RAG for PowerPoint: slide text converts, pictures of text do not
PowerPoint slide text lives in the .pptx as XML and extracts cleanly for RAG; only the screenshots and diagrams pasted onto slides need OCR.
PowerPoint is the format people most expect to fail
It looks visual, so the assumption is that a converter will need to read the slides like a person does. It does not. A .pptx is a zip of XML in exactly the way a .docx is, and every text box, title and bullet is stored as text.
Speaker notes are often the best content on the slide
Slides are deliberately terse - three words and a chart. The notes pane frequently holds the actual explanation, in full sentences, which is far better material for retrieval than the bullets. If your converter reads notes, a deck becomes much more useful than its slides suggest.
The parts that need OCR
Anything pasted as an image: architecture diagrams, screenshots of dashboards, charts exported from another tool. These are pixels on a slide and carry no text, so they extract nothing without an OCR engine - the same limit that applies to a scanned PDF.
Chunking a deck needs care
One slide is usually too small to be a useful passage, and a whole deck is too large. Grouping consecutive slides until a passage reaches a sensible size, with overlap between groups, produces far better retrieval than one chunk per slide.