What do the official documents say about PDF?
Only Google’s documentation explicitly covers PDF. The OpenAI and Perplexity crawler pages we read on 2026-11-09 do not mention PDF, so we write “not stated officially” and do not guess. None of the pages below shows a “last updated” date, so the dates are the dates we viewed them.
| Product | What the official documentation says | Status |
|---|---|---|
| Google Search | The “file types Google can index” page lists PDF among the encoded file types that need a dedicated parser, together with Word, Excel, PowerPoint, EPUB and others | indexable-file-types page, read 2026-11-09 |
| Google Search | Googlebot fetches the first 64 MB of a PDF and the first 2 MB of other supported files, measured on uncompressed data; anything beyond that is not passed to indexing | googlebot page, read 2026-11-09 |
| ChatGPT search | The crawler page covers OAI-SearchBot and other crawlers and robots.txt; it does not mention PDF or any file type | OpenAI crawler page at developers.openai.com, read 2026-11-09; PDF handling: not stated officially; the help center article returned 403: not verified |
| Perplexity | The crawler page does not mention PDF or file types | Perplexity crawler page at docs.perplexity.ai, read 2026-11-09; PDF handling: not stated officially (only this one page was checked) |
So only Google gives a clear answer to “can AI cite a PDF”: it can index one. The others make no written commitment, so do not assume they will.
What are Google’s specific rules for PDF?
All four come from Google’s Search documentation, read 2026-11-09:
| Question | What the documentation says |
|---|---|
| How it decides a file is a PDF | Generally the Content-Type response header; if the header is missing or wrong it may go by the file extension or retry with a different parser |
| How much it reads | The first 64 MB of a PDF; beyond that Googlebot stops the fetch and passes only the part already downloaded to indexing |
| Keeping it out of the index | For non-HTML files use an X-Robots-Tag response header with noindex; the documentation gives Apache and NGINX examples for PDF |
| Duplicating an HTML page | rel=canonical can be sent in an HTTP header; the documentation says it applies to non-HTML documents such as PDF |
# PDF served with the right type, a canonical pointing to the HTML page HTTP/1.1 200 OK Content-Type: application/pdf Link: <https://example.com/whitepaper/>; rel="canonical" # Keep a PDF out of the index (Google's documented way for non-HTML files) X-Robots-Tag: noindex
For general handling of duplicates see Canonical tags and duplicate content: how to set them up on a multilingual site so pages are not treated as duplicates. A file blocked by robots.txt cannot have its response header read by the crawler, so if you rely on noindex do not also block it with robots.txt; for robots.txt syntax see How do you write robots.txt for AI crawlers?.
Should you put content in a PDF?
Put the content on an HTML page and treat the PDF as a downloadable version. The reason is that only Google’s documentation states PDF support, the others do not say, and an ordinary web page does not carry that uncertainty. Here is how to handle some common kinds of content:
| Content | Recommendation | Cause |
|---|---|---|
| Whitepapers and research reports | Summary, key figures and sections on an HTML page, with the PDF as a download | The summary and figures sit in the body, so AI does not have to parse a file to quote them |
| Product manuals and spec sheets | Make an HTML table and attach the PDF separately | Specs and comparisons are easiest to read as tables |
| Developer documentation | HTML only | Documentation changes often and PDFs go stale |
| Conference papers and academic PDFs | Keep the PDF, put title, authors and abstract on a page, and link to the PDF | Google says it can index PDF; the others: not stated officially |
| PDFs behind a form | Make a summary page public; the file itself is a separate matter | Crawlers do not fill in forms; the summary page can at least be read |
For how to write content that is easier to quote see How to write content AI will quote: before and after; for developer documentation see How should developer docs be written so AI can understand and cite them?.
How do you make a PDF as readable as possible?
1. First make an HTML summary page
On the page, write the PDF’s title, a short summary, the key figures and the download link; link to the PDF’s address from this page.
2. Confirm the PDF has a text layer
Inference: a PDF made from scanned images has no copyable text, so a parser has nothing to extract. If you can select and copy text in a PDF reader, it has a text layer.
3. Set the Content-Type response header to application/pdf
Google’s documentation says it mainly goes by this header to decide the type.
4. Keep the file within the limit
Google reads the first 64 MB. Important content late in a large file may not make it into the index; put the key points first or split the file.
5. For duplicates, point to the HTML page in an HTTP header
A rel=canonical in a Link header, which applies to PDF.
6. Use X-Robots-Tag for PDFs you do not want indexed
Do not rely on robots.txt alone.
7. Use logs to confirm crawlers actually fetched it
Look in the server logs for crawler requests for this .pdf and the status code returned; see How do HTTP status codes affect AI crawlers? 404, 410, 301, 429, 5xx.
We found no AI product that publicly explains how content inside a PDF is chosen for an answer, and no statistics comparing citations of PDF and HTML, so the advice here is “the safer way”, not a known-to-work way. If your page is built by script, see Do AI crawlers render JavaScript?.
FAQ
Will AI cite a PDF?
Google’s documentation explicitly lists PDF as a file type it can index. The OpenAI and Perplexity crawler pages we read do not mention PDF and the ChatGPT help article could not be read, so whether those two fetch and cite PDFs: not stated officially.
Should content live only in a PDF?
Not recommended. Only Google officially states PDF support; the others do not say. Putting the content on an HTML page and treating the PDF as a downloadable version is safer than offering only a PDF.
Is there a size limit for PDFs?
Google’s Googlebot documentation says it reads the first 64 MB of a PDF and the first 2 MB of other supported files, measured on uncompressed data, and anything beyond is not indexed. Whether other AI crawlers have limits: not stated officially.
If a PDF and an HTML page have the same content, is that duplicate content?
You can point the PDF at the HTML page with rel=canonical in an HTTP header. Google’s documentation says this applies to non-HTML documents such as PDF. Whether other search systems honor it: not stated officially.
What if I do not want a PDF indexed?
Use an X-Robots-Tag: noindex response header. Google’s documentation says non-HTML files such as PDF, video and image use the response header rather than a meta tag. Note the crawler must be able to fetch the file: a file blocked by robots.txt cannot have the header read.
Sources
- Google · File types indexable by Google (Checked 2026-11-09; the page shows no update date)
- Google · Googlebot (Checked 2026-11-09; the page shows no update date)
- Google · Robots meta tag, data-nosnippet, and X-Robots-Tag specifications (Checked 2026-11-09; the page shows no update date)
- Google · How to specify a canonical URL (Checked 2026-11-09; the page shows no update date)
- OpenAI · Overview of OpenAI Crawlers (Checked 2026-11-09; the page shows no update date; the page does not mention PDF)
- Perplexity · Perplexity Crawlers (Checked 2026-11-09; the page shows no update date; the page does not mention PDF)
- kagentx evaluations: the data on the case study pages; each example gives its date.