The most common reason when Perplexity doesn't read the file you upload is usually not that the model is too stupid, but that the file doesn't pass through the portal first. Following the instructions in Perplexity's current Help Center, look at three things: whether the file is more than 40MB, whether the file type is supported, and whether the content triggers moderation restrictions.
This kind of problem is difficult to detect because many times the upload itself seems to have been successful, but the system does not necessarily digest the entire file in the way you imagine. Especially for long files, Perplexity prioritizes the parts it thinks are most important, rather than guaranteeing that they are read word by page. So "it didn't answer the paragraph I wanted to ask" does not necessarily mean that the document was completely missing.
First, in this order:
1. Confirm that the file is less than 40MB.
2. Check whether the file type is within the supported range, and do not default to stable resolution of all formats.
3. If it's images or complex media content, consider content moderation restrictions as well.
4. Try again with a smaller, cleaner version, such as removing irrelevant pages, compressing tables, or exporting plain text.
If you're uploading a long research report or a large PDF, don't just ask a particularly general question. Perplexity is better suited for "reading with questions" for long documents, such as direct roll call page numbers, chapters, terms, or target conclusions. The more general you ask, the easier it is to catch only the high-weight summaries that the system identifies, not the details you really care about.
There are also scenarios where the file is not broken, but the file is too cluttered. For example, scanning PDFs, more images than text, complex typography, and many nested tables can reduce the quality of parsing. In this case, saving key pages as a separate text or table is often more effective than repeatedly retransmitting the original.
Therefore, if Perplexity cannot read the file, the first reaction should not be to "change the model", but to look back at the entry conditions and the way to ask questions. Whether the file can be read stably or not is often determined at the upload step.
