Can AI search engines read PDFs on my website?

Google is the one AI search operator that documents PDF handling in detail. In its March 31, 2026 Search Central post on crawling, Google says Googlebot fetches up to 2MB for an HTML URL and up to 64MB for a PDF. Bytes past those limits are ignored: not fetched, not rendered, not indexed.

That matters because Google's AI surfaces sit on Google's index. Google states that to appear as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to show in Search with a snippet. There are no additional technical requirements, and Google doesn't describe a separate rule for PDFs.

Outside Google, the picture is thin. As of October 2026, the crawler pages from OpenAI, Anthropic and Perplexity explain which bot is for training, which is for search and which acts for a user. They don't say whether those bots parse PDFs, how much of a file they read, or how they treat a PDF next to an HTML page.

So the honest answer has two halves:

• Google: yes, within documented limits.
• Everyone else: unknown. Treat a PDF as a fact you can't confirm the engine sees.

The practical move follows from that gap. Anything you need AI answers to get right should also exist as text in a normal HTML page.

The 64MB PDF limit is generous. Most brochures, spec sheets and price lists sit well under it. The limit is per URL, so one large PDF doesn't affect other pages on your site.

The same Google post notes that any other crawler that doesn't specify a limit defaults to 15MB. For HTML, Google adds that the vast majority of the web will never hit the 2MB cap. If you want the detail on HTML payloads, see whether your page is too large for Google and the definition of a crawl payload limit.

Two things the limit doesn't tell you:

1. Inclusion. Google says indexing and serving isn't guaranteed. A fetched PDF can still be left out of an answer.
2. Gemini app behavior. The Gemini app and the Gemini API's grounding are separate Google paths, governed partly by Google-Extended, not only by Search indexing.

The next move: check in Search Console whether your important PDFs are actually indexed, rather than assuming they are.

ChatGPT search, Claude and Perplexity each run their own crawlers. OpenAI's bots page says OAI-SearchBot decides whether a site can show in ChatGPT search answers, while GPTBot covers training only. Each setting is independent of the others. Anthropic and Perplexity publish similar per-bot pages.

None of those pages, as of October 2026, describes PDF parsing. That leaves three open questions you can't answer from outside:

• Does the bot download PDFs at all, or skip them?
• If it downloads them, does it extract text from every page?
• When the same fact appears in a PDF and an HTML page, which one does retrieval use?

There is a related, well-documented gap. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so client-rendered content is invisible to them. The lesson carries over: these crawlers are simpler than a browser. Content in the most basic format, server-rendered HTML text, is the safest bet. If your site relies on client-side rendering, read whether ChatGPT can read a React website first.

Some PDFs are unreadable no matter how capable the crawler is.

• Scanned PDFs. A scan is a picture of text, not text. No AI crawler operator documents running OCR on PDFs. The same risk applies to text inside images.
• PDFs behind a form or login. If a visitor has to submit an email to get the file, a crawler can't fetch it.
• Blocked by robots.txt. A Disallow: /downloads/ rule removes the folder for the bots it names. Rules are per bot, so check each search bot, not just one.
• Blocked at the firewall. A CDN or firewall can return 403 to a crawler that robots.txt allows. OpenAI says eligibility needs both the robots.txt permission and host or CDN access for its published IP ranges.
• Header directives. PDFs can't carry a robots meta tag, so servers use the X-Robots-Tag HTTP header. A stray noindex or nosnippet set for a whole file type quietly removes every PDF from Google.

A leftover header looks like this:

X-Robots-Tag: noindex

If you see it on PDFs you want found, remove it. Google documents these controls on its robots meta tag page, and the nosnippet guide covers how they affect AI answers.

Reality: Google reads PDFs within known limits, and other engines don't say. What it means: a fact that lives only in a PDF is a fact some engines may never see. The fix is cheap.

1. List the facts that matter. Prices, plan limits, specs, dimensions, compatibility, warranty terms, opening hours.
2. State each one in HTML text. One clear sentence per fact, on a page in the raw HTML before JavaScript runs. A fetchable page with a specific fact gives an engine a sentence it can quote.
3. Keep the PDF. Link to it as the download, and keep its numbers in sync with the HTML. Contradictions between the two are a likely source of wrong answers.
4. Check access per bot. Confirm robots.txt lets OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot into the folders that hold both pages and PDFs.
5. Test the fetch. The free ChatGPT page fetch test shows what comes back as text. The AI Crawler Checker requests a URL as each AI crawler and shows which ones your firewall turns away.

If most of your PDFs are manuals or support material, see whether AI engines can cite your help center and docs. None of these steps guarantees a citation. They remove the blocks you control.

Can PDFs appear in Google AI Overviews? Google's rule for AI Overviews and AI Mode is that content must be indexed and eligible to show in Search with a snippet, with no extra technical requirements. Googlebot fetches PDFs up to 64MB. Google doesn't describe a separate PDF rule, so an indexed, snippet-eligible PDF faces the same test as an HTML page. Eligibility is not inclusion, and Google says serving isn't guaranteed.

Should I convert all my PDFs to HTML? Not necessarily. Convert the facts, not the files. Put prices, specs and other load-bearing details into server-rendered HTML pages, then keep the PDF as a download for people who want the full document. That covers crawlers that may skip PDFs without forcing you to rebuild long manuals. Keep the two versions in sync so engines don't find conflicting numbers.

How do I keep a PDF out of AI answers? For Google, send an X-Robots-Tag HTTP header with noindex or nosnippet on the PDF, since a PDF can't hold a robots meta tag. For other engines, block the relevant search bot in robots.txt, such as OAI-SearchBot for ChatGPT search. User-triggered fetchers may not follow robots.txt: OpenAI says robots.txt may not apply to ChatGPT-User, and Perplexity says Perplexity-User generally ignores it.

Does a large PDF hurt the rest of my site? Google's limits apply per URL, so a big PDF doesn't reduce what Googlebot reads on your other pages. The risk is inside the file: Google ignores anything past 64MB, so facts near the end of a very large PDF may never be indexed. Other AI crawlers don't publish PDF limits, which is another reason to put key facts in HTML.