Which tool to choose
Almost all AI assistants read PDF and Word, but they react differently to difficult files.
- If your file is a real document (created by Word, by a management software, saved as PDF by a program), practically every AI reads it. The problem, if there is one, is elsewhere.
- If the file is a scan or a photo of a paper document, you need an AI with good "image-reading" ability, or you have to convert it to text first. The AIs that see images well manage scans too; the others don't.
Before changing tools, though, understand what kind of file you have: half the problems come from the file, not the AI. Below you'll find how to recognize it.
How to do it
From a computer or a phone the principle is the same: first identify the type of file, then act.
- Figure out whether your PDF is "real" or "scanned". Open the PDF and try to select a sentence with the mouse (or by holding your finger on the phone): if the text highlights, it's a real PDF; if you can't select anything, it's a photograph of text and the AI struggles to read it.
- If it's scanned, convert it to text. Search for "free OCR online" (OCR means optical character recognition: it turns the photo of a text into real text). Upload the PDF there, download the readable version, and give that to the AI.
- If it's a Word the AI can't digest, save it as PDF (from Word: save menu, choose PDF format) and upload the PDF. PDF is the format the AIs handle best.
- If it really won't work, bypass the file: open the document, select and copy the text, and paste it directly into the chat. The operating syntax:
I can't get you to read the attached file. I'm pasting the text of the
document below. Work on it as if it were an uploaded file:
[paste the document text here]
- Check: after uploading the file, ask the AI for a reading test: "tell me what the title and the first sentence of the document are". If it reports them correctly, it's really reading; if it makes things up or says it sees nothing, the file didn't arrive as text.
A concrete example
Paola receives a contract in PDF and asks the AI to summarize it. The AI replies that the document seems empty. Paola tries to select a sentence in the PDF: nothing highlights. It's a scan.
She uploads the PDF to a free OCR service, downloads the readable version, re-uploads it to the AI. This time the summary arrives, accurate. The problem wasn't the AI but the file: a photograph of text that no assistant could read until it became real text.
When it does NOT work (and how to fix it)
If the AI says the document is empty or unreadable
It's the classic sign of a scan. Do the selection test: if you can't highlight the text, run the file through an OCR before re-uploading it. No AI reads a photograph of text while it stays a photograph.
If the file is password-protected
The AIs reject protected or encrypted files. Open the document with the password, then save a copy without protection (in many programs it's an option in the save or in the print-to-PDF) and upload that. Never give the AI the password: remove the protection yourself, locally.
If the Word has complicated tables or images
Sometimes the AI reads the text but gets the tables wrong or ignores the images. Save the file as PDF before uploading it: often the PDF preserves the layout better. If the tables remain a problem, copy and paste the table directly into the chat, row by row.
If the upload stalls or gives a technical error
Here the culprit isn't the file but the browser. Privacy extensions, ad blockers and some VPNs interfere with the upload. Try reloading the page, using an "incognito" browser window (without extensions), or another browser. Often the file goes through on the first clean attempt.
A tip from someone who actually uses it
Before fighting with a stubborn file, spend two seconds on the text selection test. It's the single check that solves most cases: if the text won't select, you already know you need OCR and you stop blaming the AI. Telling a real PDF from a scan saves you half an hour of blind attempts.
Frequently asked questions
Why does one PDF read and another doesn't, if they're both PDFs?
Because "PDF" is just the container. Inside there can be real text (saved by a program) or a photograph of text (a scan). The first reads right away, the second is an image and has to be converted to text first with OCR. Same container, opposite content.
Is online OCR safe for confidential documents?
It depends on the service and the document. For a flyer any site is fine; for a contract or a document with personal data, choose a service that states it deletes the files after conversion, or use the offline OCR feature of some programs on your computer. The rule is: don't upload to an unknown site what you wouldn't want getting out.
Is it better to upload the file or paste the text?
If the file uploads and the AI reads it well, upload it: it's more convenient and preserves the structure. If it gives problems, pasting the text is the most reliable route, because it skips all the upload hitches. The only limit of pasting is length: for very long documents the file is still better.
If the AI says it read the file, can I trust that it read all of it?
Don't take it for granted, and this is where many get burned. On long files the AI sometimes reads only the first pages and answers as if it had everything. Put it to the test: ask for something that's on the last page ("what does the document say about [a final detail]?"). If it can't find it, it didn't read all the way to the end, and that "complete" summary was only half complete.