Design systems for every user including people and LLMs | ODSP916
PDFs are now consumed by LLMs too, so they must be made accessible and structured for machine reading.
“What we see is very easy for us because we are trained up on it but LLMs don't have this and so it's very hard for them to interpret the actual meaning behind those pixels.”
A PDF SDK engineer argues that accessibility principles (perceivable, operable, understandable, robust) now serve both humans and LLMs, since LLMs see only pixels and struggle to recover semantic structure and reading order from PDFs. The practical takeaway is to extract structured text before feeding documents to models rather than relying on costly vision-based training. It's a useful engineering perspective on document-to-LLM pipelines but not a major industry announcement.