JavaScript rendering

Googlebot keeps crawling and rendering in separate queues. Every response with an HTTP 200 status is queued for rendering unless a robots meta tag says otherwise, and a headless Chromium then runs the scripts^[1]. Across more than 100,000 Googlebot fetches of nextjs.org, excluding error status codes and non-indexable pages, every HTML page rendered fully, including content fetched asynchronously and streamed React Server Components^[4]. The wait is usually short: over 37,000 rendered pages matched with server-beacon pairs gave a median delay of 10 seconds and a 75th percentile of 26 seconds, with a tail near 18 hours at the 99th^[5]. A noindex tag in the initial HTML cancels rendering entirely, so a script written to remove it never runs^[6].

The same partnership measured the main AI crawlers and found none of them executing JavaScript, covering OpenAI's GPTBot, OAI-SearchBot and ChatGPT-User, Anthropic's ClaudeBot, Meta-ExternalAgent, Bytespider and PerplexityBot^[2]. They download script files without running them, at 11.50% of ChatGPT's requests and 23.84% of Claude's^[7]. Gemini is the exception, riding Googlebot's infrastructure and rendering in full, as does AppleBot through a browser-based crawler, while Common Crawl's CCBot does not render^[3]. Volumes in the month measured were 4.5 billion Googlebot fetches across Gemini and Search, against 569 million for GPTBot and 370 million for Claude^[8].

Compiled from the Vercel and MERJ log study^[2][3], Google's rendering documentation^[1] and Microsoft's 2018 statement on bingbot^[9]. Support is a moving target and is worth rechecking against each operator's own documentation.

Google's documentation still advises server-side or pre-rendering, since not all bots run JavaScript^[10], and Google's answer surfaces add no requirement of their own: a page must be indexed and eligible for a snippet before it can appear as a supporting link in AI Overviews or AI Mode^[11]. Microsoft's 2018 guidance describes bingbot as generally able to render while still recommending dynamic rendering for sites that rely heavily on JavaScript^[9]. For assistants that read only the first response, a single-page application delivers an empty shell, and titles, prices, body copy and structured data injected after load are absent from what the model sees^[7]. MERJ's 2026 guidance frames the fix as one canonical URL served in several representations rather than a second site^[12].