anti-ai-design-patterns
Skill10 starsChanged 4 months agoMITView on GitHub
Detect and eliminate AI-generated design tells — "AI slop" and generic "AI-alien" aesthetics — so interfaces read as human-crafted and intentional. Use this skill whenever generating or reviewing any UI, component, landing page, email, or design-token set. Triggers include: single-side accent borders / left-rail cards, emoji or emoticons used as icons or section markers, default purple-to-blue gradients, everything-centered layouts, uniform rounded-lg corners everywhere, raw default Tailwind palette colors, Inter/Space-Grotesk-as-the-safe-default, cream + serif + terracotta clichés, lorem ipsum, glassmorphism on everything, uncanny uniform spacing, and any request to "make this not look AI-generated", "de-slop this UI", or "use an icon library instead of emoji". Pairs with design-fundamentals and css-styling-pixel-perfect.
In this repository14 other files it gives its agents.
- component-library-masterySkill
- css-styling-pixel-perfectSkill
- deployment-devops-workflowSkill
- design-fundamentalsSkill
- design-systems-architectureSkill
- design-to-code-workflowSkill
- design-tokens-systemSkill
- figma-expert-workflowsSkill
- frontend-framework-guideSkill
- inclusive-design-patternsSkill
- qa-testing-visual-regressionSkill
- responsive-universal-designSkill
- web-accessibility-a11ySkill
- web-performance-optimizationSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
If you cannot state *why* a color, radius, border, or glyph is there, it is probably a tell.
- Distinctive: line 82
One family per product: Never mix Lucide + Heroicons + Font Awesome — differing stroke widths and metaphors read as incoherent.
- Very specific: line 86
Never rasterize emoji into an <img> to "use it as an icon." That is the same tell with extra steps.
- Very specific: line 144
Commit to a type scale and stay on it.
- Distinctive: line 183
Do not replace an emoji tell with an over-designed icon-and-gradient tell.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Anti-AI Design Patterns
A field guide to the visual tells that mark an interface as machine-generated, and the intentional choices that replace them. The goal is not a different template — it is design that reads as deliberate: chosen neutrals, a real type system, asymmetry with reason, and iconography that belongs to one family.
Core principle: intentional beats default
An AI tell is almost never a single ugly element. It is the absence of a decision — the palette that was inherited instead of chosen, the border that decorates instead of encodes, the emoji dropped in because a real icon was more work. Every rule below reduces to one question:
Did a human decide this, or did it fall out of a default?
If you cannot state why a color, radius, border, or glyph is there, it is probably a tell.
1. The border tell — single-side accent rails
The single most common AI slop signature: a rounded card with a colored stripe on one side (usually left) and neutral borders on the other three.
/* ❌ AI TELL — accent rail on a rounded card */
.card {
border: 1px solid #e5e7eb;
border-left: 4px solid #6366f1; /* the giveaway */
border-radius: 12px;
}Why it reads as AI: the accent has no informational job — it is applied to every card regardless of state, so it decorates rather than encodes. A colored edge on a rounded corner also fights the geometry (the stripe clips awkwardly at the radius).
Fixes — pick based on intent:
/* ✅ A. If the color means nothing: drop it. Let type + spacing carry the card. */
.card { border: 1px solid var(--line); border-radius: 12px; }
/* ✅ B. If the color must encode STATE (error/warn/success): use a full token
set — background tint + text + a same-color full border, not a lone rail. */
.card--critical {
border: 1px solid var(--danger-line);
background: var(--danger-bg);
color: var(--danger-ink);
}
/* ✅ C. If you genuinely want an edge accent, make it a deliberate full-bleed
top border on a SQUARE-topped card, or an inset marker — a decision, not a default. */
.stat { border-top: 3px solid var(--accent); border-radius: 0 0 10px 10px; }Rule of thumb: a colored border must earn its place by encoding something true (severity, category, selection). If it is on everything, it is on nothing.
2. The emoji / emoticon tell — use an icon library only
Emoji as UI is the loudest AI-alien signal. It renders differently on every OS, breaks visual weight, ignores your stroke width, and cannot inherit color or a11y semantics.
{/* ❌ AI TELL — emoji as icons / section markers / bullets */}
<h2>🎨 Design</h2>
<li>✅ Ship it</li>
<button>🚀 Deploy</button>
<p>Nice work! 😄</p> {/* emoticons in product copy */}Rule: emoji and emoticons never appear in product UI. Use one icon library.
{/* ✅ Icons from a single library — inherits color, size, stroke, a11y */}
import { Palette, Check, Rocket } from "lucide-react";
<h2><Palette size={18} aria-hidden="true" /> Design</h2>
<li><Check size={16} aria-hidden="true" /> Ship it</li>
<button><Rocket size={16} aria-hidden="true" /> Deploy</button>Icon-library discipline (see references/icon-library-guide.md):
- One family per product. Never mix Lucide + Heroicons + Font Awesome — differing stroke widths and metaphors read as incoherent.
- Recommended sets: Lucide, Heroicons, Phosphor, Radix Icons, Tabler. Pick one.
- Size on a scale (16 / 20 / 24), match the icon stroke to your type weight, align to the text baseline.
- Decorative icon →
aria-hidden="true". Meaningful icon (icon-only button) →aria-label. - Never rasterize emoji into an
<img>to "use it as an icon." That is the same tell with extra steps.
Exception: genuinely emoji-native surfaces (a chat reactions bar, an emoji picker, a flag/country selector) — there the emoji is the content, not decoration.
3. The palette tell — inherited, not chosen
/* ❌ AI TELL — raw framework defaults, straight out of the box */
color: #6366f1; /* Tailwind indigo-500, untouched */
background: linear-gradient(135deg, #a855f7, #3b82f6); /* the purple→blue hero */
--gray: #808080; /* pure, hueless mid-grey */Tells: the default Tailwind/Bootstrap swatches used verbatim, the purple-to-blue gradient hero, and pure hueless greys.
/* ✅ Choose a palette. Bias neutrals slightly toward the accent hue so they
read as selected, not inherited. Define as tokens; style through tokens. */
:root {
--accent: #0c7c7c; /* a specific teal, not indigo-500 */
--ink: #1a1d23; /* near-black with a cool bias */
--ink-soft:#575d68; /* grey that leans blue toward the accent */
--line: #dce0e6; /* borders share the same bias */
}Guidance:
- Never ship raw default swatches. Shift hue/lightness even slightly so the palette is yours.
- Neutrals get a subtle hue bias toward the accent — a pure
#808080is the tell. - Spend boldness in one place; keep everything around it quiet. If the accent fights the ground, drop saturation or move to an analogous hue rather than adding a second loud color.
- Semantic colors (success/warn/danger) are a separate system from your brand accent.
4. The layout & shape tells
| Tell | Fix |
Everything centered (text-align:center on whole sections) | Use a real grid; left-align running text; center only what deserves emphasis |
One radius everywhere (rounded-lg on cards, buttons, inputs, avatars) | A radius scale (e.g. 6 / 10 / 16 / full) mapped to element size |
| Perfect symmetry, no focal point | Intentional asymmetry — an off-center hero, an uneven 60/40 split |
Uniform spacing (everything gap-4) | A spacing scale with rhythm — tight within groups, generous between them |
| Drop shadow + glassmorphism on every surface | Elevation as a system (0/1/2 levels); most surfaces are flat |
| Pill/badge on every label | Badges encode state; plain text labels for everything else |
5. The typography tell
❌ Inter (or Space Grotesk) as the reflexive "safe" choice for every project,
at one weight, one size, no scale.Fixes:
- Pair two deliberate typefaces (a display face used with restraint + a readable body face); add a mono utility face for data/labels if the subject is technical.
- Commit to a type scale and stay on it. Give headings
text-wrap: balance, body ~65ch line length, uppercase labels a touch of letter-spacing. - Inter is not banned — but if it is there because it is the default, that is the tell. Choose it for a reason.
6. The content tell
- Lorem ipsum is an instant AI/placeholder signal. Always write real, specific copy.
- Fake data that looks fake ("John Doe", "$1,234.56", "Lorem Company") — use plausible, varied, domain-true content.
- Copy written from the system's side ("webhook config") instead of the user's ("Notifications"). See
design-fundamentals. - Em-dash-heavy, evenly-hedged marketing prose — write specific, active, human copy.
7. The "AI-alien" uncanny tells
Subtler signals that something is off even when nothing is obviously wrong:
- Too consistent. Real design has intentional exceptions; machine output is eerily uniform. Vary card heights, let one item break the grid on purpose.
- Gradient text on headings, rainbow or purple→pink. Almost always a tell.
- Icon + emoji mixed in the same list. Pick one system (icons).
- Over-rounded everything including images and section containers.
- Centered max-w-prose everything with no layout ambition.
- Decorative dividers/numbers (01 / 02 / 03) on content that is not actually a sequence. Structure must encode something true.
Workflow: generate, then de-slop
- Build to spec using the other skills (
design-fundamentals,css-styling-pixel-perfect, etc.). - Run the anti-AI review pass — walk
references/review-checklist.mdagainst the output. - For each flagged item, state the decision it should encode. If there is none, remove it.
- Swap every emoji/emoticon for an icon from the chosen library; verify a single family.
- Verify tokens — no raw default swatches, neutrals biased, one accent.
When NOT to over-correct
Avoiding AI slop is not "add more decoration." Restraint is the point:
- A clean, flat, well-spaced page with one accent is never the wrong answer.
- Do not replace an emoji tell with an over-designed icon-and-gradient tell.
- A deliberately single-theme, minimal aesthetic is a valid choice, not an omission.
References
references/ai-tells-catalog.md— full before/after catalog with code.references/icon-library-guide.md— choosing and using one icon set correctly.references/review-checklist.md— the de-slop audit rubric.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

