Get your AI-built app seen by Google and ChatGPT
The builder is not the page a crawler reads. What matters is the HTML the host sends. Some of these apps send a finished document. A lot of them send a shell and fill it in the browser.
Check what AI crawlers see on one of your URLs. The result is the HTML before JavaScript runs, plus any robots.txt block for GPTBot, ClaudeBot, PerplexityBot and Googlebot.
Lovable, split by when the project was created
Lovable's May 13, 2026 post says new projects from that date are server-rendered with TanStack Start. The server runs the React tree and sends HTML on the first request. Routes that do not depend on per-request data can be prerendered at deploy. Those projects are not the empty-shell case, as long as the host actually runs the server. A static host that only uploads the client bundle throws that away.
The same post says existing projects keep working as before, and that Lovable prerenders them for a verified crawler: a headless browser renders the page and the crawler gets that HTML. Human visitors still get the SPA. Their upgrade doc is blunter about the limit. An older React and Vite app sends a mostly empty page, and tools outside the verified crawler list see very little. TanStack Start, after you upgrade, renders for every request, not only for the bots Lovable recognizes.
Lovable also published early results on their hosting: a 2.9 percent increase in organic search traffic for apps using pre-rendering, and a 98.5 percent increase in traffic from AI tools, in the window they measured. Those are Lovable's figures for Lovable-hosted apps. They are not a promise about a project you exported and deployed somewhere else. If you exported an older Vite app, check the HTML on the host you actually use.
Bolt
Bolt projects are often a Vite React app, because that is the stack people ask it for and the one a Vite starter produces. A Vite React app sends a shell unless the project was built as a server-rendered app. This page does not claim every Bolt project is a SPA. Open the published URL and view source. If the words are missing, crawlers that do not run JavaScript have nothing to keep.
Asking the builder to add a title tag does not fix an empty body. The title can be in the shell while the page content still loads in an effect. The crawler check separates those: a found title with a tiny word count is a shell that remembered its head.
The response GPTBot keeps
<html>
<body>
<div id="root"></div>
<script type="module" src="/assets/index.js"></script>
</body>
</html>No words in the body
The page after JavaScript runs
<div id="root">
<nav><a href="/products">Products</a></nav>
<main>
<h1>The page title</h1>
<p>The words you want cited.</p>
</main>
</div>Only a crawler that runs JavaScript gets this
Replit
Replit Agent will build whatever stack the prompt led it to. A Vite or client-rendered React app has the same empty HTML as any other SPA. A Next app, a Nuxt app, or a server that renders templates can already be readable. There is no single Replit rendering mode to quote, so the check is the published HTML, not the editor.
Preview URLs and production URLs can differ. A firewall on the production host can challenge GPTBot with 429 while the preview looks fine in your browser. Run the checker on the URL you want indexed.
What to do after the check
If the word count is healthy, leave the app alone. If Lovable offers an upgrade to TanStack Start and the project is one of the older Vite apps, that upgrade is the fix Lovable documents: HTML for every visitor, not only for verified crawlers. If you cannot upgrade, or the app is a Bolt or Replit client bundle you are not rewriting this week, a prerender service renders the URL for crawlers.
Pagerender does that on your domain for a week before you decide. It does not replace a robots.txt block or a firewall challenge. Those show up as their own rows in the checker. Fix the row that is actually failing.
What a crawler request passes through
Firewall
The host answers the bot
Stops on: 401, 403, 429 or 503robots.txt
The bot's token is allowed
Stops on: A matching DisallowHTML
Words arrive before JavaScript
Stops on: Under 60 wordsnoindex
The page asks to be indexed
Stops on: A noindex tagRead and indexed
Search can rank it and AI answers can cite it
Exporting the project
Lovable's prerender for verified crawlers is a feature of Lovable hosting, described in their upgrade doc. Download the repo and deploy it as static files, and that prerender step is gone. You then have the Vite shell, unless you upgraded to TanStack Start and the new host runs the server. Bolt and Replit exports work the same way: the host you deploy to is the thing a crawler hits.
Custom domains do not change the HTML. They change the name. Pointing a domain at a client bundle still serves the bundle. Check the domain you will put in Search Console, with JavaScript off.
Social previews fail for the same reason. Slack, iMessage and X read the HTML. A title set in a useEffect never arrives. Lovable calls this out as a reason for server rendering: the first response has the tags. If a link unfurls as a blank card, the crawler check will show a missing description or a tiny word count on that same URL.
Prompts that say "add SEO" usually add a title, a description and a sitemap. Those files do not render the body. A sitemap of empty URLs gives a crawler a list of shells. Ask the builder for server rendering, or for the text to be in the HTML, and then verify with view source. If the builder cannot do that inside the tool, export and put a prerender in front, or move the public pages to a framework that renders them.
Replit's preview is a fine place to click around and a bad place to judge crawlers. The production deploy can add a platform firewall. Bolt's deploy to Netlify can use Netlify's prerender extension, which their docs say serves rendered HTML to crawlers from a function in your account, with no extra charge beyond function billing, on Node 20 or later. That is a real fix for a Vite app that already lives on Netlify. It does nothing for a project you host somewhere else.
Lovable's May 13, 2026 change is the line to remember. Projects from that date are TanStack Start and server-rendered, on a host that runs the server. Older projects on Lovable hosting get a prerender for verified crawlers and a SPA for everyone else, until you upgrade. The upgrade doc says the upgrade makes the HTML the same for visitors, search, AI crawlers and link previews. If you are deciding whether the project is affected, the date and the hosting are the two questions, and view source is the check when you are not sure which you have. Do not treat a screenshot of the editor as the page. The editor runs JavaScript. The crawler does not.
Common questions
Are new Lovable projects invisible to ChatGPT?
Lovable says projects created from May 13, 2026 use TanStack Start and server-render. Those send HTML on the first response, which is what a crawler that does not run JavaScript can read. Older Vite projects on Lovable hosting are prerendered for verified crawlers, and other tools see the shell.
I exported the project. Does Lovable prerendering still apply?
Lovable's prerender for verified crawlers is part of Lovable hosting. An export you deploy yourself serves whatever that host serves. Check the URL on that host.