PDFs in Google Search and AI Answers: What They're Actually Good For

Google’s Search team devoted an entire episode of Search Off the Record to PDFs — indexing, AI, and pitfalls — with a chapter list that is more useful than most SEO checklists. It covers how PDFs get converted, how links and images inside them are extracted, size limits, whether they rank in Search and AI answers, and the recurring “restaurant menu” argument about whether a PDF should exist at all.

Here is the practical summary for anyone who has been told to turn their guide, case study or menu into a PDF to “get it indexed.”

How Google actually handles a PDF

A PDF is not a text document; it is a binary container. During indexing, Google runs it through a converter that produces an HTML representation, which is what the rest of the pipeline handles — the same way it handles a web page. That has three consequences people rarely plan for.

Text links inside the PDF are extracted. Google has previously confirmed that links within PDF documents are treated as links and pass ranking signals. So a PDF is not a black hole for link equity the way a scanned image is.

But it is refreshed slowly. Google has said it indexes PDFs but updates them slowly. If you revise the file in place at the same URL, expect the indexed version to lag for a long time.

And the formats that look best are the worst for extraction. A rasterised page — a designed layout exported as one big image — is nearly opaque. So is a scan. The version that reads beautifully in a PDF viewer may contain no machine-readable text and no extractable links at all.

On limits: the general Googlebot file limit is 15MB, but it is 64MB for PDFs and 2MB for other file types. PDFs have historically been the most generous format Google will accept, which is exactly why they accumulate.

The honest answer on ranking

PDFs do rank, and they can be cited. But the format is working against you on three fronts: no mobile-friendliness, canonicalisation problems when the same document exists as both HTML and PDF (“duplicate binary blobs,” as the chapter list puts it), and accessibility problems from rasterised or scanned pages.

The canonicalisation issue is the one that costs people most. Publishing the same guide as a web page and a downloadable PDF creates two near-duplicates with no reliable way to tell Google which one you consider primary. Google has said it generally handles this, but “generally handles” is not the same as “give me your best page.”

When a PDF is the right choice

The last row is the one that matters for off-site work. When you submit to a directory, a profile or a marketplace, you are often asked for a logo, a media kit, sometimes a one-page PDF. Those assets are fine. The link that represents you should point at a page you control — an HTML page with a visible anchor and a description — because that is the thing a search engine and a language model can read, quote and re-crawl cheaply.

The submission side of that workflow is where most of the friction lives: each platform wants its assets in a slightly different form, and doing that by hand across dozens of directories is where the plan usually stalls. That is the part https://backlinksubmit.org takes over — AI-matching relevant platforms from a handpicked directory list, submitting manually with written copy rather than generated copy, and showing you in a console which platforms have been submitted and how each one is progressing. It also makes a specific promise on indexing: about 90% of submissions get indexed, and any listing a directory never indexes is resubmitted or replaced at no extra cost.

A simple rule

Use PDFs for artefacts that need to be frozen — signed documents, specifications, decks — and use HTML for anything you want to be found, quoted or re-read. If you must publish the same content both ways, put the HTML first in your plan and treat the PDF as the download.

And if your PDF is a designed, rasterised layout, assume it contributes nothing to search or to AI answers, because the text is not in it.

FAQ

Do links inside PDFs pass ranking signals?

Google has said links within PDF documents are treated as links. That makes a text-based PDF a valid vehicle, though a slower-moving one than a web page.

Can Google index a scanned PDF?

It converts what it can, which for a rasterised scan may mean no usable text and no extractable links. If the content matters, publish it as HTML.

Is there a size limit for PDFs?

Googlebot’s general limit is 15MB, with a higher limit of 64MB for PDFs specifically and 2MB for other file types.

Does publishing both an HTML page and a PDF cause duplicate content problems?

It creates canonicalisation questions, which Google says it usually resolves. Since you cannot tell it which version is primary, prefer one format for the content that needs to be found.