Web Design · UX & accessibility
The Best Prompt for a UX Audit
The best prompt for a UX audit narrows the review to one user, one task and the screens you actually supply. This prompt has the model walk the flow step by step, then report each problem with its location, the usability heuristic it breaks, a severity rating and a specific fix. It also separates what an expert review can establish from what only testing with real users can answer.
UX Audit Prompt
Runs a heuristic review of one user flow: a step-by-step walkthrough, findings tied to usability heuristics with severity ratings, and fixes.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Intermediate
- Complexity
- Multi-part reply
- Variables
- 6
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
6 variables · fields marked * matter most
You are a UX researcher carrying out a heuristic evaluation. This is an expert review, not a usability test: you can judge the interface against established principles, but you cannot know what real users did. ## What to review Product: [Product or site] The user: [Who the user is] The task they are trying to complete: [Task being audited] Device context: [Device context] Problems the team already suspects: [Problems you already suspect] Screens, in order (descriptions, on-screen text, or attached screenshots): [Screens in order] ## Method Step 1 — Walkthrough. Go through the flow one screen at a time as the user described above. For each screen note: what the user is trying to do, what the screen asks of them, and any point where they could hesitate, misread or get stuck. Step 2 — Findings. Turn the friction points into findings. For each one give: - Location: the screen and the element, quoting its label or text - Problem: what goes wrong for the user, in one or two sentences - Heuristic: which of the ten Nielsen Norman Group usability heuristics it relates to - Severity: 4 blocks the task, 3 causes errors or serious delay, 2 causes minor friction, 1 is cosmetic - Fix: a concrete change, including replacement wording where the problem is wording Step 3 — Prioritize. Pick the five findings to fix first, weighing severity against likely effort, and say why. Step 4 — Questions for testing. List what this review could not settle and would need observation of real users or analytics. ## Ground rules - Review only what is in the material. If a screen or state is missing (errors, empty states, confirmation), list it as not reviewed instead of guessing. - No generic advice. "Simplify the form" is not a finding; name the field and the change. - Do not state user behavior as fact or invent percentages. Write "a user may" or "risk of". - Report things the flow does well only if changing them would cause harm; limit this to three. - If two findings share one cause, merge them. ## Output 1. Walkthrough: a numbered list, one entry per screen. 2. Findings: a table sorted by severity with the five columns above. 3. Fix first: five items. 4. Not reviewed, and questions for user testing.
Use it when you are
- Reviewing a checkout, signup or booking flow before spending on user testing
- Getting a structured second opinion on a design ahead of a team critique
- Turning a vague sense that a flow is confusing into a prioritized fix list
- Preparing hypotheses and tasks for a round of usability tests
Why this prompt works
One user and one task set the scope
An audit of a whole product returns a scattering of surface remarks. Fixing the review to a single person completing a single task gives the model a path to follow and a standard to judge each screen by: does this help or hinder that person at that moment.
The walkthrough comes before the findings
Experienced reviewers go through a flow in sequence before writing anything up, because many problems only show in the transition between screens. Requiring the walkthrough first makes the model account for every step and gives you a way to check that each finding traces back to a real screen.
Findings follow a fixed anatomy
Location, problem, heuristic, severity and fix are the fields a designer or developer needs to act. Quoting the element's own label stops findings drifting into generalities, and linking each to a published heuristic gives the team a shared reference when they disagree about whether something matters.
It is honest about what an expert review cannot know
The opening line and the ground rules stop the model from presenting guesses about behavior as observations. Missing screens are listed as not reviewed, and open questions are handed to user testing, so the audit feeds your research plan instead of pretending to replace it.
How to use it
- Choose one flow with a clear start and end. Run the prompt again for each additional flow instead of combining them.
- Describe the user as a single person in a situation, including device and where they came from.
- Capture every screen in order, including error messages and the confirmation step. Type out labels and microcopy exactly, or attach screenshots.
- Add suspected problems if you have them, but keep them brief so they do not steer the whole review.
- Check each finding against the real screen before sharing it. Delete any that describe something that is not there.
- Use the questions for testing as the basis for tasks in your next usability session.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | If you attach screenshots, number the files in flow order and repeat that order in the text so findings can refer to screens by number. |
| Claude | Expected to work | Anthropic's guidance recommends XML-style tags for supplied material. Wrapping each screen description in its own <screen> tag makes locations in the findings easier to trace. |
| Gemini | Expected to work | Keep the four numbered steps as written. Google's prompting guidance favors explicit instructions and a defined output format over open-ended requests. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: A small online plant shop reviewing its guest checkout on phones.
You are a UX researcher carrying out a heuristic evaluation. This is an expert review, not a usability test: you can judge the interface against established principles, but you cannot know what real users did. ## What to review Product: Fernway, an online plant shop The user: First-time buyer on a phone, arriving from an Instagram ad for a specific plant The task they are trying to complete: Buy one plant as a guest, from product page to order confirmation Device context: Mobile web Problems the team already suspects: Analytics show many sessions end at the sign-in screen. Screens, in order (descriptions, on-screen text, or attached screenshots): Screen 1 — Product page: photo carousel, name, price, 'Pot size' dropdown defaulting to 'Select', button 'Add to basket'. Screen 2 — Basket: line item, 'Promo code' field, buttons 'Continue shopping' and 'Checkout' styled the same. Screen 3 — Sign in: email and password fields, button 'Sign in', small text link 'Continue as guest' below. Screen 4 — Delivery: 11 fields including 'Company' and 'Address line 3', delivery date shown only after postcode is entered. Screen 5 — Payment: card fields, button 'Submit'. ## Method Step 1 — Walkthrough. Go through the flow one screen at a time as the user described above. For each screen note: what the user is trying to do, what the screen asks of them, and any point where they could hesitate, misread or get stuck. Step 2 — Findings. Turn the friction points into findings. For each one give: - Location: the screen and the element, quoting its label or text - Problem: what goes wrong for the user, in one or two sentences - Heuristic: which of the ten Nielsen Norman Group usability heuristics it relates to - Severity: 4 blocks the task, 3 causes errors or serious delay, 2 causes minor friction, 1 is cosmetic - Fix: a concrete change, including replacement wording where the problem is wording Step 3 — Prioritize. Pick the five findings to fix first, weighing severity against likely effort, and say why. Step 4 — Questions for testing. List what this review could not settle and would need observation of real users or analytics. ## Ground rules - Review only what is in the material. If a screen or state is missing (errors, empty states, confirmation), list it as not reviewed instead of guessing. - No generic advice. "Simplify the form" is not a finding; name the field and the change. - Do not state user behavior as fact or invent percentages. Write "a user may" or "risk of". - Report things the flow does well only if changing them would cause harm; limit this to three. - If two findings share one cause, merge them. ## Output 1. Walkthrough: a numbered list, one entry per screen. 2. Findings: a table sorted by severity with the five columns above. 3. Fix first: five items. 4. Not reviewed, and questions for user testing.
What a good response should contain
- A five-entry walkthrough that notes the guest option is visually subordinate on the sign-in screen
- A findings table quoting labels such as 'Continue as guest' and 'Submit', each with a heuristic and a 1 to 4 severity
- Specific fixes, for example proposed button wording and which delivery fields to remove or make optional
- A not-reviewed list that includes form error states and the order confirmation screen, since neither was supplied
- Testing questions such as whether buyers notice the pot size dropdown before tapping 'Add to basket'
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- A heuristic review predicts likely problems; it does not show what users do. Treat severity ratings as a starting estimate.
- Quality depends on how completely you describe the screens. Missing microcopy and states cannot be assessed.
- Models can misread screenshots, especially small text and disabled states. Verify each finding on the real interface.
- The review does not cover performance, accessibility conformance or visual brand quality in any depth.
Alternative prompts
Web Design
Accessibility Audit Prompt
Reviews page markup against WCAG 2.2, separating issues confirmed in the code from those needing manual testing, with fixes and a test script.
- ChatGPT
- Claude
- Gemini
Web Design
Website Redesign Prompt
Diagnoses an existing site from the evidence you supply, then plans what to keep, merge or remove, the new structure, a URL map and a phased rollout.
- ChatGPT
- Claude
- Gemini
E-commerce
Product Page Audit Prompt
Audits pasted product page content against the five questions a shopper asks before buying, with quoted evidence, severity, fixes and rewrites.
- ChatGPT
- Claude
- Gemini
Research
User Interview Discussion Guide Prompt
Writes a timed discussion guide for customer or user interviews: screener, opening script, behavior-based questions with probes, and a table tying each question to a learning goal.
- ChatGPT
- Claude
- Gemini
Related resources
Guide
How to Write a Good AI Prompt
The five parts of a complete prompt (goal, context, task, constraints, output format), with a worked rewrite and a pre-send checklist.
Technical guide
How to Test AI Prompts
A seven-criterion rubric and a nine-field test record for evaluating prompts, plus why this site describes results in words and publishes no scores.
Frequently asked questions
Can this replace usability testing?
No. An expert review catches violations of known principles cheaply, which makes it a good first pass. It cannot reveal mismatches between your product and how your particular users think. The prompt ends with questions for testing so the two methods feed each other.
Should I paste screenshots or written descriptions?
Either works if the model you use accepts images. Screenshots carry layout and visual weight; written descriptions make sure exact labels are read correctly. For a short flow, doing both for the screens you are most worried about gives the model the most to work with.
What are the usability heuristics the prompt refers to?
They are ten general principles for interface design published by Nielsen Norman Group, such as visibility of system status and error prevention. The prompt uses them as a common vocabulary for classifying problems. The source link on this page goes to the full list.
Sources and references
- 10 Usability Heuristics for User Interface Design — Nielsen Norman Group
- Prompt engineering overview — Anthropic
- Prompt design strategies — Google AI for Developers
Was this prompt useful?
No account needed. Do not include personal information.