Key takeaways
- Chunk long documents by section rather than uploading everything at once.
- Ask for page-referenced answers so you can verify them.
- Scanned PDFs need OCR before any model can help.
Summarising a long document is one of the genuinely useful things AI does, provided you treat the output as a map rather than the territory. Here is how to get a summary you can trust.
1Check the file has text
Open the PDF and try to select a sentence. If nothing highlights, the document is a scan. Run it through OCR first, otherwise no model can read it properly.
2Decide what kind of summary you need
Three different jobs, three different prompts:
- Orientation: "Summarise this report in 150 words for someone who has not read it."
- Decision: "List every recommendation in this document with the evidence given for it."
- Extraction: "Extract all pricing terms into a table with page references."
Generic "summarise this" produces a generic answer. Name the job.
3Chunk by structure
For anything over about forty pages, work section by section. Upload or paste a chapter, ask for a summary with page references, then combine the section summaries yourself. This produces a more accurate result than one pass over 200 pages, where detail gets compressed away.
4Demand page references
Ask for the page number beside every claim. It makes verification fast, and it discourages invention. If the answer cannot cite a page, treat it as unverified.
5Verify the numbers
Models transcribe numbers well and interpret them badly. Check every figure, date and name against the original. This is the single most important step and the one people skip.
6Ask what is missing
A good final prompt: "What questions does this document leave unanswered?" The gaps are often more useful than the summary, particularly for contracts and research papers.
Which tool
Claude handles long documents most reliably. ChatGPT is close and better if you also want image or chart interpretation. Perplexity is best when you need the summary tied to external sources. Elicit is the specialist choice for academic papers.
Tools mentioned in this guide
Every tool below has a full review with current pricing and limits.
Frequently asked questions
What is the page limit for uploads?
It varies by tool and plan, typically a few hundred pages. Chunking by chapter produces better summaries regardless of the limit.
Can it read a scanned document?
Only if the text layer exists. Run OCR first, otherwise the model is looking at pictures of words.
Written by
Scout Guides Desk
The Guides Desk writes the step-by-step material: how to brief a tool, where it fails, and which parts of a workflow still need a human. Every guide is written from a real task rather than a feature list.
Our editorial policy