Find the problems in a design before a user or an engineer does, ranked by how badly each one hurts.
critique_experience
Before: what you give it
The example input, as sent.
Landing page for Design Check, a tool that scans HTML for what makes AI-generated work read as generic. Left 60%: a very large black sans headline 'Does it read as AI-generated? Ask before you ship it.' with a gray italic serif paragraph below. Right 40%: a full-height black panel containing a small pink uppercase label 'TRY IT: PASTE HTML', a black-on-black textarea pre-filled with sample code (cut off mid-line), and a pink 'CHECK THIS HTML' button. Nav: DOCS, BILLING, ALL THE POSSIBLES. No result is shown anywhere in the first viewport. Context: Audience: developers and designers using AI to build sites. Main action: paste HTML and run a check. Desktop 1512x785, above the fold.
After: what comes back
The answer from a real run on 2026-10-24, shown in full. Nothing is edited by hand.
[critical, visual] The textarea pre-filled with sample code is described as 'black-on-black', making the text invisible against its own background. Evidence: 'a black-on-black textarea pre-filled with sample code (cut off mid-line)' (observed) Fix: Set the textarea text color to a visible contrast against the black panel (e.g. light gray or white) and verify contrast (estimated, not measured). [warning, ux] The main action button 'CHECK THIS HTML' has no visible feedback path shown, no loading, result, or error state is present anywhere in the first viewport, so a first-time user has no idea what happens after clicking. Evidence: 'No result is shown anywhere in the first viewport' (observed) Fix: Show a result panel, scroll cue, or inline state change (loading/success/error) so the outcome of the primary action is discoverable without scrolling or guessing. [warning, ux] The sample code in the textarea is cut off mid-line, which reads as a half-finished or broken state rather than an intentional placeholder, undermining trust in a dev tool meant to be precise. Evidence: 'a black-on-black textarea pre-filled with sample code (cut off mid-line)' (observed) Fix: Either show a complete, short code snippet that fits, or add a fade/ellipsis treatment that signals 'truncated' intentionally. [warning, visual] Nav item 'ALL THE POSSIBLES' breaks the naming convention of the other two nav items (DOCS, BILLING), which are plain functional nouns, making its purpose and clickability ambiguous. Evidence: 'Nav: DOCS, BILLING, ALL THE POSSIBLES' (observed) Fix: Rename to a conventional label matching its function (e.g. FEATURES, PRICING, or RESOURCES) or clarify the voice intent if the playful tone is deliberate across the whole page. [warning, product] There's a BILLING nav item visible on the landing page before any value has been demonstrated (no result shown), which may signal to a first-time visitor that this is a paid tool before they've tried it, raising friction at the point of first use. Evidence: 'Nav: DOCS, BILLING, ALL THE POSSIBLES' and 'No result is shown anywhere in the first viewport' (observed) Fix: Either de-emphasize BILLING in the first-use nav or clarify near the CTA whether the check is free, so cost concerns don't compete with trying the tool. [note, ux] The pink label 'TRY IT: PASTE HTML' functions as an instruction, but with no placeholder-state cue (e.g. 'paste your own HTML here') it's unclear whether users must replace the sample code or can just click check on the pre-filled sample. Evidence: 'a small pink uppercase label TRY IT: PASTE HTML' and 'textarea pre-filled with sample code' (observed) Fix: Add a short caption like 'Try our sample, or paste your own' to clarify that the pre-fill is usable as-is. [note, visual] The very large black headline dominates 60% of the fold, which may pull attention away from the right-panel tool (the actual task) in a first glance. Evidence: 'Left 60%: a very large black sans headline' vs 'Right 40%: a full-height black panel containing...the tool' (observed, inferred emphasis) Fix: Consider testing whether headline size can be reduced or the eye-path adjusted so the tool panel competes more evenly for first attention, since the task is to paste and check, not just to read. Summary: No PRD was provided, so this is judged against general landing-page and tool-usability principles rather than the design's stated intent. The one change that matters most is fixing the black-on-black textarea text, which as described makes the primary input invisible and would stop any user from completing the main task, this alone is a ship-blocker. Secondary issues (missing result/feedback state, truncated sample code, inconsistent nav voice) would confuse or slow users but not necessarily block them. I could not assess actual contrast ratios, real interaction states (hover, focus, loading), or what appears below the fold, since only a static above-the-fold description was given.
What you get
Findings marked critical, warning or note, each with the evidence behind it (observed or inferred) and the smallest fix, then a ship-or-not summary.
What gets better
You hear the objections in the order a stakeholder would raise them, and each one says whether it was seen in your material or guessed.
When to use it
Before you present a design to a person, or before an agent hands finished UI to engineering.
What it cannot do
It cannot measure contrast, size or motion from a description or screenshot, and says "estimated, not measured" when it mentions them. Without a brief it judges against general principles.
Run it
Two free calls a day with no key. Send your own Anthropic key in an x-anthropic-key header to lift the limit.
curl -X POST https://verdict.allthepossibles.com/api/critique_experience -H 'content-type: application/json' -d '{"artifact":{"type":"description","value":"Landing page for Design Check, a tool that scans HTML for what makes AI-generated work read as generic. Left 60%: a very large black sans headline '\''Does it read as AI-generated? Ask before you ship it.'\'' with a gray italic serif paragraph below. Right 40%: a full-height black panel containing a small pink uppercase label '\''TRY IT: PASTE HTML'\'', a black-on-black textarea pre-filled with sample code (cut off mid-line), and a pink '\''CHECK THIS HTML'\'' button. Nav: DOCS, BILLING, ALL THE POSSIBLES. No result is shown anywhere in the first viewport."},"context":"Audience: developers and designers using AI to build sites. Main action: paste HTML and run a check. Desktop 1512x785, above the fold."}'
The other reviews
- generate_state_matrix
List every state a feature can be in, including the ones nobody drew.
- map_journey
Map the stages around a goal and name the entries, exits and interruptions the flow forgot.
- audit_copy
Review buttons, errors, empty states, confirmations and notifications, and rewrite the ones that fail.
- find_assumptions
Pull the unchecked beliefs out of a brief and turn each into a question you could answer this week.
- check_terminology
Find one concept that goes by several names across your UI, docs and API.