What Google Said About PDFs in Search This Month

Every PDF uploaded to a WordPress media library becomes a public file at a /wp-content/uploads/ address with no template, no meta description and no navigation. Whether Google indexes it, how it ranks, and what happens when two copies exist are questions that rarely get asked until a price list from 2023 outranks the current product page. Google’s Search Off the Record podcast covered exactly this in an episode titled “PDFs in Search: Indexing, AI & Pitfalls” with Martin Splitt and John Mueller, and Search Engine Roundtable published a summary on 13 October 2026.

The summary is not a transcript, so everything below attributed to the episode is the summariser’s account of it. Statements from Google’s own documentation are marked as such.

PDFs are converted to HTML before indexing

According to the summary, PDFs are binary container files rather than plain text, and a converter in the indexing pipeline turns them into an HTML representation, so internal systems can treat them much like web pages. Text links inside a PDF are preserved and crawled. Images inside a PDF have no standalone URLs, so they generally cannot be processed into Google Images the way ordinary web images can. Converted PDFs are then ranked and indexed by the same core systems as HTML documents. One consequence is that a diagram that exists only inside a PDF is unlikely to turn up in Google Images. The summary also says PDFs can appear in AI Overviews and are accessible to AI models for grounding and training, but it notes that Google said little on that point, so it is thinly supported.

Size limits, and what Google actually documents

The episode said the PDF crawl limit is larger than the limit for HTML pages and that Google does not publish an exact cutoff for PDFs. Google’s overview of its crawlers and fetchers, last updated 2026-06-12 UTC, says that by default its crawlers and fetchers crawl only the first 15MB of a file, that a crawler can set a smaller limit, and that it may set a larger limit for a PDF than for HTML. The page gives no PDF figure. Numbers quoted in older coverage should be treated as superseded by that page.

The pitfalls the episode listed

The summary groups the problems into a handful that are easy to check on your own site:

• Scanned or rasterised PDFs cannot be read without OCR. The summariser’s test is to copy text out of the PDF into a text editor, and if nothing pastes, there is nothing for Google to read.
• Password-protected PDFs block search bots from reading or indexing the content.
• Identical PDFs are exact binary duplicates, which makes canonicalisation across domains difficult.
• PDFs are treated as relatively static, like images, and are recrawled less often than HTML, so the summariser advises publishing an update under a new file name or URL.
• PDFs must download fully before they render on mobile, and fixed layouts do not reflow when zoomed.
• Print versions generated by a CMS can compete with the live page and go out of date because they are recrawled less.

What to do on a WordPress site

The summariser’s own recommendation, not a Google statement, is to default to HTML for editability, mobile behaviour and faster indexing, and to keep PDFs for rigid print formatting, offline forms or paper documentation. A restaurant menu as a PDF is acceptable where an HTML version is impractical.

Where a PDF has to stay, two controls apply. A PDF cannot carry a meta robots tag, so keeping an outdated copy out of the index means an HTTP header, which is covered in the comparison of robots.txt, meta robots and X-Robots-Tag. The summary also lists rel=canonical sent as an HTTP header, PDF metadata and prominent links back to the site as ways to reduce duplicate problems, and the general failure modes of canonical signals are in the post on common canonical tag mistakes.

The first job is an inventory. Search the site for filetype:pdf and compare the results with the Media Library, then open the oldest PDF that still ranks and ask whether the current page says the same thing. The full episode is available on the Search Off the Record feed, and a PDF that is out of date and still indexed is cheaper to fix than to explain to a customer.