Does AI read my whole page or only part of it?

An AI engine rarely uses your whole page. Your content passes three filters before any of it reaches an answer.

1. The fetch limit. A crawler stops downloading at a set size. For Googlebot that is 2MB of HTML per URL.
2. Rendering. Most AI crawlers read the raw HTML your server sends. They don't run JavaScript, so anything a script adds later never reaches them.
3. Selection. The engine splits what it fetched into pieces, picks the pieces that fit the query, and may cite a single sentence.

The first two filters are in your control. The third is not. That's why the practical move is narrow: make sure the facts that matter (prices, product names, definitions, numbers) sit in the server-rendered HTML. After that, the engine decides what to use.

Google documented its limit in a March 2026 Search Central post. Googlebot fetches up to 2MB for any individual URL, excluding PDFs. For PDFs the limit is 64MB. Google says bytes past the threshold aren't fetched, rendered or indexed.

Two details matter:

• The limit is per URL. Images, scripts and stylesheets each have their own counter, so they don't eat into your HTML budget.
• The same post says Google crawlers that don't specify a limit default to 15MB. That's a Google default, not a documented limit for OpenAI, Anthropic or Perplexity crawlers.

Google also says the vast majority of the web never hits 2MB. Treat this as a rare hard failure, not a routine problem. If you suspect a bloated page (huge inline data, embedded base64 images), our guide on whether your page is too large for Google covers how to check. For documents, see how AI search engines handle PDFs.

This is the filter that removes the most content in practice. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of their requests) but don't execute them. In Vercel's words, they can't read client-side rendered content.

Independent retesting in June 2026 found the same for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Meta-ExternalAgent and Bytespider. Googlebot and AppleBot are the exceptions: they render pages.

So the same page can look complete to Google and half-empty to ChatGPT search. Common gaps:

• Prices or plan tables loaded by a script after page load
• Tabs, accordions and "load more" sections that fetch content on click
• Single-page apps that ship an empty shell and build everything in the browser

The test is simple. Compare the raw HTML with the rendered page and list what exists only after hydration. The free JavaScript rendering check does that diff, and the playbook on finding facts hidden behind JavaScript walks through the fixes.

Even when a crawler gets everything, the engine uses a fraction. Engines segment pages inside their own systems, and you can't set how they do it. Google's guide to generative AI features rejects chunking as a requirement and says there's no ideal page length.

Citation formats show how small the used piece can be. As of October 2026, Anthropic's web search tool returns span-level citations with a cited_text field of up to 150 characters. Perplexity cites whole sources by number. OpenAI's API separates the sources a model consulted from the ones it cited, and notes that sources often outnumber citations. Being read is not the same as being quoted.

What you can do is give the engine a clean sentence to lift. A specific fact stated plainly on a fetchable page ("The Starter plan costs $49 a month") is quotable. A fact implied across three paragraphs is not. For more on span citations, see what Claude's web search tool docs reveal. For live fetches, read what happens when ChatGPT opens your URL.

One edge case runs the other way. Content marked data-nosnippet is still indexed, but Bing and Google keep it out of snippets and AI summaries. Check that a template hasn't applied it to content you want quoted.

Work through these in order. Each one removes a block. None guarantees a citation.

1. Diff raw HTML against the rendered page on your pricing, product and about pages. List every key fact that appears only after JavaScript.
2. Move those facts into the server response through server-side rendering, prerendering or static HTML.
3. State each key fact in one plain sentence so it survives being cut down to a short span.
4. Scan for suppression tags: noindex, nosnippet, max-snippet:0 and data-nosnippet on content you want used.
5. Check byte size only on unusually heavy pages. Most sites can skip this.
6. Confirm crawlers can reach the page at all. A firewall can block a crawler your robots.txt allows. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Don't serve AI crawlers a different version of the page. That's cloaking, and it creates real risk. Our answer on separate pages for AI crawlers explains why one server-rendered page for everyone is the safer route.

Does a longer page get cited more often by AI? No evidence says so. Google states there's no ideal page length for its generative AI features. Length only becomes a problem at extremes, such as an HTML file over Googlebot's 2MB limit, where the excess is ignored. What matters more is whether your key facts are in the raw HTML and stated plainly enough to be quoted as a short span.

Should I split my pages into short chunks for AI? No. Google's AI optimization guide rejects chunking as a requirement. Research on chunking studies how retrieval systems split documents inside their own indexes, which publishers can't control. Restructure a page if it helps human readers. Don't expect a different layout to change how ChatGPT, Claude or Google segments it.

Can Google AI Overviews see my JavaScript content? Generally yes, because Googlebot renders JavaScript, and AI Overviews draw on Google's index. That's the gap that catches teams out: a client-rendered page can look fine in Google while ChatGPT, Claude and Perplexity crawlers see an empty shell. Test the raw HTML, not what Google shows, when you care about non-Google engines.

Can I keep part of a page out of AI answers? Yes. The documented controls are `noindex`, `nosnippet`, `max-snippet` and the `data-nosnippet` attribute. Bing says content marked `data-nosnippet` is still indexed but excluded from snippets and AI summaries, and Google supports the same attribute. These are suppression tools only. No markup makes a section more likely to be used.