DEV Community

Cover image for Why Do AI-Generated Websites All Look the Same? We Graded 72 (2026)
Landing Page Gods
Landing Page Gods

Posted on Edited on

Why Do AI-Generated Websites All Look the Same? We Graded 72 (2026)

We graded 72 live sites built with Lovable, v0 and Bolt, and then contrasted them 20 well-known SaaS sites. The grading criteria is Landing Page Gods' 14 agency criteria, which covers a range of dimensions from positioning, copy, USPs, design quality, and EEAT. The AI-built sites averaged 46 out of 100 and the SaaS sites 72. Most of the 25-point gap is in design quality, visuals, selling points, forgetability, and EEAT. Lesser criteria like icons and fonts are worth about 3 points.

8 of the 72 sites we graded

What we graded

  • 72 live landing pages built with an AI builder, 24 per builder. Lovable sites come from Lovable's blog and madewithlovable.com, v0 sites from the showcase on Vercel's community forum and v0's template gallery, and Bolt sites from Bolt's customer stories, its 2025 hackathon on Devpost and madewithbolt.com.
  • 20 well-known SaaS sites: Linear, Stripe, Resend, Clerk, Raycast, Attio, Mercury, Ramp, Notion, Supabase, Vercel, Cursor, PostHog, Retool, Loom, Figma, Airtable, Plaid, Intercom and Descript.
  • How: our landing page analyzer rendered each page on desktop and phone on 21 November 2026 and scored 14 criteria from 0 to 10, weighted into a score out of 100. Each criterion score is the median of 3 runs.

The scores

Lovable v0 Bolt All 72 AI-built 20 SaaS sites
Average score 49 47 44 46 72
Highest 69 76 63 76 82 (Stripe)
Scored 70 or more 0 1 0 1 18
Scored under 50 12 15 18 45 0

The top AI-built site, Pool Atlas (v0, 76), is an editorial ranking of hotel pools with real photography and flat, factual copy ("135 pools in 48 countries"), outscoring 17 of 20 SaaS sites.

Where the points go

Criterion (weight) AI-built average SaaS average
Positioning and copy
Positioning (12) 4.5 6.0
Selling points (10) 4.2 6.7
Buyer's terms (6) 5.5 7.2
Human copy (8) 4.1 5.5
Memorable idea (8) 4.1 7.0
Clarity
Clarity (8) 4.9 6.2
Flow (6) 5.2 6.8
Design
Design quality (12) 4.7 8.0
Type (5) 4.2 7.7
Icons (4) 4.1 8.2
Consistency (5) 6.3 8.2
Visuals (8) 4.7 8.6
Depth and EEAT
Site depth (3) 5.9 10
EEAT (5) 3.4 7.8

Scores are out of 10. The biggest losses, weighted:

  • Design quality, 4 points. The grader called 61 of the 72 designs a template layout (a centered hero, card grids and pill badges), against 3 of the 20 SaaS sites.
  • Visuals, 3 points. Every SaaS site scored 7 or more, mostly with real product UI, and 43 of the AI-built sites scored 5 or less.
  • Selling points and a memorable idea, 5 points together. 41 AI-built sites scored 4 or less on selling points and 44 on a memorable idea, against none and 1 of the SaaS sites.
  • EEAT, 2 points. 57 of the 72 scored 4 or less, with anonymous teams, testimonials without a company and no customer logos. The lowest SaaS score was 5.
  • Positioning, 2 points. The grader found the alternative missing or buried on 48 of the 72 AI-built pages and on 13 of the 20 SaaS sites, which is why even the SaaS sites average 6.

Consistency is where the AI-built sites came closest (6.3), because a component kit gives every section the same spacing, corners and color.

Shared defaults

61 of the 72 AI-built sites used a stock icon set, almost all Lucide (shadcn/ui's default), against 1 of 20 SaaS sites. Inter, Geist and the system font are fine typefaces, and the grader's type score rewards a deliberate pairing over the default stack. Together, icons and type are about 3 of the 25 points between the two groups.

The layout patterns repeat too: a pill label above the headline on 19 of the 72 (2 of 20 SaaS sites), an FAQ on 20 (2 of 20) and a "How it works" section on 9 (none). The indigo-500 button that earlier posts on this question blamed is mostly gone: 5 of 52 main buttons on the AI-built sites were indigo or violet.

Why they converge

  • The component kit. v0 writes Next.js with Tailwind and shadcn/ui, and its docs say it's "specifically trained on the default implementations of the shadcn/ui components". Lovable sites are styled with Tailwind too, and Lucide is shadcn/ui's default icon library, so the cards, badges and icons come from the same parts.
  • The brief. A builder writes from what you typed, so positioning, EEAT and a memorable idea only show up when you put them in. In a separate test of four chatbots, found in our part 2, all 12 heroes led with the brief's first selling point.

One brief, two builds

We gave Claude Code (no project files or settings) a short brief for a made-up bookkeeping startup called Tallyfarm. It chose cream, deep green and gold, a Fraunces headline and coded product UI, and scored 67, above all but 3 of the 72 showcase sites. It still scored 2 out of 10 on EEAT and 1 on site depth, as a single page with no named team, customers or contact.

Claude Code's Tallyfarm hero

Landing Page Gods (our product) built this hero from the same brief.

Landing Page Gods' Tallyfarm hero from the same brief

Part 2 of this series (coming soon) makes 8 code fixes to the Claude Code page and re-grades it.

Grade your own site

The landing page analyzer assesses any URL for free. The "quick check" does not require a signup, and the full 14-criterion grade is available with a free account.

Graded 21 November 2026. The defaults are from shadcn/ui's CLI 4.23.0 source, Tailwind's source and v0's docs.

Top comments (0)