v0 in 2026: What It Builds, the Defaults It Starts From, and How to Fix the Output

v0's own FAQ says it plainly: "output that you receive may be the same or similar to other users' output." That sentence, from the page Vercel keeps at v0.app/docs/faqs, explains most of what follows. v0 is fast because every chat starts from the same well-built scaffold. Its pages look alike for the same reason.

This article is built from v0's documentation, Vercel's own announcements and the source code of the stack v0 generates, all read on 17 October 2026. We did not build a project with v0 for it, and it reports no project metrics: where a claim comes from a document, the document is named; where it is our judgment, it reads as one.

The short version

• v0 is a Next.js generator before it is anything else. Its FAQ lists the stack it uses: Next.js, React, TypeScript, Tailwind CSS and shadcn/ui. Its design-system guide adds that "new v0 chats start with a Next.js app that uses shadcn/ui and Tailwind CSS."
• The sameness is structural. One starter for every chat, a model that Vercel's docs say "is specifically trained on the default implementations of the shadcn/ui components and may struggle with any customizations," and a FAQ that warns your output may match someone else's.
• That makes v0 a good fit where convention is a feature: internal tools, admin panels, prototypes, demos, first drafts you intend to rewrite.
• It is a poor fit, unedited, for a brand's public face. Nothing in the default says who you are.
• The defaults can be changed, and v0 now ships tools for it: Design Mode for visual edits, Design Systems 2.0 to load your own components and tokens, and a team default design system on paid plans.
• You can leave Vercel, but the path of least resistance runs through it. The FAQ says v0's code "is optimized for the Vercel platform" and that you "can still export code and deploy elsewhere."

What v0 is in October 2026

v0 has changed shape three times, and each change is dated.

The product a new user meets today is therefore a chat, a preview that runs the real app, an editor and a Publish button. v0.dev redirects to v0.app. The FAQ describes the scope as "web apps, dashboards, e-commerce, SaaS, AI-powered apps, data visualizations, 3D projects, and AI agents," and says v0 follows AI SDK 6 patterns when it builds AI features.

What every chat starts from matters more than the feature list. The Design Systems 2.0 page spells it out: "New v0 chats start with a Next.js app that uses shadcn/ui and Tailwind CSS." The older registry guide is just as direct: "v0 uses Shadcn/ui as its default component system," and, a few lines later, "v0 automatically uses shadcn/ui components in your generation."

Where v0 sits among the other builders

v0 does not try to win on language range. Its pitch, in its own FAQ, is "best-in-class Next.js expertise backed by Vercel's enterprise infrastructure." If your stack is already Next.js, Tailwind and shadcn/ui on Vercel, v0 is the shortest path from a sentence to a running app. If it is not, you are importing a stack along with the code.

Where v0's defaults are an asset

These are the projects where a shared look costs nothing and the speed is the whole point. They follow from what v0 generates, not from a tally of projects.

Internal tools and admin panels

Billing dashboards, content back offices, support consoles, CRUD screens behind a login. Convention helps here: staff already know what a table, a dialog and a dropdown look like, and shadcn/ui builds those on Radix primitives, which handle keyboard navigation and focus for you. Nobody needs the billing dashboard to carry a brand. v0's own FAQ lists "internal tools" among the things it builds.

A first draft you intend to rewrite

A generated page is easier to argue with than a blank file. "The hero should not be centered" is a faster decision than "design a hero." The discipline is to actually rewrite: keep the routing, the data wiring and the accessible primitives, replace the surface. If the plan is to recolour the draft and ship it, the result will look like the draft.

Prototypes and demos

Hackathon builds, customer-validation prototypes, a clickable demo for one meeting. v0's FAQ names "sales engineers produce custom demos instantly" as a use. The audience is small, the artifact is short-lived, and a convincing working page is worth more than a distinctive one. The risk arrives the day the prototype gets a name, a logo and a domain: at that point it is a product, and its defaults become its identity.

Learning Tailwind and React by reading generated code

v0 writes idiomatic Tailwind and standard shadcn/ui components, so its output doubles as reading material for someone new to the stack. Prompt, read the diff in the editor, change it, prompt again.

Where the defaults cost you

A brand's marketing site

If the business depends on being recognized, the default is the problem. A page assembled from the same starter, the same component shapes and the same unprompted layout choices as thousands of others gives a visitor nothing to remember. v0 can produce something else, but only if you tell it what, and keep telling it. Its docs say as much in their prompting guide: "The more specific your prompt, the better the results."

Agency and studio portfolios

A portfolio is the proof of taste. Shipping one on generator defaults sends the opposite message to the one the agency is selling. Generate the structure if you like; the visual decisions have to be visibly yours.

Sites that live on search

Google's stated position, in its guidance on AI-generated content of 8 February 2023, is that it rewards high-quality content "however it is produced," and that using automation, AI included, "with the primary purpose of manipulating ranking in search results" violates its spam policies. The tool is not the issue. A page that adds nothing its neighbors do not already have is, and default layouts with default copy are the fastest way to build one.

Editorial sites

Voice lives in the layout as much as in the words: measure, rhythm, how images sit in the column. A component kit built for SaaS dashboards has none of that. If the product is writing, design the reading page yourself.

Multilingual sites

v0's FAQ points to Next.js internationalisation patterns and recommends next-intl. That handles routing and strings. It does not handle layout: French and German strings usually run longer than English ones, and button widths, headings and nav items designed around English copy break first. Test every locale at the narrowest width before launch.

Accessibility-critical projects

shadcn/ui builds on Radix UI primitives, which handle focus management and keyboard interaction. Color is where to check. The default muted text token, oklch(0.556 0 0), works out at about 4.74:1 on white, above the 4.5:1 that WCAG 2.2 asks of body text. Accents chosen later are another matter: white text on Tailwind's blue-500 (#3b82f6) is 3.68:1 and fails; on blue-600 (#2563eb) it is 5.17:1 and passes. Those ratios are computed with the WCAG formula; run the same check on whatever palette your project ends up with.

Why v0 pages look alike

Four documented causes, none of them a bug.

• One starter. Every chat begins from the same Next.js, shadcn/ui and Tailwind app unless a design system skill replaces it.
• A model trained on the defaults. The registry guide warns that v0 "is specifically trained on the default implementations of the shadcn/ui components and may struggle with any customizations." Customized primitives fight the training.
• A shared corpus. The FAQ says v0 was trained on data sets that "primarily cover areas related to web development, particularly with React, Next.js, and modern web technologies." Everyone draws from the same well.
• Similar prompts. Most people describe a landing page the same way. The FAQ's own warning about similar output is the honest summary.

The broader mechanism, across tools, is in why every AI-generated website looks the same.

The v0 fingerprint: 14 tells and where each comes from

Some of these are documented defaults you can read in source code. Others are habits common to AI-built landing pages, with no document behind them. The table says which is which.

The first seven are the scaffold. The last seven are what the model adds when the prompt leaves space. Neither half is wrong in isolation; the cluster is the signature. The free scan checks a live page for the most visible of them: blue utility families, the blue-to-purple gradient, Inter or Geist as the page font, the three-column grid, fade-up motion and frosted glass.

For sites already shipped on v0, Lovable or Bolt, the de-AI guide walks the clean-up in order.

The 14 fixes to apply on every v0 output

Ranked roughly from the biggest visible change for the least work. Effort is a judgment, not a measurement.

If you do not have time for at least the first seven, that is a signal the project needs a different starting point, not a faster generator.

Fix 1: Choose your typefaces

Before: whatever layout.tsx loads, often the create-next-app default (Geist) or Inter.

After: a pair you chose for a reason: a serif or display face for headings, a text face for body. The specific choice matters less than the fact of choosing; avoid swapping one default for another (Geist is now its own tell, see Geist, the new Inter). Load both with next/font so they are self-hosted, and apply the heading face to h1 through h4, not just the h1.

Fix 2: Replace the accent

Before: an accent from Tailwind's blue or indigo families, or a blue-to-purple gradient on buttons and headlines.

After: two values of your own, set once as CSS variables and wired into the theme:

:root {
  --brand: oklch(0.52 0.13 245);
  --brand-accent: oklch(0.66 0.14 55);
}

Then point shadcn's --primary at it, so every component follows. Check contrast for white text on the new accent before you ship it.

Fix 3: Restyle the card primitive

Before: import { Card, CardHeader, CardContent } from '@/components/ui/card', rendered with shadcn's rounded border and soft shadow.

After: edit components/ui/card.tsx itself. shadcn/ui hands you the source, so this is one file, and every card on the site changes with it: borderless on a tinted surface, a single hairline on top, asymmetric padding, whatever suits the content. v0's docs warn that it may struggle with customized primitives, so make this change after the generation phase, not before.

Fix 4: Rebuild the hero

Before: a centered headline, a centered subhead, two centered buttons.

After: a composition that responds to the product. A left-aligned split with a real screenshot. A single block of strong color with the headline reversed out. A headline anchored to a grid line with the image bleeding off the edge. For 21 worked compositions, see the hero section catalog.

Fix 5: Break the three-card row

Before: three equal cards, icon on top, two lines of gray text.

After: one feature given the space it deserves and two smaller ones beside it, a definition list, a single table, or a sequence that follows how the product is used. If the three features are not equally important, the layout should not say they are. The argument is laid out in the card-grid problem.

Fix 6: Pick another icon set

Before: every icon from lucide-react.

After: one other open set used consistently (Phosphor, Tabler, Iconoir, Material Symbols), or icons drawn for the product. Do not mix libraries; two styles of stroke on one page look worse than the default.

Fix 7: Rewrite the calls to action

Before: "Get Started" and "Learn More."

After: the verb your product actually performs. "Run your first audit." "Book the 15-minute call." "See a sample report." It costs minutes and changes how specific the whole page feels.

Fix 8: Write a real 404

Before: the framework's stock not-found page.

After: a not-found.tsx in your own voice, with links to the three pages people most often want.

Fix 9: Make an Open Graph image

Before: no og:image, so every share shows a bare link or a generic card.

After: an opengraph-image.tsx at the root (Next.js renders it at build time with ImageResponse), and one per article or key page if you publish content.

Fix 10: Make sections asymmetric

Before: every section centered inside the same max-width container.

After: alternate alignments, let some sections use part of the width, let an image break the grid. This is a design pass over the whole page rather than a substitution, which is why it is the most expensive fix and the one that does the most.

Fix 11: Put pricing in a real table

Before: pricing as nested divs on a three-column grid.

After: a <table> with <thead>, <th scope> and rows of features. Screen readers announce it properly, it compares plans better, and it does not look like the default.

Fix 12: Write the microcopy

Before: "Loading...", "An error occurred.", "No results found."

After: copy that says what is happening and what to do next, in the product's tone. "Fetching last month's invoices." "That card was declined. Try another or contact your bank."

Fix 13: Replace the stock dialog

Before: shadcn's Dialog, centered over a bg-black/50 overlay.

After: a side panel, a bottom sheet, or an inline expanding section where a modal was never needed. Keep Radix underneath if you want its focus handling; change the shape.

Fix 14: Remove or replace fade-up motion

Before: every section fades in and rises as it enters the viewport.

After: no entrance animation (often the right answer), or one motion that belongs to the content. Fade-up on everything is covered in the motion slop piece.

For the prompt side, so the next generation starts closer, see the anti-slop prompt template. v0's Design Systems 2.0 is the durable version of the same idea: import your components and tokens once, set the design system as your team's default, and new chats start from your starter instead of v0's.

v0 next to Bolt, Lovable and Replit Agent

The four tools overlap more than their marketing suggests. What actually separates them is documented.

Sources: v0 docs, Bolt help center, Lovable docs, Replit docs, all read on 17 October 2026.

The practical reading: v0 if your stack is Next.js on Vercel and you want code you can take into your own repository through pull requests; Bolt if you want any JavaScript framework running in the browser tab; Lovable if you want hosting, a database and auth without touching a terminal; Replit if the project is not JavaScript. None of them changes the design problem: all four start from defaults, and all four need the same post-generation pass. The Bolt and Lovable write-ups are in the Bolt.new review and the Lovable review.

Pricing in 2026

What each plan costs and includes, as listed on v0.app/pricing on 17 October 2026. Prices are in US dollars, before tax.

v0 plans

New users also get $5 of one-time credits. v0's pricing documentation adds that unused monthly credits roll over to the next billing cycle and expire after 65 days, and that a $20 Premium plan still exists for older accounts but is being sunset. Credits are spent at per-model token rates, in dollars per million tokens:

The free plan is enough to see the defaults this article describes, not enough to build a project. On Plus, the model you pick decides how far the $30 goes: a million output tokens costs $50 on v0 Max Fast and $1.20 on v0 Mini, about forty times more.

Versus Bolt

Sources: v0.app/pricing and bolt.new/pricing, both read on 17 October 2026. Lovable, Replit and Cursor publish their current plans on their own pricing pages; check them on the day you buy rather than trusting any article's table, this one included.

The Vercel question

Is v0 an on-ramp to Vercel? Its documentation answers most of it.

On the Vercel side of the ledger:

• The FAQ: "v0's code is optimized for the Vercel platform."
• Publish deploys to the connected Vercel project. For the full experience, the FAQ says, "your GitHub repo should be connected to a Vercel project."
• Environment variables are managed in Vercel, and v0 uses the Vercel AI Gateway by default when it adds AI features.

On the other side:

• The output is a standard Next.js app. The FAQ: "You can still export code and deploy elsewhere."
• The GitHub integration works through branches and pull requests on your own repository, and v0 never pushes to main. Your repository stays the source of truth.
• Vercel states in the FAQ that it "doesn't own the code generated based on your queries and prompts."

So: v0 is a reasonable choice wherever Vercel is a reasonable host. Where it is not (data-residency rules, an existing platform, cost at scale), generate in v0, merge to your repository, and deploy from there. Platform features such as image optimization and edge middleware will need their equivalents on the new host; list the ones your app uses before you move.

Verdict: when to use v0, when to skip it

Use v0 when:

• The project is an internal tool, an admin panel or a dashboard.
• Your stack is Next.js, Tailwind and shadcn/ui anyway.
• You need a prototype or a demo this week.
• You want a first draft you will rewrite by hand.
• You are learning the stack and want readable examples.

Skip it, or plan a full design pass, when:

• The site is the public face of a brand.
• Design taste is what you sell.
• Organic search is a main acquisition channel and the page has nothing of its own to add yet.
• The layout has to carry an editorial voice.
• Non-English locales are primary.
• WCAG compliance is contractual and nobody will audit the result.

Use v0 with the 14 fixes when the project mostly fits but has one or two of those risks, and someone on the team has the time and the eye to make the visual decisions.

For the wider monoculture v0 accelerates, see the shadcn/ui monoculture.

FAQ

Is v0 worth $30 a month? For someone generating Next.js work every week, it can be. Plus costs $30 per user per month and includes $30 of credits (v0.app/pricing, 17 October 2026). Watch the model you pick: credits go far faster on v0 Max than on v0 Mini. If you use it occasionally, the free plan's seven messages a day show you what you would be paying for.

Can I use v0 output on non-Vercel hosting? Yes. The FAQ says you "can still export code and deploy elsewhere." The output is a standard Next.js app; deploy it anywhere that runs Next.js, and replace the Vercel-specific features your app relies on.

Does v0 produce the same page for everyone? Not the same page, but the same starting point and similar habits. Its FAQ warns that output "may be the same or similar to other users' output." Specific prompts, a design system skill, or a pass with the 14 fixes are how you move away from it.

Can I make v0 stop using shadcn/ui? The registry guide says v0 "is still capable of generating code using other component libraries, styling tools, and even frameworks," but defaults to shadcn/ui and may struggle with customized versions of its components. The durable route is Design Systems 2.0: import your own components and tokens so new chats start from your starter.

Does v0 work for non-React frameworks? React and Next.js are the default and what the starter, the docs and the training focus on. Other frameworks are possible on request; if your project is Vue, Svelte or Astro, a tool built around your framework will fight you less.

Is v0 better than Cursor or Claude Code? Different jobs. v0 generates and runs an app from a prompt, with a preview and a deploy path. Cursor and Claude Code work inside an existing codebase on your machine. v0's FAQ even documents the hand-off: connect the v0 project and your editor to the same GitHub repository and pull each other's changes.

What happens to my prompts? The FAQ says v0 may use prompts and content to improve its models unless you are on Enterprise or have opted out of training in your Vercel settings; the Business plan has the opt-out on by default. Prompts that describe unannounced products deserve the same care as any other text you send to a third party.

Can I use v0 output commercially? Vercel says it does not own the generated code, and that the output may be similar to other users' output or to third-party material, so you are responsible for checking it. The stack itself is open source: shadcn/ui, Tailwind CSS and Next.js are MIT-licensed, Lucide is ISC.

Does v0 generate accessible code? It inherits Radix's keyboard and focus handling through shadcn/ui, which is a good floor. Color contrast is the part to verify, especially for any accent you add; see the ratios in the accessibility section above.

Where are v0 updates announced? On the v0 changelog and in the v0 category of Vercel's blog.

Glossary

v0: Vercel's AI app builder. Public beta in October 2023 at v0.dev; served from v0.app since August 2025.

shadcn/ui: A set of React components built on Radix UI and Tailwind CSS, distributed as source you copy into your project. v0's default component system.

Radix UI: Unstyled, accessible component primitives underneath shadcn/ui.

Tailwind CSS: The utility-first CSS framework v0 styles with.

Lucide: The icon library shadcn/ui's components import.

Next.js: The React framework, maintained by Vercel, that every v0 chat starts from.

Design Mode: v0's visual editor: select an element in the live preview, change its styles, and apply the edit back to the code as a new version.

Design Systems 2.0: v0's way of loading your own components, tokens and starter app, saved as a skill that new chats can use.

Vercel Sandbox: The isolated virtual machine that runs v0's previews since February 2026.

Fingerprint: The cluster of defaults that makes a generated page recognizable: the 14 tells above.

Sources

All read on 17 October 2026 unless dated otherwise.

• v0 documentation: What is v0?, FAQs (including the "New v0 Update" of 4 February 2026), Design Systems 2.0, Design systems (registry guide), Design mode, Text prompting, Pricing.
• v0.app/pricing and bolt.new/pricing.
• Vercel, "v0.dev is now v0.app", 11 August 2025.
• Tim Anderson, "Build a user interface using AI prompts: Vercel v0 development kit now in public beta", DevClass, 10 October 2023.
• shadcn/ui source for Card and Button, and the theming docs.
• The create-next-app default layout.
• Google Search Central, "Google Search's guidance about AI-generated content", 8 February 2023.
• W3C, WCAG 2.2, success criterion 1.4.3.
• Bolt help center, Lovable docs, Replit docs.

No affiliate links, no sponsorship, no relationship with Vercel.

Revision note

Rewritten on 17 October 2026. An earlier version presented itself as a twelve-month field report built on about forty v0 projects and some three hundred audited sites, with timings, scores and client outcomes. None of that was documented, so all of it has been removed. The article now rests on v0's documentation and the dated sources above. The plans, prices and credit rates, corrected the same day against v0's pricing page, are unchanged.