3 days.
Total timeline: 2-5 days from briefing to report delivery.

How to conduct an expert review (step-by-step)

1. Define scope and objectives

Agree with stakeholders on what is being reviewed and why. Is this a full product audit or a focused review of specific flows? What are the business goals and known problem areas? Setting scope prevents the review from becoming an unbounded opinion piece.

2. Gather context

Review available context: user personas, analytics data (drop-off points, error rates, session recordings), prior research findings, and the product roadmap. An expert review without context produces generic observations; with context, it produces targeted, actionable findings.

3. Select evaluation frameworks

Choose which principles will guide the review: Nielsen’s 10 heuristics, WCAG accessibility guidelines, cognitive UX laws (Hick’s, Fitts’s, Miller’s), or a custom checklist. Most expert reviews combine 2-3 frameworks rather than relying on one.

4. Walk through the product as a target user

Complete the key tasks as if you were each target user persona. Do not skip edges — empty states, error messages, loading states, and permission dialogs are where most usability problems hide. Record your screen and take annotated screenshots.

5. Document each finding with structure

For every issue, record: location, description, violated principle, severity (cosmetic / minor / major / critical), and screenshot. Consistent documentation transforms the review from opinion into evidence.

6. Rate severity and prioritize

Sort findings by severity. Within each level, estimate implementation effort. The intersection of high severity and low effort defines “quick wins.” Build a 2x2 matrix of severity vs. effort.

7. Identify strengths and patterns

Call out what the product does well. A balanced report earns more trust from stakeholders than a list of complaints. Strengths matter because the team needs to know what to preserve during redesign.

8. Write the report and present

Lead with the top 3-5 findings and the business impact. End with a clear action plan.

How AI changes this method

AI compatibility: full — Expert Review is among the methods most transformed by AI tools. Automated heuristic analysis platforms can scan a page against Nielsen’s heuristics, check WCAG compliance, and flag common patterns in seconds. The human expert’s role shifts from identifying obvious violations to interpreting context, judging business impact, and connecting findings to specific user needs.

What AI can do

Automated heuristic scanning: Heurilens, UX-Ray, and HeuristiCheck scan live pages against heuristics and surface violations with severity ratings — replacing the manual checklist pass.
Accessibility auditing: axe DevTools, WAVE, and Lighthouse automatically catch 30-40% of WCAG violations, producing a baseline the expert supplements with manual checks.
Screenshot annotation: UX Critique and Capian generate annotated screenshots with issue descriptions automatically.
Pattern detection across screens: AI identifies inconsistencies in button styles, typography, spacing across an entire product.
Report drafting: An LLM generates first-draft executive summaries and severity-ranked findings, cutting report writing time from hours to minutes.

What requires a human researcher

Contextual judgment: AI flags a low-contrast button but cannot assess whether it is a deliberate design choice or a genuine accessibility failure.
Task-flow logic evaluation: AI scans individual screens but struggles with multi-step task logic — whether the sequence makes sense and whether error recovery is adequate.
Severity calibration: The actual business impact of an issue requires human judgment informed by analytics and business context.
Stakeholder communication: Presenting findings in a way that drives action is human work.

AI-enhanced workflow

Before AI, an expert review took 3-5 days. The expert spent roughly half the time on mechanical tasks (checking contrast ratios, verifying heading hierarchy, logging screenshots) and half on interpretation and recommendations.

With AI tools, the expert starts by running automated analysis on key pages to generate baseline findings in minutes. They then walk through the product focusing on contextual, flow-level, and strategic issues. The AI-generated findings serve as a checklist — the expert validates, adjusts severity, adds context, and removes false positives. The total timeline shrinks from 3-5 days to 1-2 days, with expert time concentrated on interpretation and communication.

Beginner mistakes

Reviewing without context

Walking through a product cold produces generic observations that could apply to any product. Always spend 30-60 minutes gathering context — personas, analytics, business goals — before opening the product.

Producing a complaint list instead of a prioritized report

Listing 50 issues without severity ratings overwhelms the team and leads to nothing being fixed. Use a severity-effort matrix to make prioritization actionable.

Ignoring what works

An all-negative review erodes trust. Calling out strengths demonstrates that the review is fair.

Confusing personal preference with usability principle

“I don’t like this color” is not a finding. “The CTA button has a 2.1:1 contrast ratio, failing WCAG AA” is a finding. Every observation must trace back to a principle.

Treating the review as a replacement for user testing

An expert review predicts where users might struggle, but predictions have a significant false-positive rate. Use the review to identify candidates for testing, not as the final word on usability.

Example from practice

A fintech startup preparing for Series B noticed their dashboard had a Net Promoter Score of 32, below the industry average of 45. A full usability study would have taken 4-6 weeks; the fundraising timeline required answers in 10 days.

Two independent experts each spent two days reviewing the dashboard using a combined framework: Nielsen’s heuristics, WCAG 2.2 AA, and a fintech-specific checklist. Together they identified 47 issues: 4 critical (including a bypassable confirmation dialog risking accidental transfers), 12 major, 18 minor, and 13 cosmetic. They also found that seven simultaneously-loading data widgets caused a 4-second delay correlated with a 23% session-start drop-off.

The team fixed the 4 critical and 8 highest-impact major issues in a two-week sprint. The follow-up NPS survey three months later showed a score of 48 — a 16-point improvement.

Tools

Automated heuristic analysis: Heurilens, Baymard UX-Ray, UX Critique, HeuristiCheck

Accessibility checking: axe DevTools, WAVE, Lighthouse

Screen capture and annotation: Loom, OBS Studio, Capian

Issue tracking: Notion, Airtable, Google Sheets

AI-assisted analysis: ChatGPT or Claude, Figma AI plugins

AI prompts for this method

4 ready-to-use AI prompts with placeholders — copy-paste and fill in with your context. See all prompts for expert review →.