Yes, ChatGPT can read PDFs. You upload the file and ask questions about it, request a summary or pull out specific details. It works best on text-based PDFs with clear layouts. Scanned pages, complex tables and very long documents are where it slips, so check anything important against the original before you rely on it.
Yes, with conditions
Uploading a PDF and asking ChatGPT questions about it ranks among the most useful everyday things you can do with it. Supplier terms, a long report, a policy document, meeting packs: instead of reading every page to find one answer, you ask. Done well, it saves a lot of time. Done carelessly, it gives you a confident summary of something the document doesn't quite say.
The difference is knowing what kind of file you're giving it and what kind of question you're asking.
What works and what trips it up
- Text-based PDFs work best. If you can select and copy the words in your PDF viewer, the text is there for it to read. Reports exported from Word or Google Docs are usually fine.
- Scanned documents are riskier. A scan is a picture of a page. It may be able to read the image, but blurry scans, handwriting, stamps and skewed pages raise the odds of misread words. Check names and figures carefully.
- Long documents get uneven attention. On a long file, it can lean on some sections and skim others. Ask about specific sections rather than everything at once.
- Tables need care. Columns can get misaligned when the layout is complex, so a value can end up attached to the wrong row. Always check table figures against the page.
- Charts and diagrams are hit and miss. It may describe them, but don't rely on it for exact readings.
- Confidential files need a decision first. Before uploading contracts, staff records or client data, check how your account handles what you share. I've covered this in is ChatGPT private.
Usage caps on uploads and messages vary by plan and change over time, so if you're hitting limits, my guide to ChatGPT usage limits explains what to look for.
Worked examples you can copy
The prompt matters as much as the file. Vague questions get vague summaries. Here are two common situations and the prompts I'd type.
Say you have a supplier contract
You want to know what you're signing up to, in plain English, before you talk to the supplier or your adviser. Upload the contract and try:
Summarise this contract in plain English for a small business owner. List the notice period, renewal terms, payment terms and any penalties. For each point, quote the exact clause and give its heading so I can check it.
Then follow up with:
What in this contract could cost me money or lock me in that I might miss on a quick read? Only use what's in the document. If something isn't covered, say so.
This gives you a list of questions to raise, not legal advice. Anything that matters goes to a qualified person, with the clauses it pointed you to.
Say you have a long policy or tender document
You need to know whether you qualify and what you'd have to provide, without reading every section twice. Upload it and try:
List every requirement a supplier must meet to apply, grouped by section. For each one, give the heading and quote the wording. Then list the documents I'd need to submit.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Then, because long files get uneven attention, go section by section:
Now look only at the section on evaluation. How will applications be judged? Quote the relevant lines.
Splitting the questions like this keeps it focused on one part at a time, which is where long documents most often go wrong. Tick each requirement off against the original before you build your response around the list.
Say you have a competitor's annual report
You want the strategy, not every page. Try:
Read this annual report and tell me what the company says its priorities are for the coming year, where it says it's investing, and what risks it highlights. Quote the sentence you're relying on for each point.
Then:
Which statements here are the company's plans and which are results it reports? Keep them separate.
Asking for quotes is the key move in both examples. It forces the answer back to the page and makes checking quick. For more on writing prompts like these, see my AI prompt writing guide for small businesses.
Checklist before you trust the summary
- Spot-check the quotes. Search the PDF for each quoted line. If one isn't there, treat the whole answer with suspicion.
- Check every figure against the page. Especially anything pulled from a table.
- Look for what's missing. Ask: what does this document say that your summary left out? Summaries drop caveats.
- Watch for outside knowledge. If an answer includes something you can't find in the file, it came from somewhere else. Ask it to stick to the document.
- Read the important sections yourself. The summary tells you where to look. It doesn't replace looking.
Where I'd draw the line
Use ChatGPT to find your way around a PDF quickly, pull out questions and get a first summary. Don't let it be the last reader of anything you're signing, publishing or reporting on. And if your PDF is mostly data, export the table to a spreadsheet and analyse that instead; my guide to using ChatGPT for data analysis shows how.
