The design
Gap
Building with AI has a design context problem.
State of DESIGN.md 2026
Across AI coding agents and web builders, one design request repeats: make it look more premium, more consistent, and less generic.
Based on 64K+ getdesign.md onboarding responses from developers and vibe coders exploring DESIGN.md to improve how their AI-built apps look.
AI-assisted coding lets developers, founders, and vibe coders turn ideas into apps, landing pages, and product interfaces faster than ever.
The harder question comes after the first generated version: how do you make it look intentional, premium, and consistent enough to ship?
This report is based on onboarding responses from 64K+ users who joined getdesign.md in 11 weeks to improve the design of their AI-built apps and websites with DESIGN.md files.
How are builders building now?
AI coding assistants
Claude / ChatGPT / Cursor / Replit
AI app platforms
Lovable / Bolt / v0
No-code & CMS platforms
Framer / Webflow / Wix / Canva / WordPress / Squarespace / GoDaddy
For non-technical builders, dedicated web builders are no longer the obvious default. The ones who chose 'Other / not listed' typed in their own stack. Common write-ins included Figma, Google Stitch, OpenCode, and Shopify.
0 in 0
Builds on AI app platforms
0.0%
Still starts in no-code or CMS platforms
Non-technical builders are starting outside the tools made for them.
Developers always lived in code. The surprise is everyone else: web builders were invented so people without code could ship, yet in the AI era they too go straight to the model, letting a coding agent like Claude Code or Codex build the site.
Consistent, premium design without a designer.
DESIGN.md puts design context in the prompt, and a vibe coder gets what used to take a design hire: consistency, taste, and interfaces that do not look AI-generated.
Which builder platforms are leading?
Ranked by share of all builders, tile sizes eased so every tool stays readable · the 64.3% on AI coding assistants are excluded (see 01)
Lovable dominates inside AI app platforms, but AI app platforms are only 22% across all builder categories. The bigger market still starts outside dedicated app builders.
Share within category · 22% of everyone
0.0%
Lovable among all builders surveyed
0%
Vietnam's v0 share, against a flat 23% everywhere else
The tool race is not the design race.
Dedicated builders help people ship faster, but they do not automatically make the result feel premium, consistent, or distinct. The design gap remains after the site is built.
Bar: share on each tool within the market (sums to 100) · Right: market's share of all app builders · Europe is 48 countries combined, UK shown separately · 91 additional markets are included in the aggregate
WordPress still holds the field.
Even among builders who jumped to AI, it stays the second most-used tool across no-code and CMS, and the only legacy platform anywhere near the top.
Framer leads, but the race is tight: WordPress and Canva sit just behind. No single tool owns the category, and Canva outranking Webflow shows how far design tools have moved into building real sites.
Share within category · 9.1% of everyone
Asked when builders sign up for getdesign.md to bring a DESIGN.md into their workflow.
Improve how my site looks
Use my design as a brand source
Source: getdesign.md onboarding · n = 64K+
Most builders are not asking for a new site. They want the interface they already have to look better, or closer to a design they trust. DESIGN.md turns that visual intent into context an AI builder can follow.
The web's next bottleneck is taste, not code generation.
Everyone can ship now. Almost nobody is happy with how AI-generated UIs look, and they want to fix it themselves, without hiring a designer. That takes consistency and a design source their AI can actually follow.
Want to change how their AI-generated UI looks
Building is no longer the problem. Looking good is.
Just look betterMatch a designBrand source
The goal each tool's users picked when they came to getdesign.md · bars sum to 100%
AI-builder users are here for one thing: polish. Seven in ten Lovable and v0 users just want what they shipped to look better. Move to the older platforms and the ask broadens toward matching a specific design and building a brand source, with Squarespace users the most design-system minded of all.
The build barrier is gone.
The design barrier is not.
Asked only to people already building in a web builder or platform, not to those prompting in chat.
Keep their current builder
Leave and rebuild with Claude / Gemini / ChatGPT
Source: getdesign.md onboarding · base: 31% of everyone
"No, I don't want to keep paying for web builders and want to rebuild with Claude / Gemini / ChatGPT."
Plan to leave their builder for the model
Web builder loyalty is not guaranteed when AI assistants can rebuild the same site.
Share of each tool's users who chose to stop using it and rebuild with Claude, Gemini, or ChatGPT.
Legacy CMS and hosting tools are the most exposed. AI-native builders are stickier, but even there, roughly a third say they would move back to an AI assistant.
Source: getdesign.md onboarding · base: 31% of everyone
Top 16 markets by share of all responders, from anonymized GeoIP · Europe is 48 countries combined, UK shown separately
The remaining 14.3% is spread across 120 more countries.
The long tail is real: 120 more countries follow this list.
Source: getdesign.md onboarding GeoIP · n = 64K+ · 183 countries
The web is now built by whoever can describe it.
Some are seasoned developers, but most are vibe coders and non-technical builders who never wrote a line of code, all now shipping real products just by describing them. And they all reach the same wall: what the AI builds does not look right, and even the people happiest to hand the work to a model are no longer happy with the result.
DESIGN.md is the community's answer, a design source the AI actually follows, and it closes the gap between shipping and shipping something worth showing.
Methodology
- This report looks at why builders bring DESIGN.md into AI-built websites and apps, what tools they use, and where design context is still missing.
- The data comes from required getdesign.md onboarding responses from 64K+ users who joined over eleven weeks.
- Results are reported from aggregated onboarding data. Individual user details are not included in the published analysis.
- Cross-tabulated results based on fewer than 20 respondents are not reported separately.
- Country data comes from anonymized, aggregated GeoIP. No demographic questions were asked in this wave.