I Gave the Same Landing Page Brief to 6 AI Tools. Here's What Happened.

The test is simple: write one landing page brief, give it to six AI tools, and evaluate what each produces.

The brief was for a fictional SaaS product. I kept it detailed but not exhaustively specific — the level of detail most people actually provide when they sit down to build something.

Here's the brief I used, and here's what happened.

The brief

"Build a hero section for a SaaS product called Findr. Findr is a search tool for design teams — it searches across Figma, Notion, Google Drive, and Slack simultaneously. Target audience: design leads and product designers at mid-size tech companies. Aesthetic: professional but not corporate. Dark background. The hero should communicate 'everything your team makes, instantly searchable.' Primary CTA: 'Search for free'. Secondary: 'See how it works'. Include a visual that demonstrates the product. The UI should feel premium — the kind of thing a senior designer would be proud to ship."

That's a realistic brief — specific about product and audience, directional about aesthetic, with explicit content and CTAs.

I gave this exact brief to Claude, ChatGPT, v0, Cursor, Lovable, and Bolt, with no additional prompting.

What each tool produced

Claude

Claude produced the cleanest code of any tool tested. The component was well-structured TypeScript with clearly named variables, proper separation of concerns, and no junk imports. The design itself: dark background (#0D0D0D), Inter 700 headline at 64px, a radial glow behind the headline, two CTA buttons, and a "search interface" mockup — a dark card with a search bar and three fake results.

Quality assessment: functional, clean code, middle-tier design. The search interface was the right visual concept but felt generic — a dark rounded card with placeholder text. The aesthetic was premium-adjacent but not premium.

ChatGPT (GPT-5)

GPT-5 produced a structurally similar result — dark background, Inter, radial glow — but made two interesting choices Claude didn't: it wrote better copy (the headline "Find anything your team has ever made" was more specific than my brief text), and it made the search mockup more realistic by including recognisable source icons (Figma's F logo, a simplified G for Google).

The code was slightly less clean than Claude's — some inline styles where Tailwind classes would have been cleaner — but the visual was marginally stronger.

v0

v0 defaulted to its shadcn/ui aesthetic, which is more grey than dark, more neutral than premium. The component was technically well-built — it used the right shadcn primitives and would integrate cleanly into a Next.js project — but it looked like a components library demo rather than a premium product.

Notable: v0 added a keyboard shortcut indicator in the search bar (Cmd+K), which was a smart product-design choice I hadn't specified and neither Claude nor GPT produced. It also suggested a command palette pattern rather than a traditional search bar. This showed a stronger understanding of how search UIs work in design tools.

Cursor

Cursor produced an empty component file with a comment asking me to clarify several design decisions before it would proceed: exact background colour, font stack, whether to use animations, whether the search mockup should be functional or illustrative.

This is actually the most honest response. Cursor operates as a developer tool, not a creative director. It knows that those decisions matter and that making them without direction produces generic output. In a real workflow, I'd answer the questions and get better output than any other tool. In a single-pass test, it produced the least.

Lovable

Lovable produced a full application — hero section, navigation, features section, pricing section, footer — in under 30 seconds. The design quality was the lowest of any tool tested: stock component styling, too much whitespace in the wrong places, a search mockup that was clearly a styled HTML input and nothing more.

What Lovable lacked in design quality it made up for in scope. It thought about the full experience, not just the hero. For a non-technical founder who needs a full page quickly, Lovable's output is more useful despite being less polished.

Bolt

Bolt's output was closest to Lovable's — full application with multiple sections — but the code quality was higher and the design was marginally more considered. The hero it produced had a subtle noise texture overlay that neither Claude, ChatGPT, nor v0 included — a detail that added visual richness.

Bolt also handled the responsive layout more carefully than the others, with explicit mobile breakpoints and a properly collapsing navigation.

What the results reveal

The default aesthetic is everywhere. Every tool that produced a hero section (all except Cursor) produced a dark background with a radial gradient glow behind the headline. This is the trained default. The brief said "dark background" and every model interpreted that as "#0D0D0D with a blue/purple radial gradient." Without specific colour direction, this is what AI produces.

Code quality and design quality don't correlate. Claude produced the best code and middle-tier design. Bolt produced lower-quality code but made more interesting aesthetic choices. v0 produced excellent code but the most generic design. These are separate dimensions.

Specificity of the brief shows immediately. The parts of the brief I was vague about — "professional but not corporate," "the kind of thing a senior designer would be proud to ship" — produced generic output. These are aesthetic judgments, not specifications. The parts I was specific about — "search across Figma, Notion, Google Drive, and Slack" — appeared consistently across all outputs.

The tools understand the product better than the design. Every tool correctly understood that a search interface should be the visual centrepiece of a search product's hero. None of them produced a design that looked like something a senior designer would be proud to ship — but all of them understood the product.

What a better brief would have produced

After running the test, I rewrote the brief with proper design specifications:

"Build a hero section for Findr, a unified search tool for design teams. Tech: React 18 + TypeScript + Vite + Tailwind. Background: #0A0C10 — deep navy, not black. Avoid radial gradient glow. Font: Syne (700) for headline, Inter for body. Color accent: #6EE7F7 (ice blue). Headline: 'Everything your team makes. Instantly.' in Syne 700, 68px, tracking -0.03em, white. Subheadline: 'Findr searches Figma, Notion, Google Drive, and Slack in under 100ms.' in Inter 400, 18px, white/60. Visual: a search interface card — dark (#131620), rounded-2xl, shadow-2xl — containing: a search bar with placeholder 'Search across your workspace...', keyboard shortcut badge Cmd+K, three results showing one Figma file, one Notion doc, one Slack message, each with the real icon, filename, and a two-line excerpt. Animate the results in on page load: each result slides in from the right, 80ms stagger. CTA: 'Search for free' in bg #6EE7F7 text-black, pill. Secondary: 'See it in action' ghost. No trust bar — the product is pre-launch."

Running this brief through Claude produces output that looks significantly better than anything any of the tools produced from the first brief. The design decisions — Syne instead of Inter, ice blue instead of generic blue, a specific visual with real source icons — change the quality of the output more than the tool does.

The conclusion

The tool you choose matters less than the brief you write.

The gap between Claude's output and Lovable's output, given the same vague brief, is smaller than the gap between a well-specified brief and a vague brief in the same tool.

Cursor's refusal to proceed without more information is actually the most sophisticated response. It's treating the brief like a real design project — where decisions about background colour and font stack are not incidental but load-bearing. The tools that just proceeded and produced the default aesthetic were being helpful in the wrong way.

The most productive way to use any of these tools is to make the design decisions yourself, specify them precisely, and let the AI handle the implementation. The AI is better at implementation than it is at creative direction. The designer is better at creative direction than the AI is.

That division of labour is the state of the art in 2026.