# Does AI cite PDFs?

> Menus, price lists, brochures, white papers, forms. Lots of small businesses keep their best answers in PDF files. Can ChatGPT, Claude, Perplexity or Google's AI answers read them and point people to them? Here is what the official docs say, what nobody has confirmed, and how to set your PDFs up so they help rather than hide your answers.

Updated October 12, 2026 · AI Syndicate team · https://www.aisyndicate.com/blog/does-ai-cite-pdfs/

**Quick answer:** Partly. Google documents that it indexes PDFs, reads **the first 64MB** of each one, and treats their links much like HTML links, so a text-based PDF can appear in Google Search and, in principle, Google's AI answers. OpenAI, Anthropic and Perplexity say **nothing** about PDFs on their crawler pages, so whether their bots read your PDFs is unconfirmed. Scanned PDFs with no real text are the weakest case everywhere.

**Our take:** Make the PDF the download button, and keep the answer itself on a normal web page. Google spells out how it handles PDFs, but the AI companies publish nothing at all about them, and their crawler pages only talk about web pages. A plain HTML page is the one format every engine documents. So we put the facts on a page, offer the PDF as a copy, and point the PDF's canonical at the page.

## What are the key facts about AI and PDF files?

| Fact | Detail | Source |
| --- | --- | --- |
| Google indexes PDFs | "Adobe Portable Document Format (.pdf)" is on Google's list of indexable file types | Google |
| PDF size limit | Googlebot crawls "the first 64MB of a PDF file," against 2MB for other supported files | Google |
| Google's AI features | A page must be indexed and eligible for a snippet; no extra technical requirements | Google |
| Links in PDFs | Treated "similarly to links in HTML"; nofollow is not possible inside a PDF (2011) | Google |
| PDF titles in results | Google uses the file's title metadata and the anchor text of links to it (2011) | Google |
| Scanned text | Google "may" use OCR on text stored as images; rule of thumb: if you can copy the text, it can be indexed (2011) | Google |
| Keep a PDF out of Google | Send an X-Robots-Tag: noindex HTTP header with the file | Google |
| Canonical for a PDF | A Link HTTP header with rel="canonical" works for non-HTML files such as PDFs | Google |
| OpenAI, Anthropic, Perplexity | Their crawler pages do not mention PDFs or file types: unconfirmed | OpenAI, Anthropic, Perplexity |
| Claude with an uploaded PDF | Each page is turned into an image and its text is extracted alongside it | Anthropic |
| US government policy | Agencies must "prioritize HTML and use PDFs only when necessary" | Section508.gov |

## Does Google index PDF files?

Yes. Google's list of file types it can index includes "Adobe Portable Document Format (.pdf)," grouped with Word, PowerPoint and other files that "require a specific parser to extract the human-readable text." Google decides the file type from the Content-Type header your server sends. [Google](https://developers.google.com/search/docs/crawling-indexing/indexable-file-types)

PDFs also get a bigger size allowance than web pages. Googlebot "crawls the first 2MB of a supported file type, and the first 64MB of a PDF file." Once it hits the limit it stops and only indexes what it already downloaded. [Google](https://developers.google.com/search/docs/crawling-indexing/googlebot) So a long annual report is unlikely to be cut short, but a 200MB scanned catalogue could be.

Google's most detailed PDF guidance is an older Search Central blog post from September 2011. It says Google has indexed PDFs since 2001, that PDFs can rank "similarly to other webpages," and that links in PDFs "can pass PageRank and other indexing signals." It also notes you cannot mark a link inside a PDF as nofollow. [Google](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results) That post is 15 years old, so treat it as Google's last word on the details, not a fresh promise.

> **In plain words:** Google can find, read and rank a normal text PDF. It is a real search result, not a dead end.

## Can a PDF show up in Google AI Overviews or AI Mode?

Probably yes, but Google has not said so directly. Its AI features page says a page only has to be "indexed and eligible to be shown in Google Search with a snippet," and that "there are no additional technical requirements." [Google](https://developers.google.com/search/docs/appearance/ai-features) The page does not mention PDFs or file types either way.

Since Google indexes PDFs and can show them in Search, an indexed PDF meets the stated bar. **Our advice, not a vendor statement:** we would expect a PDF to be usable as a source in AI Overviews and AI Mode the same way it is in Search, but we have no Google statement that confirms it for PDFs specifically.

The same controls apply. Google lists nosnippet, max-snippet and noindex as ways to limit what its AI features show. [Google](https://developers.google.com/search/docs/appearance/ai-features) For a PDF you can't add a meta tag, so those rules go in an HTTP header instead (see below). For the wider picture, read [how to get cited in Google AI Overviews](/blog/how-to-get-cited-in-google-ai-overviews/).

## Do ChatGPT, Claude and Perplexity read PDFs on my website?

Nobody outside those companies can say for sure. We read the official crawler pages from OpenAI, Anthropic and Perplexity on 12 October 2026. None of them mentions PDFs, file types or non-HTML documents. They only talk about web pages and content. So any claim that these bots do, or don't, read your PDFs is **unconfirmed** by the vendors.

What each page does say:

- **OpenAI.** OAI-SearchBot is "used to surface websites in search results in ChatGPT's search features." ChatGPT-User fetches pages for a user's request and is "not used for crawling the web in an automatic fashion." GPTBot collects content that may train OpenAI's models. [OpenAI](https://developers.openai.com/api/docs/bots)
- **Anthropic.** Claude-SearchBot "navigates the web to improve search result quality." Claude-User visits sites when people ask Claude questions. ClaudeBot collects content for training. [Anthropic](https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler)
- **Perplexity.** PerplexityBot is "designed to surface and link websites in search results on Perplexity." Perplexity-User fetches a page when a user asks a question. [Perplexity](https://docs.perplexity.ai/guides/bots)

One thing is documented: these assistants can read a PDF a person hands them. Anthropic's developer docs say that when you give Claude a PDF, the system "converts each page of the document into an image" and extracts the text alongside it. [Anthropic](https://platform.claude.com/docs/en/build-with-claude/pdf-support) That is about uploaded files, though, not about a bot crawling your website. Don't confuse the two.

We could not find a Bing page that says how Bing handles PDFs, so we make no claim about Bing or Copilot here.

## Can AI read a scanned PDF?

Sometimes, and you should not count on it. Google's 2011 guidance says it "may process the images with OCR algorithms" when a PDF's text is stored as pictures. Its rule of thumb: if you can copy and paste the text into a text document, Google should be able to index it. It also says images inside PDFs were "not indexed" and suggests making HTML pages for images you want found. Password-protected or encrypted PDFs are excluded. [Google](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results)

A scanned menu or flyer is just a photo of words. A crawler that does not run OCR sees an empty file. Do the copy-paste test yourself: open the PDF, try to select a sentence. If you can't, neither can most software.

The fix is the same one that helps blind visitors. WebAIM explains that accessible PDFs are usually "tagged" PDFs, which carry the document's structure for screen readers, and that "an untagged PDF would not be considered 'accessible.'" [WebAIM](https://webaim.org/techniques/acrobat/) The W3C's PDF techniques add that images in tagged PDFs should carry an alternate description, text "that can be vocalized by text-to-speech technology." [W3C](https://www.w3.org/WAI/WCAG22/Techniques/pdf/PDF1)

**Our advice, not a vendor statement:** export PDFs from the original Word or design file, not from a scanner. If a scan is all you have, run it through OCR to add a real text layer, then check the result. Our guide to [whether AI can see your images](/blog/can-ai-see-your-images/) covers the same problem for photos on web pages.

## Should I publish an HTML version of my PDF?

Yes, for anything you want AI answers to quote. Google's own guidance says: "Whenever possible, we recommend serving a single copy of your content." If you keep both, tell Google which one you prefer, for example by listing it in your sitemap or by "specifying the canonical version in the HTML or in the HTTP headers of the PDF resource." [Google](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results)

Google's current canonical docs confirm the header method works for "non-HTML documents such as PDF files." You add a Link header to the PDF's response that points at the preferred URL, using an absolute address. Google supports this "for web search results only." [Google](https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls) A header for a PDF whose main version is a web page looks like this:

`Link: <https://www.example.com/menu/>; rel="canonical"`

The US government reached the same view for its own sites. Section508.gov says federal policy requires agencies to "prioritize HTML and use PDFs only when necessary." [Section508.gov](https://www.section508.gov/create/pdfs/)

**Our advice, not a vendor statement:** the web page should hold the full answer: prices, hours, specs, the key numbers. Put a clear "Download the PDF" link on that page, and link back from the PDF to the page. Because the AI companies only describe their bots in terms of web pages, an HTML page is the safer bet for ChatGPT, Claude and Perplexity. If you want a lighter text copy for AI crawlers as well, see [Markdown pages for AI crawlers](/blog/markdown-pages-for-ai-crawlers/).

## How do I stop a PDF from showing up in search and AI answers?

Send an X-Robots-Tag header with the file. Google says to use it "to block indexing of non-HTML resources, such as PDF files," because you can't put a robots meta tag inside a PDF. Any rule you can use in a robots meta tag also works in this header. [Google](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag)

Google gives server examples. On Apache, a `<Files ~ "\.pdf$">` block with `Header set X-Robots-Tag "noindex, nofollow"`. On NGINX, a `location ~* \.pdf$` block with `add_header X-Robots-Tag "noindex, nofollow";`. [Google](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag)

Two traps:

- **Don't block the PDF in robots.txt as well.** Google says that if crawling is blocked, indexing rules "will not be found and will therefore be ignored." [Google](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag) The file must stay crawlable for noindex to work.
- **noindex is a Google rule.** OpenAI, Anthropic and Perplexity document robots.txt controls for their bots, not X-Robots-Tag. To keep a PDF away from them, block the path for their user agents in robots.txt. [OpenAI](https://developers.openai.com/api/docs/bots) [Anthropic](https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler) [Perplexity](https://docs.perplexity.ai/guides/bots)

The most common real problem is the opposite: an old price list PDF still indexed and still being quoted. Noindex it, or better, replace it at the same URL with current numbers.

## What should be on my PDF checklist for AI search?

| Check | Done when | Source |
| --- | --- | --- |
| Real text, not a scan | You can select and copy a sentence from the PDF | Google |
| No password or encryption | The file opens without a password | Google |
| Correct file type | Your server sends the PDF with a PDF Content-Type header | Google |
| Title metadata set | The PDF's Title property names the document clearly | Google |
| Descriptive links to it | Links say "2026 price list (PDF)", not "click here" | Google |
| Under the size limit | The PDF is well under 64MB; the key facts are early | Google |
| Tagged and accessible | The PDF is tagged and images have alternate text | WebAIM; W3C |
| HTML page holds the answer | A normal web page has the same key facts and links to the PDF | Google; Section508.gov |
| Canonical set | Duplicate PDFs send a Link rel="canonical" header to the page | Google |
| Old PDFs handled | Outdated PDFs are replaced or send X-Robots-Tag: noindex | Google |

## How do I check whether AI can find my PDFs?

Start with Google, because it is the only engine that documents PDF handling. Search `site:yourdomain.com filetype:pdf` to see which of your PDFs Google has indexed; Google documents the `filetype:` operator on its file types page. [Google](https://developers.google.com/search/docs/crawling-indexing/indexable-file-types) Look for old versions that should be gone, and for titles that come out as file names like "scan_0042.pdf".

Next, check your server logs to see whether GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot have ever requested a .pdf URL. Our guide to [finding AI bots in your server logs](/blog/how-to-see-ai-bots-in-your-server-logs/) shows how. If they only ever fetch your web pages, that's one more reason to put the answer on a page.

Our free [AI visibility check](/#tool) reads your pages the way AI crawlers do and flags pages that come back empty or blocked. Then try the [Prompt Simulator](/prompt-simulator/): ask the question your PDF answers, such as "what does a kitchen repaint cost at...", and see which source each AI engine cites. If it quotes an old PDF, or nothing of yours, you know which page to build next.

## FAQ

### Can ChatGPT read PDFs on my website?

Unconfirmed. OpenAI's crawler page describes OAI-SearchBot, ChatGPT-User and GPTBot but never mentions PDFs or file types. To be safe, put the key facts on a normal web page and offer the PDF as a download. [OpenAI](https://developers.openai.com/api/docs/bots)

### Does Google index PDF files?

Yes. PDF is on Google's list of indexable file types, and Googlebot crawls the first 64MB of a PDF, compared with 2MB for other supported files. [Google](https://developers.google.com/search/docs/crawling-indexing/indexable-file-types) [Google](https://developers.google.com/search/docs/crawling-indexing/googlebot)

### How do I noindex a PDF?

Send an X-Robots-Tag: noindex HTTP header with the file, and keep it crawlable in robots.txt so Google can see the header. Google gives Apache and NGINX examples. [Google](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag)

### Is having the same content in HTML and PDF duplicate content?

Google recommends one copy where possible. If you keep both, mark the preferred version, for example with a rel="canonical" Link header on the PDF that points to the web page. [Google](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results) [Google](https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls)

### Can AI read a scanned PDF?

Not reliably. Google said in 2011 it may use OCR on text stored as images, with the rule of thumb that copyable text can be indexed. Other AI crawlers publish nothing on this. Add a real text layer and tag the PDF. [Google](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results) [WebAIM](https://webaim.org/techniques/acrobat/)

## Sources

1. [File types indexable by Google](https://developers.google.com/search/docs/crawling-indexing/indexable-file-types) — Google Search Central. Read Oct 12, 2026.
2. [Googlebot](https://developers.google.com/search/docs/crawling-indexing/googlebot) — Google Search Central. Read Oct 12, 2026.
3. [PDFs in Google search results (September 2011)](https://developers.google.com/search/blog/2011/09/pdfs-in-google-search-results) — Google Search Central Blog. Read Oct 12, 2026.
4. [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) — Google Search Central. Read Oct 12, 2026.
5. [Robots meta tag specifications](https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag) — Google Search Central. Read Oct 12, 2026.
6. [How to specify a canonical with rel="canonical" and other methods](https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls) — Google Search Central. Read Oct 12, 2026.
7. [Overview of OpenAI Crawlers](https://developers.openai.com/api/docs/bots) — OpenAI. Read Oct 12, 2026.
8. [Does Anthropic crawl data from the web, and how can site owners block the crawler?](https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler) — Anthropic. Read Oct 12, 2026.
9. [PDF support](https://platform.claude.com/docs/en/build-with-claude/pdf-support) — Anthropic. Read Oct 12, 2026.
10. [Perplexity Crawlers](https://docs.perplexity.ai/guides/bots) — Perplexity. Read Oct 12, 2026.
11. [PDF Accessibility: Defining PDF Accessibility](https://webaim.org/techniques/acrobat/) — WebAIM. Read Oct 12, 2026.
12. [PDF1: Applying text alternatives to images with the Alt entry in PDF documents](https://www.w3.org/WAI/WCAG22/Techniques/pdf/PDF1) — W3C. Read Oct 12, 2026.
13. [Create accessible PDFs](https://www.section508.gov/create/pdfs/) — Section508.gov (US GSA). Read Oct 12, 2026.