Verified against ChatGPT · 2026-08-08
Run a UX audit that ties every flagged issue to a specific funnel drop-off, not a generic heuristics checklist
Turns a page-by-page walkthrough of your signup or checkout flow into a prioritized list of UX issues, each one tied to where users actually abandon rather than a generic Nielsen-heuristics scorecard.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are auditing a specific web flow before a redesign sprint starts, not producing a generic heuristics scorecard. I will describe the flow screen by screen; you will find the issues that plausibly explain where real users are dropping off, not every stylistic nitpick you can find. FLOW BEING AUDITED Free-trial signup flow for a project-management SaaS, four screens from pricing page to first workspace created. SCREENS IN ORDER Screen 1: pricing page with three tiers. Screen 2: email + password form, no social login. Screen 3: 6-field company-details form, all marked required. Screen 4: empty workspace with no onboarding checklist. KNOWN DROP-OFF DATA 62% of signups who reach screen 3 never submit it; screen 2 to 3 conversion is 91%. PRIMARY USER AND GOAL A solo freelancer trying to see if the tool can replace their spreadsheet before a client call in twenty minutes. PHASE 1 — WALK THE FLOW Go screen by screen in the order given. For each screen, state the one decision or action the user is meant to take, and whether the screen's layout makes that action the obvious next step or makes the user hunt for it. Do not evaluate visual polish in this phase — only whether the intended action is findable and unambiguous. PHASE 2 — MATCH ISSUES TO THE DROP-OFF DATA If drop-off data was given, work backward from it: for the screen(s) with the worst known drop-off, generate the two or three most plausible UX reasons a user would abandon there specifically, not a generic list of best practices that could apply to any screen. If no drop-off data was given, say so explicitly and flag which screen you'd want data for first before trusting your own guess. PHASE 3 — PRIORITIZE Rank every issue found by estimated impact (how many users it plausibly affects) times ease of fix, not by how visually obvious the issue is. A subtle field-validation bug that silently blocks submission ranks above a spacing inconsistency, even though the spacing issue is easier to spot in a screenshot. WHAT NOT TO DO Do not pad the audit with a fixed set of heuristics applied uniformly regardless of whether they're relevant to this flow. Do not flag copy tone or branding preferences as UX issues unless they demonstrably block or confuse the user's next action. OUTPUT FORMAT A table: Screen | Issue | Why it plausibly causes drop-off | Estimated impact | Fix effort | Priority. Follow it with a two-sentence summary naming the single highest-priority fix and why it beats the others.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
GPT-5.1's default behavior when asked to "audit a UX flow" without constraints is to reach for a stock heuristics list (Nielsen's ten, or a generic accessibility-and-clarity checklist) and apply it uniformly to every screen, because that's the highest-probability completion for the bare instruction and it doesn't require the model to reason about what's actually causing abandonment on this specific flow. Anchoring Phase 2 to real drop-off data forces a different mode: instead of pattern-matching to generic best practices, the model has to work backward from an observed number and generate causal hypotheses that specifically explain that number, which produces genuinely diagnostic reasoning rather than a checklist recitation. Explicitly permitting "no data given, flag it" matters because otherwise the model will confidently rank issues by impact using invented percentages, which reads as authoritative but is fabricated; naming the absence of data is more useful than a confident-sounding guess dressed up as analysis. The impact-times-effort prioritization instruction counters a specific failure mode where visually obvious issues (spacing, color contrast, copy tone) get ranked above invisible ones (a validation rule that silently rejects valid input) simply because they're easier for a language model to describe vividly — forcing an explicit effort/impact axis breaks that bias toward describability over actual severity. The table-plus-two-sentence-summary format also matters practically: a bare table lets a stakeholder skim priorities, while the forced two-sentence takeaway prevents the audit from ending as an undifferentiated list where every row reads as equally urgent.
What you get back
Screen 3 | All 6 fields marked required, including 'company size' and 'phone number' | Users evaluating the tool solo have no company-size answer ready and abandon rather than guess | High (62% drop from data) | Low (make 4 of 6 fields optional) | P1. Screen 4 | Empty workspace, no first action suggested | User doesn't know what to do next and closes the tab | Medium | Low | P2. Summary: making the four non-essential fields on screen 3 optional is the single highest-leverage fix — it directly targets the only screen with measured drop-off data and requires no new design work.
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — branding & design, built and maintained for you — that's Scult's day job.
EXPLORE BRANDING & DESIGN
