Can AI Crawlers Read JavaScript? Why Some Sites Look Blank to ChatGPT

Two versions of every web page

Every page exists twice. There is the HTML the server sends, and there is the page a browser builds after it runs the scripts in that HTML. On a traditional site the two are nearly the same: the words are in the HTML from the start. On a site built to render in the browser, the HTML can be a near-empty shell with one instruction: go fetch the content and draw it.

A person never notices the difference, because a browser runs the scripts in a fraction of a second. A crawler that reads only the raw HTML notices everything. It records a page with a title, a navigation bar and no prices, no services, no reviews. Then an engine asked about your business has nothing to quote, and it quotes someone else. That is one of the quieter reasons a site ranks on Google but is missing from AI answers, a pattern we describe in your site ranks on Google and is invisible to AI.

A crawler that does not run scripts sees the page you shipped, not the page you designed.

What Vercel's 2024 crawler study found, and how old it is

The best public evidence comes from one source: an analysis Vercel published on December 17, 2024, based on crawler traffic across its hosting network. It is useful and it is dated. Vercel is a hosting company with a view of its own customers' traffic, not an independent lab, and the crawlers it describes may have changed since. We treat it as medium confidence and say so.

One detail is worth knowing. Vercel found that ChatGPT's and Claude's crawlers do download JavaScript files (11.50% and 23.84% of their requests, respectively) but do not run them. Fetching a script is not the same as executing it. Vercel's own advice was blunt: server-render the content that matters. Google says much the same in its JavaScript guidance, noting that not all bots can run JavaScript. The full crawler list, with what each one is for, is on the AI crawlers page.

Sites and widgets most likely to go blank

This is not a whole-site verdict. Many sites are fine everywhere except one or two important pieces that a script drops in after load. Those pieces are often exactly the ones an engine would quote.

Test your own pages with view-source and curl

You do not need a tool or an agency to check this. You need a sentence from your page and two minutes. Pick a sentence that matters: a price, a service name, a line from a review.

1. 1

   Open the raw HTML

   In Chrome, Edge or Firefox, type view-source: in front of the page address and press Enter. This shows what the server sent, before any script ran.

2. 2

   Search for your sentence

   Press Ctrl+F (Cmd+F on a Mac) and search for the sentence. If it is there, non-rendering crawlers can read it. If it is not, they probably cannot.

3. 3

   Check from the command line

   Run curl -s followed by the page address, piped to grep -i and your phrase in quotes. No output means the phrase is not in the HTML the server returns.

4. 4

   Try it as the crawler

   Add curl's -A option with a user agent containing GPTBot or OAI-SearchBot. If the response differs, or returns an error, a firewall or CDN rule is treating AI crawlers differently.

5. 5

   Look for your structured data

   Search the view-source page for application/ld+json. If your business markup is missing there but shows in browser tools, a script is adding it.

6. 6

   Turn JavaScript off and look

   In your browser's developer tools, disable JavaScript and reload. What remains is roughly what a non-rendering crawler gets.

How Next.js puts the words in the HTML

Next.js, the framework DataCram builds on, was designed for this problem. In its App Router, pages and layouts are Server Components by default: they render on the server and arrive as finished HTML. Pages that do not depend on request-time data are prerendered at build time and served from a CDN. Interactive pieces are added as Client Components on top of content that is already in the page.

Titles, descriptions and structured data live in code and ship in the HTML, not through a tag manager. That is what we mean when we say a site is readable by every crawler on the table above. The same build choices make pages fast, which is why rendering sits inside our site speed and Core Web Vitals work too.

A fair warning: Next.js does not make a site readable by magic. A developer can still fetch content in the browser after load, or wrap the important text in a widget, and end up with the same blank page. The framework makes the right thing the default. It still has to be built that way, and tested with the steps above.

Where WordPress stands, fairly

A standard WordPress site renders each page in PHP on the server, so the post and page content arrives in the HTML. On this particular question, a well-built WordPress site is fine, and we would rather say so than pretend otherwise. The risks come from what gets added on top.

Our larger case for Next.js is about speed, upkeep and security, and it is laid out in WordPress vs Next.js. For AI readability alone, the test matters more than the platform. Run view-source on your key pages, whatever they are built on.

Fixes, from a free afternoon to a rebuild

If the test shows missing text, the fix depends on how much is missing. Start at the top of this list and stop when the important facts are in the HTML.

• Write the key facts as plain page text. Hours, prices, services, areas and a few review quotes, typed into the page rather than pulled from a widget. Often an afternoon.
• Move structured data into the page source. Output the JSON-LD from the site or CMS instead of a tag manager.
• Keep important answers on the page, visible. Put FAQ answers in the HTML and keep the critical ones out of tabs that load on click.
• Turn on server rendering or prerendering. Many frameworks support it; on a browser-rendered app it is a development project, not a setting.
• Rebuild when the shell is the whole site. A single-page app with no server rendering is usually cheaper to rebuild than to patch. DataCram's custom builds start at $1,750, fixed price.

Rendering is one item on a longer list, and fixing it is part of our AI search optimization work. A GEO audit checks it alongside crawler access, entity consistency, reviews and content, and the AI visibility check shows whether the engines name you once the words are readable. For the rest of the work, start at the generative engine optimization guide.