Loading
The web scan is free and needs no account.
114 articles on recognising AI-default patterns, building a design system, and shipping frontends that look intentional.
A 🚀 floating in a `h-12 w-12 rounded-full bg-blue-100` circle above every feature card is the cheapest tell in AI-built sites — it's what generators output when nobody installed an icon set.
Replit Agent takes a description to a deployed app with a database, secrets and hosting in one workspace. What its documentation says it does, the default interface to replace, the platform lock-in, and how effort-based pricing bills you.
Six grayscale logos at opacity-60, scrolling left forever on a 30-second keyframe — and not one of them is a real customer. The "Trusted by" strip is the most reflexively shipped, least earned element on any AI landing page. Here is the exact code, why it persuades nobody, and the one line that beats it.
"Powerful analytics without the complexity" is the most overused headline structure on the AI-built web. It names a generic benefit and a generic objection, says nothing, and you can swap the nouns into any product.
The v0 hero you shipped can land in the red on mobile. Here's the exact map from each slop pattern (framer-motion bundles, JS-gated reveals, a multi-megabyte stock hero) to the audit it tanks, and the fix.
Every card on every AI-built site rises 4px and grows a shadow on hover. The exact two classes, why generators reflexively reach for them, and what to do instead.
Ten startups, ten products, one homepage: the page generator defaults add up to, scrolled from sticky nav to footer, with every tell named where it shows up.
Six checks in the browser's developer tools that show whether a "hand-coded" site is a v0, Lovable, Bolt or Framer export with a new logo: the rendered font, the accent color, the class strings, the stack, the badges left in the source, and a scan.
In 2019 a near-black landing page was a statement made by maybe a dozen companies. By 2026 it's what a model returns when you type "modern SaaS." This is the seven-year lineage — Linear, Vercel, shadcn, v0 — of how one studio's taste became everyone's default.
One under-specified brief, three AI builders. How Webflow's AI site builder, Framer AI and v0 each decide layout, color, type, copy, motion and code when you leave them open, from their own documentation, and the line to add to the brief for each.
No, using Cursor or v0 won't get you penalized — but shipping the unedited default will cost you trust and conversions. Twelve real questions founders ask about AI-built sites, each answered bluntly with the part nobody wants to say out loud.
One slop hero, rebuilt one change at a time. Full before-and-after code, every decision named, every diff copy-pasteable. No theory — just the eleven edits that move a hero off the assembly line.
El `#3b82f6` es el acento más repetido de toda la web y delata que lo construyó una máquina. Aquí tienes ocho alternativas concretas, con su sector y el método para que pasen el contraste.
#3b82f6 ist die meistverbaute Akzentfarbe im Netz und der sicherste Beleg dafür, dass niemand sich entschieden hat. Hier kommt der praktische Weg zu einem Akzent, der nach Mensch aussieht: aus echten Fakten abgeleitet, in OKLCH getunt, kontrastgeprüft, als eine einzige Variable verdrahtet.
O #3b82f6 virou a cor de destaque mais comum da web e grita "isso foi feito por IA". Veja como derivar um accent de verdade, controlar tudo em OKLCH e ainda passar no teste de contraste.
Le bleu #3b82f6 est devenu la signature des sites générés à la chaîne. Voici comment dériver une vraie couleur d'accent, la piloter en OKLCH et passer le contraste, avec huit accents concrets et les secteurs où ils marchent.
Esa barra de navegación flotante de cristal esmerilado que genera v0 o Lovable ya no dice "premium": dice "generado". Te enseño a distinguir el cristal por defecto del cristal con criterio, y lo que de verdad cuesta `backdrop-filter` en un móvil de gama media.
Die mattierte Glas-Navleiste aus jedem v0- oder Lovable-Starter ist das neue Verräter-Muster. Warum Modelle reflexartig zu backdrop-blur greifen, was es wirklich kostet und wie du Glas-aus-Absicht von Glas-aus-Faulheit unterscheidest.
A barra de navegação em vidro fosco virou o sotaque do v0 e do Lovable: opacidades redondas, bordas iguais nos quatro lados e blur sobre fundo chapado. Veja como diferenciar o vidro de fábrica do vidro feito por gente.
La barre de nav en verre dépoli flottant n'est plus un signe de raffinement, c'est l'aveu d'un v0 ou d'un Lovable laissé sur ses valeurs par défaut. Voici comment repérer le verre par défaut et comment en faire du verre choisi.
Ese cohete metido en un disco azul claro encima de cada tarjeta de feature es una de las huellas de IA más fiables de la web. Te explico por qué los generadores lo eligen y qué poner en su lugar.
Ein Emoji in einem hellblauen Kreis über jeder Feature-Card ist 2026 der zuverlässigste KI-Fingerabdruck im Web. Hier steht, warum Generatoren ihn produzieren, warum er billig wirkt und was du stattdessen baust.
Aquele círculo `bg-blue-100` com um emoji no meio do card de features é uma das digitais de IA mais fáceis de achar na web hoje. Some com ele e seu site para de parecer gerado por máquina.
Un emoji posé dans un petit disque bleu clair, c'est l'une des signatures IA les plus fiables du web. Voici pourquoi les générateurs y reviennent toujours, pourquoi ça fait cheap, et quoi mettre à la place.
Veinte caracteres de Tailwind que aparecen en todas las landings generadas por IA: la misma tarjeta que sube 4 píxeles y echa sombra. Te enseño a verlo, por qué penaliza el rendimiento y qué poner en su lugar.
`hover:-translate-y-1 hover:shadow-xl` klebt auf jeder KI-gebauten Karte, egal ob sie klickbar ist oder nicht. Warum die immer gleiche 4-Pixel-Anhebung nach Maschine aussieht und welche Hover-Gesten stattdessen etwas bedeuten.
Todo card sobe 4px e ganha sombra no hover, do botão ao logo do rodapé que nem clicável é. Vinte caracteres de Tailwind viraram a impressão digital da era do build com IA — e dá pra trocar por gestos que realmente significam algo.
Toutes les cartes montent de 4 pixels au survol et gagnent une ombre, qu'elles soient cliquables ou non. Voici pourquoi ce réflexe trahit une page générée par IA, et comment lui substituer des hovers qui veulent dire quelque chose.
Apple lo dirigió celda a celda; v0 lo escupe sin pensar. Cómo el bento grid pasó de rejilla de features cuidada a la forma por defecto que delata una web generada por IA.
Das Bento-Grid war mal Apples handgesetzte Feature-Vitrine. Heute ist es der Default, den v0 ausspuckt, wenn es nichts Konkretes zu sagen hat: vier Kacheln mit dem gleichen Reis drin.
A Apple inventou um grid de feature dirigido por arte; o v0 transformou ele no formato padrão que sai quando o modelo não tem nada específico a dizer. Como identificar o bento de IA na classe CSS e o que usar no lugar.
En quatre ans, le bento est passé de la grille de fonctionnalités ciselée par Apple à la forme que v0 recrache quand il n'a rien de précis à dire. Voici comment le repérer en cinq secondes, et quoi faire à la place.
Cada sección de las landing generadas por IA da el mismo saltito de 16 píxeles al hacer scroll. No es casualidad: es la firma del motion sin criterio. Te explico por qué pasa y cómo animar con intención.
Jeder Abschnitt hüpft 16 Pixel nach oben und blendet ein, 500 Millisekunden, ease-out, getriggert bei 30 Prozent Viewport. Fade-in-up ist das text-gray-600 der Animation. Hier ist, warum die Maschine genau hier landet und wie du Bewegung mit Absicht baust.
Toda seção sobe 16 pixels e dá um fade ao entrar na viewport: mesma duração, mesmo easing, mesmo gatilho. Esse é o motion slop da IA, e dá pra reconhecer em 30 segundos.
Chaque section qui monte de 16 pixels et s'estompe, toujours la même durée, le même ease-out : c'est le text-gray-600 du mouvement. Voici pourquoi l'IA y tombe à chaque fois, et comment animer avec une vraie intention.
Ocho de cada diez landings de SaaS arrancan con el mismo fondo casi negro, el mismo subtítulo en gris zinc-400 y el mismo resplandor índigo detrás del titular. El oscuro no es el problema: el oscuro generado por defecto sí.
Acht von zehn neuen SaaS-Landingpages laden in dasselbe Beinahe-Schwarz: zinc-950, weiße Headline, ein einzelner Indigo-Glow. So erkennst du den generierten Dark-Hero und kommst mit fünfzehn Minuten Arbeit wieder raus.
O padrão da IA deixou de ser o gradiente azul-roxo e virou near-black: mesmo zinc-950, mesmo text-zinc-400, mesmo glow indigo. Veja como manter o escuro e perder a digital de página gerada.
Le hero bleu-violet sur fond blanc est mort, vive le near-black. Huit landing pages SaaS sur dix tournent aujourd'hui le même zinc-950, le même accent indigo et le même glow centré. Voici comment garder le sombre sans porter l'empreinte de la machine.
Es el titular más repetido de la web generada por IA: una promesa que encaja en cualquier producto y, por tanto, en ninguno. Te enseño a detectarlo y a escribir el titular que solo tu producto puede firmar.
"Mächtige Analytics ohne die Komplexität" ist kein Versprechen, sondern ein Platzhalter. Warum Sprachmodelle reflexartig zu dieser Headline-Struktur greifen und wie du sie durch eine Zahl, einen Namen oder ein echtes Ergebnis ersetzt.
"Analytics poderoso sem a complexidade" serve para qualquer empresa do mundo — e é exatamente por isso que não serve para a sua. Como detectar a estrutura de título mais batida da web feita por IA e escrever uma que só você poderia assinar.
"Analytique puissante sans la complexité." Le titre le plus surexploité du web généré par IA décrit zéro produit. Voici comment le repérer et le remplacer par une promesse que seul votre produit peut tenir.
The complete forensic catalogue of typographic tells that mark a site as AI-generated, from Inter and Geist defaults to leading-relaxed, tracking-tight, missing tabular figures, and unchosen Tailwind gray. 62 numbered entries, each with the AI default and the human fix.
Three in five AI-startup landing pages now load Geist-Variable.woff2 from /_next/static/media/ — not because anyone chose it, but because Vercel's toolchain emits it by default. Here's how to spot the tell and what to ship instead.
Ask Claude Code for a landing page and you get #3b82f6, Inter, and rounded-2xl in 40 seconds. Here are the four files and prompts that move the model off the statistical center for good.
Leave the look open and every AI builder fills the gap with its default. What v0, Lovable, Bolt and Framer AI each start from, what they do when your prompt says nothing about design, and how to override them, from their own documentation.
A fintech founder spots the from-blue-500 to-violet-600 gradient in seconds and the retainer is gone. Here's the four-step workflow to run AI at full speed and still hand off work no client can read as machine output.
A stranger's site, one URL, ninety seconds. The CSS ships in plaintext and DevTools reads it back resolved. Here are the six computed properties that tell you whether the "bespoke design system" came out of a generator.
Apple's bento grid was art-directed craft. Four years later it's the shape v0 returns when it has nothing specific to say — rounded-3xl tiles, two-stop slate gradients, and a backdrop-blur over nothing.
A floating frosted-glass nav pill — bg-white/10, backdrop-blur-md, border-white/20, rounded-2xl — is now generated so often it reads as "AI" before it reads as "premium." Here's how to spot glass-by-default and how to ship glass that looks like a person made it.
Every AI-generated SaaS hero now loads on #09090b — zinc-950, indigo accent, radial glow, tracking-tight 7xl headline. Here's why the machines reach for near-black, and the five fixes that cost fifteen minutes.
Every section on an AI-built page does the same 16px hop, 500ms, ease-out, fired at 30% viewport. The library changes; the fingerprint doesn't. Here's why the machine always lands here, and how to ship motion that actually decided something.
Framer's AI generates pages you edit on a design canvas, which makes generic layouts cheap to break. Its costs are the copy and imagery it fills in for you, and a site that lives on Framer. What its documentation says, the four tells, and the prices.
#3b82f6 is the most-shipped accent on the web, and untouched blue-600 announces that nobody chose it. Here's how to derive a real accent, build it in OKLCH, and pass contrast.
A senior dev would never wrap `price.toFixed(2)` in a try/catch — but Cursor will, every time. Twenty-three specific patterns that mean an LLM wrote the code, each with a sample, the reason the model defaults to it, and a refactor recipe.
Open ten startup pages launched last week, squint, and they collapse into one composite hallucination: a from-blue-600 to-purple-600 hero, three lg:grid-cols-3 cards, a Lucide Check, a footer ending in "Built with care". A field guide to why the web looks like one product, what public data shows, and the cheapest ways to break the pattern.
A seller lists 80 Midjourney prints before lunch at the cost of one generation; a 10,000-article content farm spins up for about $115. This is the money map of the 2026 slop economy: who mints it, who sells it, and who profits no matter who wins.
Anthropic's own guide says Claude defaults to generic designs with white and purple backgrounds unless told otherwise. Nine levers, a master prompt, use-case prompts and a 60-second audit that steer any model off Tailwind blue, Inter and the three-card grid.
Bolt runs Node.js in your browser tab, which makes it quick for JavaScript prototypes and fragile when the code leaves the tab. What its documentation says it does, where its pages turn generic, the WebContainer limits, and where updates are posted.
Your Lovable site got 12 likes and one DM that said "looks AI." Here are 35 prioritized fixes — with real hex codes, shadcn class strings, and paste-ready code — to break the AI visual signature in 30 minutes, 4 hours, or one weekend.
Five times since 2011, a framework promised to set the web free and instead dressed half of it in the same costume. Here is the mechanism — and why the LLM training pipeline now runs that cycle in months instead of years.
Twenty-one verifiable signs, each scored 1 to 5, that separate a v0 or Lovable export from genuine custom work. Open DevTools, check the hex, count the cards, read the headers — and get a number between 0 and 75.
An AI builds the same SaaS page every time: Inter, a blue-purple gradient, three rounded cards, a bg-blue-600 CTA. Here are 73 specific, code-backed patterns that pull a site off the centroid — and the eight to ten you actually need.
HouseFresh says Google's March 2024 core update cost it 91% of its search traffic; Retro Dodo reported an 85% drop from September 2023. The helpful content updates of 2022 to 2026, dated from Google's own records, who got hit, and the 2026 playbook.
The centered pill, gradient headline and button row is what v0, Lovable and Bolt defaults add up to. Here are 21 structural alternatives, each with a name, code, and a usage rule.
Lovable builds and hosts full-stack apps on one fixed stack, and asks about palette, type and layout only if your prompt leaves them open. What its documentation says it does, where the generic look comes from, the 13 tells, and the way out.
Three identical cards with a "Most Popular" badge on the middle one: the default every generator reaches for. Here are 15 layouts that aren't that, each with a code skeleton, a real example, and a warning.
Open ten AI-startup landing pages and you get the same slate-50 card, the same outline button, the same Inter, the same blue-to-violet hero. A field study of how one 2023 idea became the fingerprint of the AI-built web — and nine concrete ways to use shadcn without shipping slop.
#3b82f6, the violet-to-pink gradient and three rounded-2xl cards are the Comic Sans of 2026: a watermark that says "a model made this." Where the defaults come from, what the public data shows, why people spot them in a glance, and twelve palettes that escape them.
What v0 is, from its own documentation: the Next.js, shadcn/ui and Tailwind starter every chat begins with, where that makes it fast, why its pages share a look, and fourteen fixes for the defaults.
Where vibe coding actually works (internal CRUD, webinar landings) and where it costs you the brand (homepages, pricing pages, anything that has to age). With the post-vibe stack serious teams converged on.
O que o termo significa, quem o popularizou e quando, por que sites feitos com IA compartilham os mesmos padrões, o que nosso scanner encontrou em três páginas públicas e como sair deles.
Cosa significa il termine, chi l'ha reso popolare e quando, perché i siti fatti con l'IA condividono le stesse impostazioni predefinite, cosa ha trovato il nostro scanner su tre pagine pubbliche e come uscirne.
Was der Begriff bedeutet, wer ihn wann bekannt gemacht hat, warum KI-gebaute Websites dieselben Voreinstellungen teilen, was unser Scanner auf drei öffentlichen Seiten gefunden hat und wie du davon wegkommst.
Qué significa el término, quién lo popularizó y cuándo, por qué las webs hechas con IA comparten los mismos valores por defecto, qué encontró nuestro escáner en tres páginas públicas y cómo salir de ahí.
Ton site IA a la bonne silhouette mais n'importe quel dev le reconnaît en trois secondes. Voici 23 modifications concrètes, code à l'appui, pour casser la signature Lovable / v0 / Bolt sans tout recoder.
Clic droit, Inspecter, #3b82f6 : la signature v0/Lovable/Bolt se lit en un onglet. Une grille de 17 signes pour transformer l'intuition « ça sent l'IA » en un score sur 60 défendable.
Anthropic le reconnaît lui-même : sans consignes, Claude produit des designs génériques sur fond blanc et violet. 9 leviers, un system prompt maître, 20 prompts prêts à coller et un audit final pour sortir Claude, GPT-5 et Gemini de leur moyenne.
Ce que veut dire le terme, qui l'a popularisé et quand, pourquoi les sites faits par IA partagent les mêmes réglages par défaut, ce que notre scanner a trouvé sur trois pages publiques, et comment en sortir.
Le #3b82f6, le dégradé from-purple-500 to-pink-500 et la grille de 3 cards rounded-2xl sont devenus la signature involontaire de l'IA. D'où viennent ces réglages, ce que montrent les données publiques, pourquoi on les repère en un coup d'œil, et 12 palettes pour en sortir.
Every AI coding tool falls back to a default when the prompt leaves room. Where each one's default comes from, what each gives you to override it, according to its own documentation, and what Sailop's free scan checks today.
"Effortlessly streamline your workflow" is a staple of AI hero subheads and almost never in copy written by someone with skin in the game. Here are 40 such phrases, why each is a tell, and the concrete line to write instead.
Comic Sans was never the problem; using it by default was. Tailwind's blue family is going the same way on AI-built pages. Where the default comes from, why it costs you, eleven accents outside the blue band, and how to make the change stick.
Open a handful of AI-built landing pages and most will share several of these tells: the blue-to-indigo hero, three identical cards, animate-pulse on the popular tier. Here is each giveaway, the exact class that produces it, and the fix.
What the term means, who popularised it and when, why AI-built sites share the same few defaults, what our scanner found on three public pages, and how to get off those defaults.
Inter, #3B82F6, three cards in a row: the five tokens that mark a site as AI-built. Here's how to rewrite them in an afternoon without touching the component logic.
v0 ships the same tokens to everyone: Inter, #3B82F6, rounded-lg, a 3-column grid. Here are all nine defaults and the exact CSS to override each one.
Tailwind ships 242 color shades and AI output leans on a handful of them. The result: a generation of sites built on an infinitely flexible tool that all reach for bg-blue-600, rounded-lg, and shadow-md. The fix isn't dropping Tailwind. It's using the rest of it.
Bolt.new ships the same #667eea→#764ba2 hero, Inter-everything type, and "Get Started Free" CTA in nearly every project. Here are the seven tells and the exact fix for each.
Most "design" MCP servers are shadcn generators with extra steps. These ten detect, validate, or generate against your real project — screenshots, contrast ratios, font load impact, and a scanner that flags your fourth identical grid-cols-3.
Prompt any LLM for "a SaaS landing page" and you get blue-600, Inter, rounded-2xl, three feature cards, and fade-up on scroll — every time. This catalogs all 87 of those patterns, why each happens, and the exact fix.
After `bg-`, a model is far likelier to write blue than fuchsia, not because blue looks better but because that is what it learned. Here is the token-probability mechanism that makes AI sites alike, and why temperature can't fix it.
A from-scratch procedural design system generator in TypeScript: seeded mulberry32 PRNG, an HSL palette that dodges the 200-290 "AI band," WCAG-enforced shade scales, and golden-ratio spacing. Real code, real numbers, no templates.
What the HTTP Archive's 2025 Web Almanac measured about AI-built sites: Lovable and Vercel subdomains in the crawl, the rise of "AI purple" and why it tracks Tailwind, no significant surge in gradients, shadows or fonts, and Inter's climb to about 1.5% of pages.
Type "build me a features section" into v0 and you get three `rounded-lg shadow-md p-6` cards every time. Here's why users skip cards two and three, and five layouts that earn a read instead.
A blue-to-purple gradient, Inter, three rounded cards, fade-up on scroll: the visual fingerprint of AI-generated frontends. Here is how to detect it, prevent it, and rip it out of an existing codebase.
Ninety-plus patterns that mark frontend code as machine-made — from #3B82F6 primaries to fade-up-on-scroll — each with the AI default, why it tells, and what a human ships instead.
A paid theme, a free shadcn starter and a page an AI wrote in a minute can share the same blue, the same three-card grid and the same fade-up motion. Why templates converge, and a five-step check to run on any template before you build on it.
Your coding agent lints TypeScript and runs tests, but it will happily ship bg-blue-500 and transition: all for the thousandth time. An MCP server gives it the one sense it lacks: knowing what generic looks like.
Move your primary hue from 217 to 28, swap grid-cols-3 for a 5fr_3fr split, and kill every transition-all. Six measurable changes that drop a Sailop score from 78 to 29.
ESLint will green-light a `bg-blue-500 rounded-lg shadow-md transition-all` card without blinking — valid code, maximum AI slop. Here is the linter built for the layer ESLint can't see.
A dark rectangle with red, yellow, and green dots is a fixture of AI-generated dev tool landing pages. It is a decorative lie that fakes a macOS window nobody can click. Here is why it's an instant AI tell and what to ship instead.
Blue-500 primary, Inter, grid-cols-3, transition-all duration-300: the same seven defaults show up in every AI-generated site. Here's how to name and measure each one.
Inter font, #3b82f6 buttons, a grid-cols-3 of identical cards. AI design converges because of mode collapse and Tailwind defaults baked into training data — and the fix is constraints, not freedom.
Open DevTools, inspect three elements, and look for one timing value: transition-all 0.3s ease-in-out. If all three match, you're reading AI output. Here's the full 30-second field guide.
73 NEVER/ALWAYS rules that separate default from intentional: hue 200-290 is banned, #ffffff is banned, transition: all is banned. Each one is a measurable line between slop and craft.
Claude Code reaches for blue-500, Inter, and grid-cols-3 by default. Wire in the Sailop skill and the same prompts ship #c2592e, Bitter, and an asymmetric 2fr 3fr grid instead.
A typical AI-written project-management landing page scores 84/100 (Grade F) on every slop dimension. Here is the line-by-line CSS that takes it to 18: burnt sienna instead of blue-500, Bitter instead of Inter, 5px gaps instead of the 4px grid.
Inter is the best font on the web, which is exactly why your site disappears into the crowd that uses it. Here are 10 pairings, the four CSS settings AI never touches, and the one move that instantly reads as designed.
Sailop hashes a seed string into a deterministic RNG, then generates a full design system in HSL with a hard exclusion zone over the blue-indigo-violet AI band. Same seed, same palette, every value WCAG-checked and off the Tailwind defaults.
Your pipeline blocks a missing semicolon but waves through #3b82f6 buttons and transition-all everywhere. Here's how to fail the build on generic design — with a real GitHub Actions workflow and a score threshold.
Give an AI total freedom and it picks bg-blue-500 every time. Give it "any hue except 200-290" and it picks #c2592e. The fix for AI homogeneity isn't smarter models. It's tighter constraints.