Audit a landing page with nine AI design-critique prompts, surface 28 issues

@aliscodes repeatedly put his own landing page through a set of AI prompts, handing the same page to nine different "design-critique" prompts that surfaced 28 previously unnoticed issues — including one that undermines first-screen credibility. What makes the experiment worth noting is that it never asks the model "does this look good" in the abstract; instead it breaks a subjective design audit into concrete, actionable tasks: copy, visual hierarchy, typography, accessibility, and audience positioning.

Key findings

In the "is this page failing on design or copy" test, the model judged the problem to be copy rather than design, and @aliscodes accordingly rewrote the vague "sessions expire randomly" into a more concrete statement of user consequences. Another prompt asked "what here looks templated or AI-generated"; the model called out the template-like elements item by item, and the author realized his color scheme was almost entirely unchanged Tailwind defaults. For accessibility, the AI found six WCAG AA contrast failures, with the brand indigo missing the threshold by just 0.03 — fixable by changing a single hex value.

Visual hierarchy and typography

Analyzing the page by visual flow toward the conversion goal, the model noted that the most eye-catching element was a 72px decorative number, while the "book a call" CTA didn't appear until the sixth focal point — a mismatch between visual hierarchy and the conversion goal; under a "treat this page as a competitor to dissect" prompt, it also flagged a free code-review entry buried deep in the page. The typography audit showed 35 text combinations and 6 weights, with the main headline ranking only sixth largest; the price `$499` used a 56px font with 84px line height, reading more like body text than a display title. The author also had the AI check margins, padding, font sizes, and line heights against a unified 8pt system, and asked for a simplified type scale in the manner of a design director's first review.

Method and limits

The thread also shows why prompt design itself matters. When finding design references for a "fix AI-generated SaaS codebase" service page, the model initially misread the target audience and only corrected itself after follow-up; another prompt asked it to check layout problems at a 375px width. Overall, @aliscodes treats the AI as several reviewers with distinct lenses — design director, competitor, copy editor, accessibility inspector — turning subjective page feedback into an actionable issue list.

2026-07-15 ~ 2026-07-15 · 11 related posts