Lovable SEO: what still keeps your app out of Google and ChatGPT

Yes. A Lovable app can rank on Google and be cited in ChatGPT. Since 13 May 2026, Lovable has handled most of the rendering that used to leave crawlers with an empty page. Lovable SEO now comes down to four things you control:

1. The stack. TanStack Start sends full HTML to everyone. The older React and Vite stack relies on pre-rendering for verified crawlers.
2. The host. That pre-rendering is part of Lovable hosting. Moving the app means checking the HTML again.
3. The domain. One primary custom domain, named in the canonical tags and the sitemap.
4. The public pages. What the product does, who it is for, and the price, in the text a crawler receives.

You do not need a rewrite, and Lovable says a project it hosts does not need an external pre-rendering service.

Does Lovable have SEO built in?

Yes, for the foundations. Lovable’s docs say a published app returns readable HTML to search engines and AI crawlers. HTML is the page as text, before a browser draws it. A crawler is a program that fetches pages for a search engine or an assistant. If its first response is empty, it has little to keep.

New projects created from 13 May 2026 use TanStack Start. New Enterprise projects followed on 22 June 2026. TanStack Start uses server-side rendering (SSR): the server builds the finished page, so every visitor and every tool gets the same HTML on every request.

Older projects use React and Vite. When a verified crawler asks for a published public URL, Lovable pre-renders the page and returns the HTML. The list covers Google and Bing, the AI search crawlers of ChatGPT, Perplexity, Claude and Gemini, social preview bots, and verified SEO tools. Tools outside that list see very little, unless they ask for a Markdown copy, which Lovable serves on both stacks.

Lovable also runs an SEO and AI search review. It checks the sitemap, robots.txt, titles and descriptions, structured data, headings, canonical tags, indexing, and AI crawler access. A canonical tag names the address you want treated as the real copy of a page. Ranking still depends on the content, the domain, and links from other sites. Lovable says so too.

Which Lovable stack is your app on?

Ask in the project chat: “What stack is this project on?”

On TanStack Start, a buyer, a link preview, and a search crawler see the same full page.

On React and Vite, the pre-rendered snapshot usually includes content that loads with the page. Content that loads late, often from the database, may be missing. A pricing page can be stored without the price. A shared link can show a blank preview, because LinkedIn, Slack, X, and WhatsApp read the HTML and mostly do not run JavaScript. An outside SEO scanner is not a verified crawler either, so it may call the page empty while Google receives the rendered copy. Treat the scanner as a hint.

You can upgrade in place. Lovable says most upgrades cost 10 to 35 credits. The live site keeps the old version until you publish, and you can restore the earlier version from history if the result is wrong. I would upgrade whenever search or link previews matter.

How do you check what Google sees, and what should be public?

Run this on the pages that sell: home, pricing, and the page that says who it is for.

1. Verify the site in Google Search Console. Lovable can do that from the SEO and AI tab.
2. Open URL Inspection and paste one full address. You get the copy Google stored and a live test.
3. Open the HTML and read it as a customer. Look for the price, who it is for, and the offer. If those sentences are missing, Google may not have them.
4. Compare the canonical you declared with the one Google selected. Fix any mismatch first.

Google’s JavaScript guide says content has to be in the rendered HTML before Google can index it, and rendering can wait in a queue. SSR and pre-rendering put the words in the first response, which also helps bots that never run JavaScript. “URL is on Google” means the page may appear. It does not mean it ranks.

Those selling pages have to be public and complete. Say what the product does, who it is for, and the price or how pricing works. Give one use case. Answer the objection buyers already raise. Put it in the visible text, not in a panel that fills in after a database call.

Dashboards and customer data stay behind the login. Google says to send a clear “not allowed” status for pages a crawler cannot open. ChatGPT can only cite what it can read.

What changes when you move your Lovable app off Lovable hosting?

Lovable’s pre-rendering belongs to Lovable hosting. The export does not bring it along, so treat the move as a fresh launch for search.

Lovable’s deployment guide splits the move by stack:

• React and Vite leaves as static files. The browser builds the page, and nothing pre-renders it for crawlers. Upgrade to TanStack Start before you leave, or make sure the new host returns real HTML on the first response some other way.
• TanStack Start needs a host that runs a server. Lovable names Vercel, Netlify, and Cloudflare Pages. A host that only stores files drops server-side rendering.

Test on a temporary address first. Open a page and confirm the HTML arrives filled in. That check is separate from clicking around in a browser.

Your lovable.app address stays with Lovable and keeps serving the last version you published there. When the new host checks out, point the custom domain at it. Then confirm that old addresses redirect, canonical tags name the new primary domain, and the sitemap lists that domain, not the Lovable host.

Which address is your real site?

Use one primary custom domain. Lovable lets you connect several. With a primary set, the others redirect to it with a temporary HTTP 302, because you can change the primary later. Lovable hosting does not offer a permanent 301 between connected domains. If you unset the primary, every connected domain serves the project and search credit is split.

Put the primary domain in the canonical tags and the sitemap. Lovable’s review flags canonicals that point at the homepage when they should not, and sitemaps that point at another host.

Custom domains are on paid plans. A lovable.app address is fine for a demo. Search presence belongs on a domain you control. Once it is live, verify it in Search Console and submit the sitemap. If you change the primary, verify the new one and run the review again.

Some addresses never appear in search: branded workspace URLs (app name, then workspace, then lovable.app), unpublished projects, and projects whose access is limited to the workspace or named people.

What makes ChatGPT, Perplexity or Google’s AI answers cite a page?

No file buys a citation. The work is to make a public page worth quoting.

Google says AI Overviews and AI Mode use the same foundations as ordinary search. A page has to be indexed and allowed to show a snippet. No new file or special markup is required. If you add structured data, it should match the visible text.

OpenAI runs separate crawlers with independent settings. OAI-SearchBot places a site in ChatGPT search answers. Block it in robots.txt and the site will not appear there. GPTBot is for training. You can allow search and block training. ChatGPT-User is a visit a person starts, not the switch for search.

PerplexityBot builds Perplexity’s search results. Perplexity-User fetches a page when someone asks a question and generally ignores robots.txt. Allow the search bot. A robots.txt change takes about a day to apply at OpenAI, and up to a day at Perplexity.

Lovable says llms.txt is not required. I would spend the time on the page instead: a direct answer to a buyer’s question near the top, facts in the visible text, structured data that matches them, and links from other sites. None of it forces a citation.

What does Lovable’s SEO review not tell you?

Lovable says the review does not judge content quality, search intent, backlinks, competitor strength, or whether the topic is worth targeting. It no longer measures page speed. For that, use PageSpeed Insights on the published URL. A green review means the checks it knows how to run pass.

That is Goodhart’s law, which I wrote about in nothing more practical than a good theory. A measure that becomes the goal stops telling you what you cared about. The goal is a buyer finding the page and recognising the offer.

On my own site, scripts/check_seo.py blocks any change where a page has more than one H1, no meta description, a canonical that does not match the sitemap, or more than one structured-data block. After a merge, the site notifies participating search engines through IndexNow. The script stops a broken page shipping. It does not know whether the page is worth reading. Lovable’s review has the same boundary.

If your Lovable app is about to go live, or you are moving it off Lovable hosting and you want customers to keep finding it, send me the link. I take AI-built products to production. Here is how I work. There is more about me on the homepage.

See you space cowboy...