- According to an analysis by Vercel and MERJ, none of the major AI crawlers execute JavaScript, including GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, and PerplexityBot.
- If it isn't in the delivered HTML, it doesn't exist for these crawlers. A page can work for people and still be empty for AI systems.
- Google does render JavaScript, but with a delay, via a queue. Google itself recommends server-side rendering or pre-rendering, because not every bot executes JavaScript.
- The fix is architectural: core content, headings, links, and metadata belong in the server-delivered HTML, while interactivity can stay in the browser.
More and more people put their questions straight to ChatGPT, Perplexity, or Google's AI Mode. Answers arrive instantly, often without anyone clicking through to a website. Visibility no longer depends on rankings alone, but on whether AI systems can read your content in the first place.
This is exactly where many modern websites trip over a technical detail: JavaScript rendering. We know the problem firsthand. Our previous website served empty HTML without JavaScript and was virtually invisible to AI crawlers, so we rebuilt unitedads.de as a statically generated site. This article explains why the problem occurs, how to spot it, and how to fix it.
On many websites, content is only built in the browser
Frameworks such as React, Vue.js, or Angular enable fast interactions and flexible interfaces. The trade-off: the actual page content often no longer sits in the HTML the server delivers. Instead, JavaScript assembles it in the browser, after APIs have been called and state has been built.
For people, that's no problem: the browser renders everything reliably. A crawler that doesn't execute JavaScript, however, often sees nothing but an empty shell:
<!doctype html>
<html>
<head><title>Ihre Website</title></head>
<body>
<div id="root"></div>
<script src="/assets/index.js"></script>
</body>
</html>
The major AI crawlers don't execute JavaScript
AI systems don't read websites the way a person does in Chrome or Firefox. They load the delivered HTML and analyze its text and structure. Vercel and the consultancy MERJ analyzed requests to nextjs.org and across Vercel's network. The finding: none of the major AI crawlers execute JavaScript. The study covered OpenAI's GPTBot, OAI-SearchBot, and ChatGPT-User, Anthropic's ClaudeBot, Meta-ExternalAgent, ByteDance's Bytespider, and PerplexityBot.
The crawlers do download JavaScript files (11.5% of requests for ChatGPT, 23.84% for Claude), but they don't execute them. The exceptions are Gemini, which runs on Googlebot's infrastructure, and Applebot, which renders like a browser.
The rule is simple: If it isn't in the HTML, it doesn't exist for these systems. A page can work perfectly for people and still be empty for AI systems. The study dates from December 2024; your own server logs will show whether any crawler behaves differently today.
Google renders JavaScript later, via a queue
Google now handles JavaScript well, with caveats. It processes JavaScript pages in three stages: crawling, rendering, and indexing. Before rendering, the page goes into a queue, which according to Google can take a few seconds or longer. Failed API calls, slow dependencies, or complex client-side logic can lead to content being captured only partially.
That's why Google itself still recommends server-side rendering or pre-rendering: it makes pages faster for users and crawlers, and not every bot can execute JavaScript. Counting on Google to cope anyway means optimizing for a single recipient.
In AI search, readability decides whether you qualify as a source
The technical problem isn't new. What's new is how content gets used. AI systems cite websites directly, lift passages, and build answers from them. The question is no longer whether a page ranks third or fifth, but whether it qualifies as a source at all.
If an AI can't read your content, you won't appear in its answers. For Google Search itself, Google's guide to AI search is clear: optimizing for AI Overviews and AI Mode is still SEO, and the page must be indexed and eligible to show a snippet. ChatGPT, Claude, and Perplexity add another factor: their crawlers only see the HTML from the server.
Your page's source code reveals the problem
A simple test is usually enough. View your page's HTML source or load it with JavaScript disabled. If there's hardly any real content, that's a warning sign.
It's especially critical when text, headings, or internal links are injected later via JavaScript, or when the title and meta description are set dynamically. AI crawlers see none of it.
Visibility starts on the server
The answer isn't to abandon JavaScript, but to use it properly. What matters is that the server delivers the core content of each page, through server-side rendering, static generation, or a hybrid approach. Machines need to find the content directly in the HTML.
Interactivity and complex logic can still run in the browser, but they must not determine whether content exists at all. For blogs, knowledge pages, and corporate websites, this isn't a technical detail. It's a decision about your visibility.
Seven steps to make your content readable for AI crawlers
Check which content a crawler can actually retrieve. How the page looks in the browser is not enough to tell.
1. Check what actually arrives in the HTML
Before rebuilding anything, get clarity. Open key pages without JavaScript or fetch them via curl. What counts isn't what the browser shows after rendering, but what the server delivers. If headings, body copy, or internal links are missing there, you have a structural problem.
2. Deliver core content from the server
Everything needed to understand the page belongs in the delivered HTML: page title, main headings, explanatory text, lists, product descriptions. JavaScript can enhance this content but shouldn't be what creates it. For content pages, that means server-side rendering or static generation.
3. Separate content from interactivity
A common mistake: text only appears once a certain state is set or an API call succeeds. A clear principle works better: content first, logic second. Filters, tabs, or sorting can change the presentation, but they shouldn't generate the information.
4. Use a clean, semantic HTML structure
AI systems rely on structure. A page built from nested <div>elements is harder to interpret than one with clear headings, paragraphs, and lists. That means a single <h1>, logically structured <h2>and <h3>levels, real <p>paragraphs, and links as <a href="...">.
5. Set metadata on the server
Many AI crawlers can't see title tags, meta descriptions, or structured data set via JavaScript. Key meta information always belongs in the server-delivered HTML. Without that context, your chances of being used as a source drop.
6. Write quotable passages
AI systems work with clear statements. Pages made up only of controls, cards, or dynamic components offer little usable text. What helps are paragraphs that answer a question directly or explain a concept. Those are exactly the passages AI answers quote or paraphrase.
7. Test the page from an AI's point of view
The best test: what's left when no JavaScript runs? Is the page's purpose still clear, and can a reader tell what it's about without clicking or interacting? If not, that's not an edge case but a structural risk. Also check your server logs to see which AI crawlers request your pages.
Conclusion: HTML is the most important interface again
The key question used to be: can Google index my page? Now there's a second one: can an AI understand my content without a browser? If you can't answer yes with confidence, you risk being left out of AI answers, however good your product and content are.
Modern frameworks still work for this; they just need to be used more deliberately. We made that choice for our own website: every page delivers its full content as HTML, and JavaScript runs only where a feature needs it. To learn how AI systems choose their sources, read our article on Generative Engine Optimization.
