There isn't one rendering rule for AI crawlers
The answer depends on the client. Google Search documents JavaScript rendering, while the public bot overviews reviewed here for OpenAI, Anthropic and Perplexity don't specify whether their named crawlers render JavaScript. A blanket yes or no would go beyond those documents.
Rendering means processing the page in a browser environment, where scripts can build or change its content. Downloading a JavaScript file doesn't establish that the client executed it, and fetching a URL successfully doesn't establish that the information added by that script was available.
If essential information depends on code running after the HTML arrives, look for evidence about the client that retrieves it and the conditions under which it visits. A provider's product name alone doesn't identify that route.
What the providers actually document
Bot documentation often explains why a client visits and how website owners can control access. That purpose is useful to know, but a search crawler, a training crawler and a user-triggered fetcher can have different jobs within the same service.
The table separates those documented roles from an explicit rendering statement. Google Search is included as a familiar comparison because its documentation does describe JavaScript rendering; that statement applies to Google Search and shouldn't be extended to every Google AI product.
This is a review of the cited documentation, checked on 12 November 2026, alongside the dated study discussed below. It doesn't report a new experiment with the providers' crawlers.
| Service | Clients and documented purpose | JavaScript rendering |
|---|---|---|
| OpenAIOpenAI bot overview | OAI-SearchBot for search, GPTBot for potential training use and ChatGPT-User for user-triggered visits. | Not specified in the cited bot overview. |
| AnthropicAnthropic crawler guidance | ClaudeBot for training, Claude-SearchBot for search and Claude-User for user-directed retrieval. | Not specified in the cited crawler guidance. |
| PerplexityPerplexity crawler guidance | PerplexityBot for search and Perplexity-User for fetching pages in response to a user question. | Not specified in the cited crawler guidance. |
| Google SearchGoogle: JavaScript rendering and crawling | Googlebot crawls pages for Google Search, with rendering described as a separate processing step. | Yes. Google Search documents JavaScript rendering with evergreen Chromium. |
'Not specified' means the cited document doesn't state the capability. It isn't a finding that the client can't render JavaScript.
Why crawler research needs a date
On 17 December 2024, Vercel and MERJ published crawler research that reported no JavaScript rendering observed for the AI crawlers they studied, including OpenAI bots, ClaudeBot and PerplexityBot. Their investigation used nextjs.org and Vercel network data, with additional checks on Resume Library and CV Library.
That's useful empirical evidence for those clients and sites at that time. It supports a dated statement about what the researchers observed, rather than a permanent claim about every visit made by those services. Their distinction between downloading scripts and executing them is also important when interpreting server logs.
Keep the publication date and method alongside the conclusion when using a study to make an implementation decision. The 2024 observation doesn't resolve an unspecified capability in current documentation.
A browser visit doesn't prove crawler support
An AI application can use more than one tool to reach the web. Cloudflare's web fetch documentation, updated on 12 October 2026, describes retrieving a page over HTTP and converting its HTML into readable content. It directs agents towards a browser tool when the task requires JavaScript-rendered content or browser interaction.
An application could read a dynamic page through a browser even if an HTTP fetch receives only its initial HTML. That doesn't establish the behaviour of an unrelated provider's crawler.
When someone says that an assistant read their JavaScript page, establish how it reached the information before applying that result elsewhere. A successful browser visit leaves the rendering behaviour of the service's indexing crawler unresolved.
Rendering doesn't settle whether the information arrives
Even an explicit rendering capability leaves questions about the page being visited. Google documents a rendering queue and explains that blocked resources can prevent it from rendering a page's JavaScript. Its guidance therefore gives website teams conditions to check alongside the capability itself.
Consider a product page that displays specifications after a request to another service. A rendering client would still need that request to succeed, and a specification behind a tab might require a further interaction. Those are possible dependencies to investigate, rather than proof that a named crawler succeeds or fails on your page.
Keep the observed result specific. If the specification is absent from the returned content, record that absence and the route used to retrieve it. Calling the entire website 'invisible to AI' would hide the practical question of which information was missing and under what conditions.
Check the client you need to reach
Start with the outcome you need, such as making a public service description available to a search retrieval client. Find the provider's current guidance, identify the relevant client and record the URL and date of that guidance. Where the rendering behaviour is unspecified, keep that uncertainty visible in your assessment.
For a practical check, preserve the page URL, time and response, along with the real client or tool used. A request from your own machine with a bot's user-agent string can help inspect your server's response to that header, but it doesn't reproduce the provider's rendering system or prove that its crawler visited.
Putting essential public information into readable HTML removes its dependence on later JavaScript execution for clients that can retrieve that response. You can verify that delivery change directly, then assess any search or assistant outcome separately.


