PDFs in Search: Indexing, AI & Pitfalls

Should you publish your content as a PDF or stick strictly to standard HTML? Does Google treat PDF files differently when it comes to indexing, ranking, and AI search features? In this episode of Search Off the Record, Martin Splitt and John Mueller from the Google Search Relations team demystify how Google Search handles PDF documents. They break down some of Google's internal pipelines, explain why putting an LLMs.txt file inside a PDF is a bad idea, explore common technical traps like rasterized pages and canonicalization issues, and share practical advice on managing PDFs for SEO, AEO (SEO), and user experience. In this episode, you’ll learn: Google's PDF converter pipeline: How Google converts binary PDF containers into HTML pages during indexing so the rest of the pipeline can process them normally. LLMs.txt as a PDF: Why saving an LLMs.txt file as a PDF is a bad idea, even if you like LLMs.txt. File limits & Project Gutenberg: Why crawling limits exist, and how HTML and PDF size limits function in the crawl pipeline. Rasterized pages & accessibility: How pages saved as images destroy text readability, ruin screen reader accessibility, and prevent Google from extracting text content. Canonicalization & "binary blobs": Why identical PDF downloads across different sites make canonicalization harder and how using rel="canonical" HTTP headers or embedded header links can help. PDFs for SEO in AI Overviews: How PDFs are parsed, ranked, and referenced in Google Search, AI Overviews, and broader AI systems for training and grounding. Key takeaways for SEOs & developers: HTML is preferred, and PDFs work: HTML offers better mobile responsiveness, zoom reflow, and faster loading. Internally, Google converts indexable PDF text into HTML so they can rank normally and appear in AI features. Test for text indexability: Highlight and copy text out of your PDF into a text editor; if text doesn't copy, it might be a rasterized image and can't be indexed easily. For restaurant menus: A PDF menu is better than no menu…