You open your site in the browser. Everything looks perfect: the layout loads, the content appears, images land right where they should. Flawless. But there are visitors hitting your site who don't use a browser. And those silent visitors are deciding, right now, whether your brand exists or not for millions of people.
Those visitors are AI crawlers. The same ones that power ChatGPT, Claude, Perplexity and Google's AI summaries. And they work very differently from your browser.
The restaurant metaphor
Picture two restaurants.
In the first one, you sit down and the waiter brings the dish ready to eat. You see the food, smell the aroma, and you can start eating right away. That's how Server-Side Rendering works: the server builds the complete page and delivers everything ready for whoever requested it.
In the second restaurant, the waiter brings the raw ingredients, a recipe, and a portable stove. "Cook it yourself, everything's there." If you're a human with time and patience (or a browser with JavaScript), it works. But if you're an AI crawler in a hurry, with five seconds to understand what's on the plate before moving to the next restaurant, it looks at the table, sees raw ingredients, and leaves thinking the restaurant is empty. That's how Client-Side Rendering works for AI.
The numbers that should worry you
This isn't speculation. Vercel and MERJ analyzed over 500 million GPTBot visits (ChatGPT's crawler) and found not a single instance of JavaScript execution. It even downloads the site's code files (in about 11% of visits), but never runs any of them. Anthropic's ClaudeBot? Same behavior. PerplexityBot? Same.
These crawlers arrive, read the HTML the server delivers in the first response, and leave. If your site's content only appears after JavaScript builds the page in the browser, for them the page is blank.
And the volume is massive. GPTBot alone generates 569 million requests per month on Vercel's network. ClaudeBot, 370 million. Together, they account for about 20% of Google's volume. This isn't marginal traffic.
The cruelest paradox: first on Google, invisible to AI
Here's the part that hurts. Your site can rank first on Google and, at the same time, be completely invisible to ChatGPT, Claude and Perplexity.
How? Google invested billions of dollars over more than a decade to build infrastructure capable of executing JavaScript and "seeing" pages the same way a browser does. AI crawlers didn't make that investment. They operate with tight time budgets (between one and five seconds per page) and rendering JavaScript at that scale would be prohibitively expensive.
This isn't a bug that will be fixed soon. It's an engineering decision.
| Server delivers ready (SSR) | Browser builds (CSR) | |
|---|---|---|
| What arrives first | Page with content | Empty page + code |
| Does Google see it? | Yes | Yes (with delay) |
| Does ChatGPT see it? | Yes | No |
| Does Claude see it? | Yes | No |
| Does Perplexity see it? | Yes | No |
The two-minute test
Want to know if AI can read your site? Do this:
In your browser, press Ctrl+U (or Cmd+U on Mac). This shows the page source code, exactly what an AI crawler receives. Search for your site's main content. If it's not there, if instead you see just a <div id="root"></div> and a bunch of .js file references, your content is invisible to every AI except Google.
Another way: disable JavaScript in your browser (in developer tools). If the page goes blank or shows only a skeleton, that's exactly what AI sees.
Why can Google do it and the others can't?
Because Google maintains a massive infrastructure called the Web Rendering Service, built over more than a decade, that works like thousands of virtual browsers running in parallel. When Googlebot visits your site, it "opens" the page as if it were an invisible Chrome, executes the JavaScript, waits for the data to load, and only then indexes the content.
No other AI crawler has this. Anthropic's official documentation, published in May 2026, states it directly: the web fetch tool "does not support sites dynamically rendered via JavaScript." What the server delivers is what Claude reads. Period.
The only partial exception is Google's Gemini, which inherits Googlebot's infrastructure. Beyond that, the entire AI ecosystem reads only what arrives in the first HTML.
What to do about it
If your site was built with plain React, Vue or Angular (those applications that load everything via JavaScript), there are solutions:
The most robust is migrating to frameworks that deliver ready-made HTML from the server, like Next.js (for React), Nuxt (for Vue) or Angular Universal. They let you keep all the interactivity users expect while ensuring that core content arrives in the first HTML. It's the cleanest long-term bet.
For those who need a quick fix, there are pre-rendering services that intercept crawler requests and serve a static version of the page. It works as a shortcut: the site stays the same for human users, but crawlers receive a readable version.
And for cases where a full migration isn't feasible, you can adopt a hybrid approach: keep JavaScript for interactive elements (menus, filters, animations) and ensure that essential content (text, links, structured data) is in the HTML the server delivers.
Ranking on Google is no longer enough
The visibility game has changed. When someone asks ChatGPT about your market, asks Perplexity for a product comparison, or consults Claude about best practices in your industry, the answer comes from sites that AI could actually read. If your site delivers a blank page to those crawlers, who shows up in the answer is your competitor.
The good news: the problem is known, testable and fixable. The bad news: every week without action is a week where AI is citing someone else in your place.
If this kind of analysis about how technology changes the rules of the game interests you, the MacarenoNet newsletter delivers content like this every week. Straight to your inbox, no fluff.
Sources