v0 vs Lovable vs Bolt vs Framer AI: Which AI Website Builder Fights Its Own Defaults
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.
Does your site carry these defaults? Paste its address: the check takes a few seconds, free, no account.
On this page
Give any AI website builder a one-line brief, "a landing page for a SaaS that helps teams track project deadlines," and it has to fill every gap you left: the typeface, the palette, the layout, the copy. What fills those gaps is the tool's default, and the default is the slop.
So the question worth answering is not which builder produces the prettiest page from a blank prompt once. It is what each builder does, by design, when your prompt leaves the look open, and how easy it makes it to replace what it chose. Those are documented, so this comparison is built from v0's, Lovable's, Bolt's and Framer's own documentation, read on 17 October 2026.
An earlier version of this page described the four pages each tool "returned" for that brief, down to their headlines and hex codes. We hold no record of those runs, so that account has been removed. The ranking below is an editorial judgment from documented behavior, and says so.
What a vague brief tends to produce
Across generated landing pages, the same handful of choices keeps coming back when nobody asks for anything else:
- A centered hero: headline, one-line subhead, a solid primary button and an outline button beside it.
- A three-column feature grid, each card with an icon in a rounded square, a bold title and two lines of gray text.
- Lucide icons.
- Blue, indigo or a blue-to-purple gradient somewhere, usually the hero or the main button.
- Inter or Geist as the only typeface.
- Three pricing tiers, the middle one raised with a "Most Popular" badge.
- A four-column footer: Product, Company, Resources, Legal.
- Social proof that nobody supplied: invented testimonials, round-number stats.
Strip the brand names off, and the hero of such a page reduces to something like this (an illustration of the pattern, not a captured output):
<section className="max-w-6xl mx-auto px-6 py-24 text-center">
<h1 className="text-5xl font-bold tracking-tight">
Never miss a deadline again
</h1>
<p className="mt-4 text-lg text-muted-foreground">
The all-in-one platform for teams that move fast.
</p>
<div className="mt-8 flex justify-center gap-4">
<Button>Get Started</Button>
<Button variant="outline">Learn More</Button>
</div>
</section>The mechanism is covered in why every AI-generated website looks the same: absent a strong brief, a model regresses to the middle of what it has seen. The four tools differ in how they let you, or make you, get away from that middle.
v0 (Vercel)
What it starts from. Its docs are explicit: "New v0 chats start with a Next.js app that uses shadcn/ui and Tailwind CSS," and v0 "automatically uses shadcn/ui components in your generation." The shadcn/ui shapes come with it: cards drawn as rounded-xl border bg-card shadow-sm, buttons at h-9 rounded-md, pill badges, Lucide icons. shadcn's own theme is grayscale, so a blue accent, when it appears, is the model's choice or yours, not the library's.
What it does when the look is open. Nothing special: it generates. The registry guide adds that v0 "is specifically trained on the default implementations of the shadcn/ui components and may struggle with any customizations."
How you override it. Design Systems 2.0 imports your components, tokens and a starter app, and on paid plans a team owner can make one the default for new chats. Design Mode lets you select an element in the live preview and change its styles. The code goes to your GitHub repository through pull requests, so a developer can finish the job anywhere.
Lock-in. Low for the code (a standard Next.js app, which Vercel says it does not own), higher for the workflow, which publishes to Vercel.
The full write-up: v0 in 2026.
Lovable
What it starts from. One fixed stack: TanStack Start (React, Tailwind) for apps created since 13 May 2026, React and Vite before that, PostgreSQL through Lovable Cloud or Supabase. "Lovable does not offer a choice of framework or database."
What it does when the look is open. This is where Lovable differs from the others. Its design guidance documentation says that when "your prompt leaves the look open," Lovable shows design questions before it builds: a palette, a typeface pairing and a layout, each with up to four options, from "asymmetric" and "broken grid" to "card grid." It preselects the first option in each. Select Skip all and "Lovable then chooses reasonable defaults."
How you override it. Answer the questions or name the look in the prompt (Lovable then "builds directly"), keep rules in project or workspace knowledge, connect a design system, or import from Figma.
Lock-in. Hosting and backend are Lovable's by default; the code leaves through Git sync on any plan or a zip on paid plans. The Edit with Lovable badge stays on free-plan sites.
The full write-up: Lovable in 2026.
Bolt.new (StackBlitz)
What it starts from. Any JavaScript framework that runs on the frontend, with Node.js for the backend, running in WebContainers in your browser tab. Bolt does not publish a fixed visual default.
What it does when the look is open. It builds, and its own blog is candid about the result: "if you keep things vague and surface-level, you'll end up with a workable but cookie-cutter app" (Bolt blog, 25 September 2025).
How you override it. Project knowledge (persistent across chats), an agents.md file, Figma frames as design reference, visual edits in the preview, and, on Teams plans, an uploaded design system.
Lock-in. Low: download the project or push it to GitHub. Free-plan sites carry a "Made in Bolt" badge.
The full write-up: the Bolt.new review.
Framer AI
What it starts from. Not code: a page on Framer's canvas. Framer describes its AI as an agent that "creates editable pages, sections, copy, and visuals directly in your project," which you keep refining "in the conversation, or take over on the canvas at any point" (framer.com/ai).
What it does when the look is open. It generates a design on the canvas. The difference from the other three is what happens next: every typeface, color, spacing and breakpoint is a visual control rather than a line of code, so changing the default does not require a developer.
How you override it. Edit the canvas directly, or keep instructing the agent. Framer also documents reusable "skills" for design systems and writing.
Lock-in. The highest of the four. Framer's own comparison with Webflow lists code export as a Webflow feature, not a Framer one; the site lives on Framer's hosting. The "Made in Framer" badge disappears once you connect a custom domain or move to a paid plan (Framer help).
The full write-up: the Framer AI review.
The comparison table
| Tool | Starts from | When the look is left open | Main override | Code out | Free-plan branding |
|---|---|---|---|---|---|
| Lovable | Fixed stack: TanStack Start, React, Tailwind | Asks about palette, type and layout first | Design answers, knowledge, design systems | Git sync, zip on paid plans | Edit with Lovable badge |
| Framer AI | A page on Framer's canvas | Generates a design you edit visually | Direct canvas editing | No code export | Made in Framer badge |
| v0 | Next.js, shadcn/ui, Tailwind | Generates from its starter | Design Systems 2.0, Design Mode | GitHub pull requests | Not documented |
| Bolt | The JavaScript framework you name | Generates; "cookie-cutter" if vague | Project knowledge, agents.md, Teams design systems | Zip, GitHub | Made in Bolt badge |
Sources: v0 docs, Lovable docs, Bolt help center, framer.com, read on 17 October 2026.
Ranking: which builder does most to keep its defaults off your page
An editorial judgment from the documentation above, not a measurement of output.
1. Lovable. The only one of the four that documents asking about the look before it builds when the prompt does not say. Answering those questions honestly (and not accepting the preselected first options) is the cheapest anti-slop step any of these tools offers. Its weakness is that the stack is fixed and the hosting is Lovable's.
2. Framer AI. It does not ask, but what it produces is a canvas, and on a canvas every default is one click from being replaced, by a designer, without code. The price is lock-in: no code export.
3. v0. The strongest single default of the four (one starter, a model trained on shadcn/ui's stock components) and the strongest escape route for teams willing to set one up: Design Systems 2.0 and a team default. Without that, a vague prompt gets the house style.
4. Bolt. The most open stack and the most honest vendor about the problem, but nothing in its documented flow intercepts a vague prompt, and the design-system feature is limited to Teams plans. A precise first prompt and an agents.md file are on you.
What all four share
None of these tools is the problem; the unsteered default is. Every one of them can produce something distinct when the brief is distinct: a specific reference ("a 1970s Swiss train timetable," not "modern"), a banned list ("no gradients, no Inter, no centered hero"), a real brand color, and actual product copy instead of invented testimonials.
If you have already shipped a page from one of them, the defaults are removable: see de-AI-ing a Lovable, v0 or Bolt site, and run the free scan on the live page to see which of the common signatures it carries.
So the practical ranking is: whichever tool puts your decisions where it will read them. Framer if you want to art-direct on a canvas and accept its hosting. v0 if you write code and will set up a design system. Bolt if you want an open JavaScript project and will write a precise brief. Lovable if you will answer its design questions instead of skipping them, and delete any testimonial you did not write before you publish.
Revision note
Rewritten on 17 October 2026. The earlier version reported the result of typing one brief into the four tools "on a Tuesday afternoon," with each tool's headline, colors and invented testimonials. There is no record of those runs, so they have been removed. The comparison now rests on each tool's documentation, and the ranking is labeled as an editorial judgment.
Scan a live page
Eight AI defaults, the evidence for each, and a fix kit. Free, no account.
Read the whole site, with a fix prompt per default
Get told when the CLI ships