Source: skill files (frontend-design, taste-skill, soft-skill, ui-ux-pro-max, critique, audit); § Beyond UI from raw/How_to_De-Slop_Every_AI_Output_Forever_With_1_Skill.md and raw/8_Claude_Skills_I_Can_t_Live_Without.md (Ben AI, added 2026-10-28)

A consolidated reference of every banned pattern across all 23 design skills, with the specific alternative to use instead. AI-generated UI has recognizable fingerprints — predictable color choices, default fonts, symmetrical layouts, and decorative patterns that real designers do not use. This guide is the antidote. Pin it. Reference it before every UI review.

Visual and Color

Banned PatternWhy It FailsDo This Instead
Purple-to-blue gradientsThe #1 AI aesthetic fingerprintIntentional palettes from ui-ux-pro-max database
Cyan on dark backgroundsInstantly reads as “AI dashboard”Tinted neutrals with a single warm or cool accent
Neon glows / neon accentsCheap sci-fi aesthetic, overused by AISubtle accent colors at full opacity --- no glow
Pure black (#000)Harsh, unnatural on screensTinted near-black (#0a0a0a, 1a1a1a with warm/cool cast)
Pure white (#fff)Flat, sterile feelTinted near-white (#fafaf9, f8f8ff)
Gray on colored backgroundsCreates visual mudUse a shade of the background color (color-mix or oklch)
Gradient textThe most recognizable AI decorative choiceWeight and size hierarchy for emphasis
Dark mode with glowing accents as defaultThis is the AI default --- not a deliberate choiceChoose light or dark intentionally based on product context
Glassmorphism everywhereAI uses glass as a default surfaceReserve glass for one or two accent elements max

Typography

Banned PatternWhy It FailsDo This Instead
InterDefault of defaults --- signals zero design thoughtGeist (technical), Outfit (friendly), Satoshi (modern)
RobotoGoogle’s default, ubiquitous in AI outputCabinet Grotesk (editorial), Plus Jakarta Sans (clean)
Arial / Open SansSystem fallbacks, not design choicesAny intentional font from the ui-ux-pro-max 57 pairings
System defaults”I did not think about fonts”Every project deserves a deliberate font choice
Monospace as aesthetic shorthandLazy “dev tool” signalingUse monospace only for actual code display
All-caps subheaders everywhereA single trick repeated becomes monotonousWeight contrast, size contrast, color contrast
Identical text sizes across levelsNo hierarchy, everything looks the sameModular type scale with clamp() fluid sizing

Layout

Banned PatternWhy It FailsDo This Instead
Three equal-width cards in a rowThe single most common AI layoutAsymmetric bento grid, masonry, or varied widths
Hero metric template (big number + label + trend)Every AI dashboard starts with thisUnique data presentations --- inline stats, contextual numbers, data narratives
Everything centeredReads as timid, lacking confidenceLeft-aligned content with asymmetric negative space
Cards wrapping everythingCards should group related info, not be universal containersMix full-bleed sections, offset panels, inline content
Nested cards (card inside card)Unnecessary visual depthFlatten hierarchy --- use spacing and borders
Edge-to-edge navbarGeneric, no personalityFloating navbar with top-4 left-4 right-4 spacing
Identical hero sectionBig heading + subtext + CTA + stock imageAsymmetric split, offset media, integrated demo, editorial layout
4-column footer with newsletterEvery AI-generated marketing pageFocused footer with only what users actually need

Motion

Banned PatternWhy It FailsDo This Instead
Bounce / elastic easingLooks playful in isolation, cheap in productionease-out-quart / ease-out-quint / ease-out-expo
Layout property animation (width, height, top, left)Triggers layout recalculation, causes janktransform and opacity only (GPU-composited)
Linear / ease-in-out as defaultsFeels mechanical, lifelessExponential easing curves feel natural
Instant state changes (no transition)Jarring, feels broken150-300ms for micro-interactions, 500-800ms for entrances
Animation for decorationDistracting, adds no informationEvery animation must communicate state change or draw attention
h-screen for full viewportBreaks on mobile (address bar), causes layout shiftsmin-h-[100dvh] for dynamic viewport height

Content and Placeholder Data

Banned PatternWhy It FailsDo This Instead
John Doe / Jane SmithInstantly signals placeholderReal-sounding names: “Sarah Chen,” “Marcus Rivera”
99.99% / 10x / 500% improvementUnrealistic, no one believes theseRealistic numbers: “23% faster,” “4.2 rating,” “$847/mo saved"
"Elevate” / “Seamless” / “Unleash” / “Supercharge”AI copywriting clichesSpecific value propositions tied to the actual product
Lorem IpsumScreams “not finished”Write real copy --- even rough copy beats placeholder
Title Case On EverythingAI defaults to title case for headingsSentence case for most headings, title case sparingly
”Getting Started is Easy” + 3-step processEvery AI landing page has thisShow the actual product experience, not a meta-description of it

Components

Banned PatternWhy It FailsDo This Instead
Generic card with icon + title + descriptionThe AI default componentVaried containers: inline stats, list items, media cards, data rows
Pill badges on everythingOverused, adds visual noiseSubtle indicators: color dots, weight changes, position-based hierarchy
Sun/moon dark mode toggleUnnecessary UI --- respect system preferenceUse prefers-color-scheme by default. Add manual toggle only if needed, with a less generic icon.
Modal for everythingInterrupts flow, annoying on mobileInline expansion, slide-over sheets, bottom sheets, dropdown panels
Identical testimonial cards in a rowSee “three equal-width cards”Staggered layouts, pull quotes, inline social proof
Progress bars with rounded ends and gradientsAI-generated progress indicator signatureSimple bars with square ends, or use numbers/percentages directly

Icons

Banned PatternWhy It FailsDo This Instead
Emoji as UI iconsNot professional, inconsistent across platformsPhosphor Light, Heroicons, or Lucide SVG icons
Using Lucide exclusivelyCreates visual monotonyMix one or two complementary icon sets
Rocket emoji/icon for “launch”AI clicheContextual icons tied to the actual action
Inconsistent stroke widthsMixing thin and thick icons looks carelessPick one icon set and stick with it throughout the project
Icons without labels (icon-only buttons)Ambiguous, fails accessibilityIcon + text label, or icon with tooltip at minimum

Code and Technical

Banned PatternWhy It FailsDo This Instead
div soup (divs for everything)No semantics, hurts accessibility and SEO<nav>, <main>, <section>, <article>, <aside>, <header>, <footer>
h-screenBreaks on mobile browsers with address barmin-h-[100dvh]
Arbitrary z-index (z-[9999])Unmaintainable, z-index warsDefined scale: z-10 (base), z-20 (dropdown), z-30 (sticky), z-50 (modal)
Broken Unsplash URLssource.unsplash.com is unreliablepicsum.photos/seed/{descriptive-name}/800/600 for consistent placeholders
Inline styles everywhereUnmaintainable, overrides are painfulTailwind classes or CSS custom properties
!important as a fixSpecificity problem, not a solutionFix the cascade --- more specific selectors or restructure

Beyond UI: a de-slop gate for every output (added 2026-10-28)

Creator: Ben AI · URL: https://www.youtube.com/watch?v=yuToKYHsAYA · Duration: 11:22 · Platform: YouTube · Uploaded: 2026-06-16 (humanizer and fact-checker tips from https://www.youtube.com/watch?v=bXnRA3pJavE, 2026-04-18)

The tables above cover UI. Ben’s de-slop skill applies the same idea to every AI output a person or team sends out: emails, LinkedIn posts, newsletters, proposals, support replies, internal docs. He built it for two reasons. First, his own quality bar drops when he is tired or rushed, and the skill holds the standard you set once (“a spell check, but for slop”). Second, each employee has a different quality bar and their AI agrees with them, which he connects to a16z’s “institutional AI vs individual AI” article (unread here) and the term “workslop.”

  • Two check layers. A universal layer catches widely agreed tells: em dashes, AI writing patterns and vocabulary, factual errors (claims checked against authoritative web sources), contradictions and unreadable writing. A company layer checks your own standard: tone of voice, brand identity (and visual guidelines for proposals), alignment with strategy and company facts, and whether the asset does its job.
  • Process. Classify the output type (marketing post, support reply, internal doc), pick the checks that apply, and read one reference file per check (an AI-writing-tells guide, a fact-check method, a voice guide with rules such as “no guru energy” and “no bold claims without backing”). Then three subagents grade the output separately, for less bias, and the skill returns one of three verdicts: good to go, good to go but consider these fixes, or not ready. Each issue comes with its source file and a suggested fix. You choose which fixes to apply.
  • Demos. A post-call follow-up email got “good to go with fixes”: some AI-writing signs, a claim that didn’t match the call transcript, and small readability and voice issues. A LinkedIn post got “not ready” for breaking hard rules. He also ran it on a newsletter and a sales proposal.
  • Wire it in. Make it the last step of writing skills, make it a team rule before anything AI-made goes out, and log every run (to the second brain or the skill folder) so recurring failure patterns across the business show up.
  • Scope. Ben says it is a detector, not a perfect fixer: it lacks the context for every possible asset type, but it makes the team aware of the bar. The free version has blanks for your voice and brand; a paid builder skill walks you through filling them.
  • Lighter options from the same creator: chain humanizer as the last step of every copy skill (“Can you update my LinkedIn post skill to always run the Humanizer skill after it’s generated the initial output?”), and run a fact-checker skill on drafts, on your own outlines before publishing, and on other people’s articles. See Ben AI’s Claude Toolkit.

To build the company layer quickly, write one reference file per standard (voice, brand, facts) and have the grader cite the file each issue breaks.

Before the gate (added 2026-11-09). The de-slop skill catches slop after drafting. Deslopify Your AI (Nufar Gaspar on The AI Daily Brief, raw/How_to_Deslopify_Your_AI.md) covers the steps before it: state your own view and confidence before prompting (“I’m about 60% sure the problem is onboarding. Push back hard before you agree with me”), ask for several directions instead of one, fix review checkpoints in advance, and run a persona review board. Gaspar calls humanizers “slop trying not to look like slop”, because “humanized text is still flagged about three-quarters of the time” by AI detectors.

Key Takeaways

  • The “AI color palette” (cyan + purple + neon on dark) is the single biggest giveaway. Eliminate it.
  • Font choice matters more than any other single decision. Switching from Inter to an intentional font transforms the feel.
  • Three equal-width cards in a row is the most common AI layout. Use asymmetry.
  • Content quality kills AI slop faster than visual polish. Real names, realistic numbers, specific copy.
  • Every banned pattern has a specific alternative --- this is not about avoiding things, it is about choosing better things.
  • Print this list and check it before every UI review until the patterns become instinct.
  • For writing and other non-UI output, run a two-layer gate (universal tells plus your company’s own standard) before anything goes out, graded by separate subagents with a clear verdict (Ben AI’s de-slop skill, § Beyond UI).

Try It

  1. Screenshot your most recent AI-generated UI
  2. Walk through each table in this guide and tally how many banned patterns appear
  3. For each match, note the “Do This Instead” column
  4. Rebuild with explicit instructions to Claude: “Use Geist font, not Inter. Use asymmetric layout, not equal cards. Use warm neutrals, not gray. Use ease-out-quart, not bounce.”
  5. Compare the before and after --- each eliminated pattern compounds into a dramatically different result
  6. For copy, take your last AI-drafted email or post and ask Claude to check it in two passes: universal AI-writing tells and factual claims, then your own voice and brand rules, with a verdict of good to go / fix first / not ready (the § Beyond UI pattern)