Journal · Guide
Eight AI website builders, six checks: what leaves with you and what doesn't
Eight vibe-coded builders against six checks an owner can run. None passes all six on its own pages. The table says which thing each one is missing, so you know before it is your turn.
On this page ▾
Your website was built in a chat window. Someone typed what they wanted, a builder wrote the code, and a site appeared at a link. That is a vibe-coded site, and there is nothing wrong with it until the day you need something the chat cannot do: read the page without JavaScript, change a title without asking a bot, move the site, sell the business, stop paying.
This piece puts eight of those builders against six checks. The checks are the ones an owner can run, not a developer. Every cell below comes from the vendor's own pages, read on September 10, 2026, with prices re-read on September 23. None of the eight passes all six on its own documentation. Each is missing something the owner pays for later, and the point of the table is to know which thing before it is your turn.
The six checks
- Readable without JavaScript. When a machine asks for the page, does it get the words, or an empty shell that only fills in once a browser runs the code? Search engines and the AI assistants read the first kind reliably. The second kind depends on the vendor's promises.
- Per-page SEO. Can you set the title, description, canonical link, share image, schema, sitemap and robots file for each page yourself, without asking a chatbot to do it?
- Your own domain. Does the site connect with an A record and a CNAME at your registrar, leaving your email and nameservers alone?
- The exit. When you leave, what leaves with you: the code, the data, both, or neither?
- When you stop paying. What stays up, what goes dark, and whose badge comes back?
- The price. The number after the asterisk, not the number on the card.
The table
Pass means the vendor's page says so. Fail means the vendor's page says the opposite. Unverified means the vendor does not say, which is not the same as no.
| Builder | 1 Readable | 2 Per-page SEO | 3 Own domain | 4 The exit | 5 Stop paying | 6 Price |
|---|---|---|---|---|---|---|
| Lovable | Pass for projects made after May 13, 2026; older ones are shells except to verified crawlers | Ask in chat; no fields; canonical and schema unmentioned | Pass: A record plus TXT, DNS stays at your registrar | Code yes (ZIP, GitHub); Cloud database, auth and storage no | Site stays up; badge returns on next publish | $25/mo, $21 annual |
| Emergent | Unverified | Unverified | Pass: two A records plus CNAME, no nameserver change | GitHub only; no ZIP; database export undocumented | Stays up only while 50 credits are in the account | $20/mo, $17 annual; 50 credits per deployed app per month |
| Bolt | Fail by default; SEO Boost pre-renders on paid | No fields; SEO Boost only | Pass: CNAME and ALIAS, no IP published | Code yes (ZIP, GitHub); Bolt database stays unless claimed into Supabase | Stays up; cannot republish or reconnect a domain on Free | $25/mo; 28% off yearly |
| v0 | Pass: Next.js, server-rendered by default | Code, written on request; no product fields | Pass: through your own Vercel project | Code yes (ZIP, GitHub), deploys to your own Vercel account | Vercel hosts it; v0 lapse unverified | $0 plus Vercel Pro $20/mo for commercial use; v0 Plus $30/mo |
| Replit | Unverified; depends on what the agent built | Ask the agent; no settings | Pass: A record plus TXT per hostname | Code yes (ZIP, GitHub); secrets, database data and deploy settings no | Cancel and custom domains are removed; Starter apps go down after 30 days | $20/mo, $18 annual, plus usage |
| Base44 | Fail: client-rendered; crawlers get a snapshot | Pass: per-page fields, schema, sitemap, robots | Pass: ALIAS or A record; MX and TXT untouched | Front end only; ejected project has an empty database | Apps stay live; paid features fall away | $16/mo billed annually; monthly price not shown |
| Firebase Studio | Pass: Next.js on App Hosting | Code, via the agent; unverified as a feature | Pass: A record plus TXT | Code yes (ZIP, GitHub); Firestore data and Auth users no | Studio shuts down March 22, 2027; deployed apps keep running | $0; closed to new users since June 22, 2026 |
| Anything | Unverified | Ask the agent; site-level fields in v1 only | Pass: single A record plus CNAME on Vercel DNS | Code yes (ZIP); database yes as a SQL dump; GitHub sync "coming soon" | Stays published; branding returns; custom domain publishing disabled | $19/mo billed annually ($24 monthly on Sep 10) |
Lovable
Good at getting from a sentence to a working app in an afternoon, with a hosting story that does not embarrass anyone. The check it fails is the one that changed on a date. Apps created on or after May 13, 2026 use a server-rendered framework, and Lovable's own page says every request returns fully rendered HTML for humans and crawlers. Projects made before that date are single-page shells; a pre-rendered copy is served to verified crawlers only, and the docs say third-party scanners will see the regular shell. Titles and descriptions are generated per page and changed by asking Lovable in chat. The exit is a ZIP or a GitHub sync for the code. The Cloud database, auth and storage do not come with it, and the docs say there is no one-click migration to your own backend. Pro is $25 a month, $21 on the annual plan.
Emergent
Good at full-stack apps with a database behind them, from one prompt. We have moved sites off it, so this section is written from the exit side. Emergent's pages do not say whether a published page serves full HTML without JavaScript, and they do not document per-page title, description, canonical, schema, sitemap or robots. The exit is a push to GitHub; no ZIP is documented, and there is no page on exporting the managed database. The line that decides the check is on the help site: deployments stay active as long as 50 credits are in the account, even after cancelling. Each deployed app uses 50 credits a month. Standard is $20 a month, $17 annual, with 100 credits.
Bolt
Good at a fast front end, and the hosting has been Bolt's own since August 2025. The framework is chosen for you. Pages are client-rendered unless SEO Boost is on, which pre-renders HTML ahead of time and is disabled by default. There is no per-page meta panel. The exit is a ZIP that runs with two commands, and a two-way GitHub sync. The Bolt database stays behind unless you claim it into a Supabase account you own. Downgrade to Free and the site stays online, but you cannot republish or reconnect a domain, and the badge returns. Pro is $25 a month, 28 percent off on yearly billing.
v0
Closest to passing. v0 writes Next.js, and Next.js renders pages on the server by default, so the words are in the page before any script runs. The catch is the second check: there is no product screen for metadata, so titles, descriptions, sitemaps and robots are code you ask v0 to write. The exit is real: ZIP download, a GitHub branch that merges on publish, and a deployment into a Vercel project in your own account. Vercel's own line is that it does not own the code. The price is not v0's card. Vercel's Hobby plan is for personal, non-commercial use, so the commercial floor is Vercel Pro at $20 a month, with v0 itself free at that level.
Replit
Good at anything, which is the problem for this table. The agent picks the stack, so whether a given site serves full HTML depends on what it chose that day. SEO is done by asking the agent to add the tags and the files. The exit is a ZIP and a GitHub sync, minus secrets, database data and deployment settings. On the Starter plan a published app goes down after 30 days. Cancel a paid plan and the docs say custom domain connections are removed and published apps may go offline. Core is $20 a month, $18 annual, and deployments bill by usage after the included credits.
Base44
The best SEO controls of the eight: per-page title and description, a toggle to hide a page from the sitemap, platform schema you can replace with your own, and a sitemap and robots file generated on custom domains. It fails the first check on its own words. Base44 apps are client-side rendered, and the platform serves crawlers a rendered version. Browsers get the shell, machines get a snapshot. The exit is a command that downloads the front end and the entity schemas, and the docs say the ejected project has an empty database. Starter is $16 a month billed annually; the pricing page does not display a monthly figure.
Firebase Studio
Do not start here. Google closed Firebase Studio to new signups on June 22, 2026, and shuts it down on March 22, 2027, when remaining workspace data is deleted. Apps already deployed keep running, and the underlying Firebase services are not affected. For the record: it built Next.js and published to App Hosting, which renders on the server, so the readability check passed. Metadata was code. The exit was a ZIP and a push to GitHub, without Firestore data or Auth users. If a site of yours lives here, leave on your schedule, not theirs.
Anything
Formerly Create.xyz. Next.js on Vercel's infrastructure. Whether pages ship as full HTML is not stated. SEO in the current version is code the agent writes on request. The exit is the most complete of the eight: a ZIP of the code and a SQL dump of the database, which makes it the only one that hands the data back. GitHub sync is listed as coming soon. Downgrade and the site stays published with the branding back and custom-domain publishing off. Pro is $19 a month billed annually; on September 10 the monthly figure was $24.
Three patterns
The exit is where they diverge. Code leaves almost everywhere: seven of eight offer a ZIP or a repository. Data almost never does. One builder exports the database as a file. Two leave it undocumented. The rest say plainly that it stays.
Readability is a date, not a product. A Lovable site from April and a Lovable site from June carry the same logo and are different objects to a machine. Bolt is a shell unless a switch is on. Base44 is a shell with a snapshot for crawlers. The question is never which builder, it is which build.
The price on the card is not the price. Credits per deployed app. Usage billing after the included amount. A hosting plan underneath the builder plan. An annual figure shown where a monthly one is not.
If one of these built your site
Three things, in the order that finds the expensive problem first.
- Turn JavaScript off in the browser and load your home page. If the words are there, the first check passes.
- Look for the export. A ZIP, a GitHub button, a database download. Note which of the three you cannot find.
- Read the vendor's page on what happens when you cancel. The sentence you are looking for names your domain or your badge.
That takes ten minutes and tells you what you own.
If it fails a check
That is a fix with a price. One page, $750, from the Selection of services. Builds are quoted. The check itself you have already run.
Sources: vendor documentation and pricing pages, read September 10, 2026; prices re-read September 23, 2026. Where a vendor page was silent, the table says unverified.
Related Reading
AEO vs SEO vs GEO: 7 Differences That Decide Who Gets Named
SEO competes for a position on a page of links. AEO competes for a mention inside a single answer. Seven differences decide which one you're winning.
AI Watermarks: When They Matter, When They Don't
Every image you make with the big AI tools now carries an invisible serial number. Here's when that matters for a business, and when it doesn't.
Google Preferred Sources: How to Get Your Website Selected
What it is, why it matters, how to check your site, and exactly how to let readers choose you.