Skip to content

Do AI crawlers render JavaScript? What ChatGPT and Claude see

Google renders JavaScript, but the crawlers behind ChatGPT, Claude and Perplexity mostly read only the HTML your server sends. Here is who renders what, a five-minute test, and the fixes ranked by effort.

The seoscore team 9 min read

On this page
  1. Which crawlers render JavaScript and which don't
  2. Why a page can show in AI Overviews but not in ChatGPT
  3. What usually goes missing
  4. How to see your page the way an AI crawler does
  5. Is your framework or site builder at risk?
  6. How to fix it: from quick wins to a full rendering change
  7. Check your page with seoscore
  8. After the fix: confirm AI assistants use your pages
  9. Next steps

Mostly, no. Googlebot, Bingbot and Applebot run JavaScript before they read a page. The crawlers behind ChatGPT, Claude and Perplexity (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User and PerplexityBot), along with Common Crawl's CCBot, have read only the HTML your server sends in public log studies. If your text, prices or structured data appear only after scripts run in the browser, those assistants see a blank or half-empty page, even when Google indexes it perfectly.

This is an AEO problem. AEO (answer engine optimization) is whether AI engines can read, trust and quote a page, and reading comes first. Our AEO guide covers the rest. This article assumes the bots are already allowed in. If you haven't checked, start with how to allow AI crawlers in robots.txt.

One caveat. A developer reported that OAI-SearchBot and GPTBot started sending Next.js prefetch requests on 25 September 2026, requests a crawler sends only after it has run JavaScript. That evidence comes from one person's sites, OpenAI hasn't confirmed it, and others have not seen it at scale. Even if it holds, it covers two bots from one company. Build as if AI crawlers read raw HTML.

Which crawlers render JavaScript and which don't

Crawler Operator What it feeds Runs JavaScript?
Googlebot Google Search index, which AI Overviews and AI Mode draw on Yes (documented)
Google-Extended Google Not a crawler: a robots.txt token controlling whether Google's crawl is used for Gemini Not applicable: relies on Google's existing crawlers
Bingbot Microsoft Bing index and Copilot; Bing is reported to be among ChatGPT search's sources Yes (documented)
Applebot Apple Siri and Spotlight Yes
GPTBot OpenAI Model training No in most logs; one unconfirmed report of rendering
OAI-SearchBot OpenAI ChatGPT search index Partial: very rare script fetches
ChatGPT-User OpenAI Live fetches when a chat needs a page No
ClaudeBot Anthropic Model training No
Claude-User Anthropic Live fetches for a user's question No
PerplexityBot Perplexity Perplexity search index No in logs; one test disputes this
Perplexity-User Perplexity Live fetches for a user's question No evidence that it does
CCBot Common Crawl Open web archive used in many training datasets No: stores raw responses

Where this comes from:

  • Google documents its crawl, render and index process in JavaScript SEO basics. Google-Extended has no user agent of its own; Google's existing crawlers do the fetching (Google's crawler list).
  • Vercel and MERJ's study, The rise of the AI crawler (December 2024), found that none of the major AI crawlers rendered JavaScript. OpenAI's and Anthropic's bots downloaded script files but never ran them. Applebot rendered pages like Googlebot.
  • A 30-day server-log study by Forge, ending 24 September 2026, logged ChatGPT-User reading 37,045 pages with no script fetches, and found no script fetches from Claude-User, ClaudeBot or PerplexityBot either. OAI-SearchBot fetched scripts 10 times in 1,738 page requests.
  • Glenn Gabe's case study found a fully client-side rendered site looked blank to ChatGPT, Perplexity and Claude.
  • The Perplexity exception: searchviu's test found Perplexity knew a price that only JavaScript inserted. It may have got it from another source, so treat Perplexity as unconfirmed rather than safe.

OpenAI, Anthropic and Perplexity don't say in their crawler documentation whether their bots run scripts. Every "No" above is observed behaviour, not a promise, and it can change.

Why a page can show in AI Overviews but not in ChatGPT

AI Overviews are built from Google's search index, and Googlebot renders pages before they go into it. If Google can run your scripts, AI Overviews can quote text that exists only after JavaScript has run.

ChatGPT, Claude and Perplexity either build their own indexes or fetch your page live when someone asks a question. Both routes read the raw HTML. ChatGPT search can also lean on third-party search results, and Bing renders JavaScript, so you might see the odd ChatGPT citation. But when ChatGPT-User opens the page to read it, it gets whatever your server sent.

That is why a site can rank well in Google, appear in AI Overviews and still never be quoted by the assistants. It also matters for GEO (generative engine optimization): whether AI assistants name and cite your brand in their answers. An assistant can't cite a page it couldn't read. The GEO guide explains how the two fit together, and our guide to getting cited in AI Overviews covers the Google side.

What usually goes missing

The content that disappears is anything your server doesn't put in the first HTML response:

  • Client-side rendered single-page apps. The server sends an empty <div id="root"> and a script tag. Everything else is built in the browser.
  • Text loaded after an API call. Product descriptions, articles or FAQs fetched from a headless CMS once the page loads.
  • Tabs, accordions and "read more" sections whose hidden text is inserted by JavaScript on click.
  • Infinite scroll and lazy-loaded sections. Anything that appears only when a user scrolls.
  • JSON-LD injected through Google Tag Manager. Google can read it (Google's guidance), but a crawler that doesn't run scripts never sees it.
  • Titles, meta descriptions and canonicals set by JavaScript. The bot sees the defaults, often a generic site title.
  • Prices, stock and reviews from third-party widgets.
  • Cookie walls and consent scripts that replace the page until someone clicks.

How to see your page the way an AI crawler does

These tests take about five minutes per page. Pick a page you want quoted, such as a guide, product page or FAQ.

1. Compare view-source with the inspector

Open the page, then open view-source (Ctrl+U on Windows, Cmd+Option+U on a Mac). That is the raw HTML. Search it for a sentence from the main text. If the sentence appears in the inspector (right-click, Inspect) but not in view-source, JavaScript is putting it there.

2. Turn JavaScript off

In Chrome DevTools, open the command menu (Ctrl+Shift+P or Cmd+Shift+P), type "Disable JavaScript" and reload. What's left is roughly what a non-rendering bot gets. A blank page or a loading spinner is your answer.

3. Fetch the page as GPTBot

From a terminal, request the page with a crawler's user agent and save what comes back:

curl -s -A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.1; +https://openai.com/gptbot" https://www.example.com/your-page > gptbot.html

grep -c "a sentence from your main text" gptbot.html

A count of 0 means the sentence isn't in the raw HTML. Copy current user-agent strings from OpenAI's bot page and Perplexity's bot page, since version numbers change. Two limits: a curl request doesn't come from OpenAI's IP addresses, so a firewall that checks IPs may answer differently; and it shows only what the server sends, not what a bot might do afterwards.

4. Compare word counts

In the browser console on the normal page, run document.body.innerText.split(/\s+/).length. Then do the same with JavaScript disabled. If the second number is a fraction of the first, most of your content depends on scripts.

5. Check the answer and the structured data

In the raw HTML, look for the short answer you want quoted, your FAQ text, the <title>, the meta description, the canonical tag and any application/ld+json block. Each one missing is something an AI crawler can't use.

6. Read your server logs

Logs show which AI bots really fetched which pages:

grep -E "GPTBot|OAI-SearchBot|ChatGPT-User|ClaudeBot|Claude-User|PerplexityBot" access.log | awk '{print $7}' | sort | uniq -c | sort -rn | head

If bots fetch your pages but the raw HTML is empty, rendering is your blocker. If they never appear, check robots.txt and your firewall first.

Is your framework or site builder at risk?

Setup Default behaviour What to check
React with Vite, or Create React App Client-side rendering: empty HTML shell High risk. Plan prerendering or a move to a server-rendering framework
Vue single-page app Client-side rendering High risk, same as above
Next.js, Nuxt, SvelteKit, Remix Server rendering or static generation by default Components that fetch data in the browser after load
Astro Static HTML with little or no JavaScript Client-only islands that hold main content
Angular Client-side unless SSR was chosen Whether @angular/ssr is set up
Gatsby Static HTML at build time Client-only routes and runtime data fetching
WordPress, classic themes Server-rendered PHP Plugins that load tabs, sliders or reviews by script
WordPress with heavy page builders Mostly server-rendered Builder widgets that insert content on load
Shopify Liquid themes are server-rendered App widgets for reviews, prices and FAQs
Wix, Squarespace, Webflow Pages generally served as HTML Third-party embeds and custom code blocks

Whatever the setup, the tests above beat the table. A modern framework can still ship a page whose key text is fetched in the browser.

How to fix it: from quick wins to a full rendering change

1. Put the key answer in the server HTML

The cheapest fix is to make sure the opening answer, the main body text and the FAQ text are in the template, not fetched later. On many sites this means moving a CMS call from the browser to the server.

2. Hard-code JSON-LD

Output structured data in the page template rather than through a tag manager:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Ltd",
  "url": "https://www.example.com"
}

Wrap it in a script tag of type application/ld+json in the server-rendered head or body.

3. Render tabs and accordions in HTML, hide them with CSS

Keep all the text in the HTML and let CSS or the native details element handle showing and hiding:

<details>
  <summary>Do you deliver on Saturdays?</summary>
  <p>Yes. Orders placed by 2pm on Friday arrive on Saturday.</p>
</details>

4. Set titles, descriptions and canonicals on the server

If a router library sets them in the browser, move them to the server or build step so the first response carries the right values.

5. Prerender static routes

For a single-page app, a build-time prerender step can save each route as HTML. It suits pages that don't change per visitor, such as guides, product pages and landing pages.

6. Move key templates to SSR or static generation

Server-side rendering (SSR) builds the HTML on each request; static generation builds it once at deploy. In Next.js, fetch in Server Components rather than in useEffect. In Angular, ng add @angular/ssr adds server rendering. Start with the templates you most want quoted.

7. A prerender service as a stopgap

A prerender service serves bots a rendered snapshot while visitors get the normal app. Google calls this dynamic rendering "a workaround and not a long-term solution" and recommends server-side rendering, static rendering or hydration instead. Use it to buy time, keep the snapshot's content the same as what visitors see, and plan the real fix.

Check your page with seoscore

The free AEO and GEO checker runs this test for you: paste a URL and one of its checks is whether the content is there without JavaScript. The same report covers the other reasons an AI engine might skip a page: AI crawler access in robots.txt, snippet rules, llms.txt, entity markup, freshness, citations, direct answers, FAQ markup and answer blocks. The checker guide explains each result.

One page tells you about one template. A full SEO audit scores your site's pages out of 100, with AEO worth 15 of those points, and ranks every fix by the points it adds, so a rendering problem sits in the list beside everything else competing for your time. Every check and weight is on the methodology page.

After the fix: confirm AI assistants use your pages

Re-run the curl and word-count tests on the live page, then watch your logs. Live fetchers such as ChatGPT-User, Claude-User and Perplexity-User will see the new HTML the next time they fetch the page. Search indexes update after a recrawl. Training crawls only shape future model versions, so expect months rather than days there. Nothing guarantees a citation; fixing rendering removes a blocker.

To see whether it's working, AI visibility asks ChatGPT, Claude, Gemini and Perplexity the questions your customers ask and scores how often, and how early, they name your site, with every answer and the sources cited. AI mentions, on Starter and up with a free trial, reports where Google AI Overviews and ChatGPT cite your site, which of your pages, the questions asked and the other sites cited beside you. For manual methods, see tracking brand mentions in ChatGPT and how ChatGPT picks sources.

Next steps

  1. Run view-source and the JavaScript-off test on your three most important pages.
  2. Curl one of them as GPTBot and look for your opening answer, title, canonical and JSON-LD.
  3. Fix the quick wins first: server-side answer text, hard-coded JSON-LD, HTML accordions.
  4. Plan prerendering or SSR for templates that are still empty.
  5. Check bots are allowed in with the robots.txt AI crawler guide, then read the AEO guide, the AI Overviews citation guide and the GEO guide.

Questions people ask

Will allowing AI crawlers in robots.txt fix a JavaScript problem?

No. Robots.txt decides whether a bot may fetch a page; it says nothing about what the bot finds there. A bot that is allowed in but receives an empty HTML shell still has nothing to quote, so you need both access and server-rendered content.

Is serving AI bots a prerendered snapshot a form of cloaking?

Google says it doesn't generally treat dynamic rendering as cloaking as long as the rendered version carries the same content visitors see. The risk comes from showing bots something different, such as extra text or keywords. Keep the snapshot identical in substance and treat it as a temporary measure.

Does Gemini render JavaScript?

Gemini's answers in Google products draw on Google's index, which Googlebot builds after rendering pages. When Gemini fetches a URL that someone pastes into the app, tests disagree on whether it runs scripts. Server-rendered content is safe either way.

Can llms.txt make up for a JavaScript-heavy site?

Not reliably. An llms.txt file can point AI tools to plain-text or markdown versions of your pages, but there is little evidence that the major assistants read it. Fix the HTML first and treat llms.txt as an extra; our llms.txt article covers what it can and can't do.

Sources

  1. Traffic from ChatGPT jumped! Is it because its crawlers now run JavaScript? - DEV Community dev.to
  2. Do AI Crawlers Render JavaScript? - Digital Drift www.digitaldrifthq.com
  3. developers.google.com/search/docs/crawling-indexing/javascript/javascript-seo-basics developers.google.com
  4. developers.google.com/search/docs/crawling-indexing/google-common-crawlers developers.google.com
  5. The rise of the AI crawler - Vercel vercel.com
  6. Can AI crawlers run JavaScript? What ChatGPT really sees www.forgeaibotlog.com
  7. AI Search and JavaScript Rendering - How client-side rendering causes visibility and ranking problems in ChatGPT, Perplexity, Claude, and others [Case Study] www.gsqi.com
  8. AI crawlers & JavaScript rendering: 39 bots compared www.searchviu.com
  9. developers.google.com/search/docs/appearance/structured-data/generate-structured-data-with-javascript developers.google.com
  10. platform.openai.com/docs/bots platform.openai.com
  11. developers.google.com/search/docs/crawling-indexing/javascript/dynamic-rendering developers.google.com

seoscore

Will AI answer engines quote you?

Run the free AEO checks on any page and see what to fix first.

Check my AEO
More on AEO