JavaScript SEO in 2026: Google renders your page, most AI crawlers don't
Google runs your JavaScript before it indexes a page. In Vercel and MERJ's study the major AI crawlers did not, so a page built by scripts can be fine on Google and blank to ChatGPT. How each crawler treats JavaScript, a 30-second test, and the fixes from cheapest up.
JavaScript SEO used to mean one question: can Google see what your scripts put on the page? It now has two audiences. Google runs your JavaScript before it indexes a page. Most AI crawlers do not, so a page that is fine on Google can be blank to ChatGPT, Claude and Perplexity.
In our check of 315 small-business homepages, 49 (16%) had fewer than 80 words in their HTML. In a browser, nearly all of them looked like complete websites.
How Google handles JavaScript
Google describes three phases: crawling, rendering and indexing (JavaScript SEO basics). Googlebot fetches the HTML and puts the page in a queue; when Google has capacity, an up-to-date headless Chromium renders the page and runs its JavaScript, and the rendered result is what gets indexed.
That works, with caveats Google states itself:
- Pages can wait in the render queue for seconds, and sometimes a good deal longer.
- Google does not render JavaScript from files or pages that robots.txt blocks, so a rule against your script or API folders can leave the rendered page empty.
- Google only discovers links that are real
<a>elements with anhref. - Google still recommends server-side rendering or pre-rendering, because it makes pages faster and because not every bot can run JavaScript.
How AI crawlers handle it
The best public measurement is Vercel and MERJ's study of AI crawler traffic across Vercel's network, published in December 2024. Their finding: "none of the major AI crawlers currently render JavaScript." The crawlers they tested included OpenAI's OAI-SearchBot, ChatGPT-User and GPTBot, Anthropic's ClaudeBot and PerplexityBot. They did download JavaScript files, ChatGPT's crawler spending 11.5% of its fetches on them and Claude's 23.8%, but never ran them.
| Crawler | Runs your JavaScript? |
|---|---|
| Googlebot (Google Search, and Gemini through it) | Yes, in a delayed render step |
| Applebot | Yes, through a browser-based crawler |
| OAI-SearchBot, ChatGPT-User, GPTBot (OpenAI) | No, in the December 2024 study |
| ClaudeBot (Anthropic) | No, in the December 2024 study |
| PerplexityBot | No, in the December 2024 study |
Crawlers change, so treat the table as dated. But the safe assumption for 2026 is the old one: if your words are not in the HTML the server sends, an AI assistant probably cannot quote them.
What goes missing
- Whole pages, on single-page apps built with React, Vue or Angular that render in the browser. The HTML is an empty
<div id="root">and a script tag. - Sections loaded later: reviews, FAQs, menus, prices or opening hours pulled in by a widget after the page loads.
- Text behind clicks: tabs and accordions whose content is fetched only when someone opens them. (Content that is in the HTML but hidden until a click is fine.)
- The title and description, when a script sets them. Crawlers that do not run it see the same title on every page.
- Links, when navigation is built from click handlers instead of
<a href>, or uses hash routes like/#/services.
The 30-second test
- Open your page in Chrome on a computer and press Ctrl+U (Cmd+Option+U on a Mac). That is the HTML your server sent, before any JavaScript ran.
- Press Ctrl+F and search for a sentence you can see on the page, such as your opening line or your phone number.
- Found it? Crawlers can read it. Not there? Then it is added by JavaScript, and most AI crawlers will not see it.
If you prefer a terminal, request the page under a crawler's name and search the result:
curl -s -A "OAI-SearchBot" https://www.example.com/ | grep -c "a sentence from your page"
Or paste the address into our AI visibility checker, which counts the words in the HTML and shows what each crawler received.
Fixes, from cheapest to most work
- Put the essentials in the HTML. Even on a script-heavy site, the HTML template can carry a plain paragraph with who you are, what you do, where, and how to reach you. It costs nothing and gives every crawler something true to read.
- Load widgets' content on the server. Reviews, menus and FAQs can usually be written into the page when it is built, with the widget left to add interaction on top.
- Prerender at build time. Render your pages to HTML during the build and let the browser take over from there. We did this for our own React homepage: how our React homepage showed Google an empty page.
- Use a framework that renders on the server, such as Next.js, Nuxt, SvelteKit or Astro, for new builds whose job is to be found.
Serving crawlers a separate prerendered copy of each page also works, but it means two versions of every page to keep in step, and the day they drift apart, crawlers and visitors see different businesses.
A short checklist for developers
- Every indexable page returns its main text, title, meta description and canonical in the initial HTML.
- Links are
<a href>elements; routes use the History API, not#fragments. - Unknown URLs return a real 404, not the app shell with a 200.
- robots.txt does not block the JavaScript, CSS or API paths that Google needs to render.
- Structured data is in the HTML, not injected after load.
Questions
Does Google index JavaScript content?
Yes, after rendering it, which can take from seconds to longer. Content that only appears after a click or a scroll, or that depends on blocked files, can be missed.
Can ChatGPT read JavaScript websites?
In the December 2024 measurements, OpenAI's crawlers downloaded JavaScript but did not run it. If your words only exist after JavaScript runs, assume ChatGPT search cannot quote them, and put them in the HTML.
Is a site builder like Wix or Squarespace safe?
Builders that send finished HTML are fine. Run the Ctrl+U test on your own pages rather than trusting the platform: the widgets you add are where text most often goes missing.
This is day 4 of 20. The study behind it: what we found checking 400 small-business websites.