Google converts PDFs to HTML before indexing them, Splitt and Mueller say

A PDF a brand publishes on its site is not a different thing in Google's index. Martin Splitt and John Mueller, both on Google's Search Relations team, said on their 13 October Search Off the Record episode that Google runs every PDF through a converter that produces an HTML version, and the HTML version is what the indexing pipeline reads. The podcast is titled "PDFs in Search: Indexing, AI & Pitfalls".

What the converter does

Splitt described the step directly. "We have a converter," he said, "and we are taking that and producing an HTML version." After the conversion, the PDF ranks on the same signals as any other page, embedded links in the PDF are crawled like ordinary links, and the content is eligible for AI Mode and AI Overviews in the same way an HTML page would be. The practical consequence is that a PDF with selectable text behaves like a web page; a scanned PDF, where the text is a picture, does not reach the index at all without optical character recognition.

The limits the hosts named

Three limits came up in the episode. The crawl size allowance for a PDF is larger than for an HTML page, Mueller said, without naming a number, so a long document still fits. Images inside the PDF probably do not reach image search, and the hosts were explicit that this is Google's current behaviour and not a guarantee it stays. Two exact-duplicate PDFs make canonicalisation harder, because Google has to pick one and the two copies compete.

Mueller gave one direct recommendation, that HTML is "probably better" when a publisher has the choice, and used a restaurant menu as the example: an HTML menu loads faster on a phone, is easier to update, and does not need the converter step at all. The episode also warned against publishing an llms.txt file as a PDF, because llms.txt is meant to be plain text an AI model can read directly, and wrapping it in a PDF adds a conversion step the file exists to avoid.

What a team should do with this

A team deciding whether to publish a report, a datasheet, a specification, or a menu as a PDF now has a clear reason to publish the HTML version alongside it, or in its place when the audience will reach the content from a search result. Every signal Google uses on an HTML page, including structured data, canonical tags, mobile-friendliness, image alt text and the links the author wrote, has to survive the PDF-to-HTML conversion to be used, and the hosts did not promise all of them do. A PDF stays worth publishing when the format is the point, for example when the reader will print it or send it, and when preserving the layout matters. For discovery, the HTML version is the one that reaches the index without a middle step.

Source

PDFs in Search: Indexing, AI & Pitfalls, Search Off the Record. Roger Montti summarised the episode for Search Engine Journal.