The best AI for summarising PDFs in 2026: our pick, a runner-up and a free option

A fluent summary that misses one clause can cost more than reading the document yourself. These picks are chosen for summaries you can check, with a workflow that keeps every claim tied to its page.

The short answer

The best AI for summarising PDFs in November 2026 is Google's Gemini Notebook, called NotebookLM until 16 July 2026, because it answers only from your file and cites the exact passage, and its free Standard plan takes sources of up to 500,000 words or 200 MB. Anthropic's Claude is the runner-up when you need judgement on what a document means. ChatGPT's free plan reads only about 12 pages at a time.

Our pickGemini Notebook (formerly NotebookLM): cited answers from your PDF, free
Runner-upClaude: judgement on long documents, Opus 5.5 on paid plans
Best free optionClaude free plan, for a second editorial reader
File sizeGemini Notebook: up to 500,000 words or 200 MB per source
Watch out forChatGPT Free: 27,000-token context, about 12 pages of input

The best AI for summarising PDFs, in short

Summarising a PDF sounds like one job, but it is really two: getting the gist of a long document quickly, and getting a summary you can rely on when you act on it. The second is where most tools let people down, because a fluent summary can quietly skip the one clause that matters. Our picks are chosen for reliability first.

  • Our pick: Gemini Notebook (formerly NotebookLM), because every answer is grounded in your file and cited back to the passage, and the free plan handles very long PDFs.
  • Runner-up: Claude, for summaries that need judgement, such as "what should we push back on in this contract?", and for turning the summary into a finished document.
  • Best free option for a single quick summary: Claude's free plan, since our pick is already free and you may want a second, more editorial reader.

Our pick: Gemini Notebook

Why. Google renamed NotebookLM to Gemini Notebook on 16 July 2026, and the core idea did not change: it answers only from the sources you add, with citations you can click to see the exact passage. That is what you want from a PDF summary. Each source can be up to 500,000 words or 200 MB, the free Standard plan allows 50 sources per notebook, and it accepts PDFs alongside Word, PowerPoint, Google Docs and web links. Beyond a written summary, it can turn the same file into an Audio Overview (a podcast-style summary in more than 80 languages), a slide deck, a report, a quiz or flashcards.

Who it suits. Anyone summarising long or important documents: annual reports, tenders, research papers, policy documents, board packs. It is especially good when you have several PDFs on one topic and need to compare them. Full details are in our Gemini Notebook guide.

The catch. It knows nothing beyond your file, so it will not tell you that a figure in the PDF is out of date. Some features, such as the Longer Audio Overview length and Interactive mode, are English only. On personal accounts Google says your content is not used to directly train its foundational models unless you give feedback, so avoid the thumbs up and down buttons on sensitive notebooks; there is no training toggle.

Runner-up: Claude

Why. Claude, from Anthropic, is at its best with long documents and judgement across a lot of material. Where Gemini Notebook is a careful librarian, Claude is closer to a sharp colleague: ask it what a supplier contract means for you, which assumptions in a business case look weak, or how two versions of a policy differ, and it gives a reasoned answer. Claude can also create files, so a PDF summary can become a briefing document or a slide deck in the same chat. Paid plans from Pro upwards get Claude Opus 5.5, released on 22 September 2026.

Who it suits. Managers, lawyers, analysts and consultants who need an opinion on the document, not only a shorter version of it. Compare it with the others in ChatGPT vs Claude and Claude vs Gemini, or read the Claude guide.

The catch. Usage limits: long PDFs and heavier models use your allowance faster, and every plan has a weekly limit. On Free, Pro, Max and Ultra your chats can be used for training unless you switch off Help improve our AI models in Settings, under Privacy.

Best free option: Claude's free plan

Why. Gemini Notebook is already free, so it would be the obvious answer twice. If you want a second reader for a single document, Claude's free plan includes file uploads, Claude's Sonnet and Haiku models, web search and up to five Projects, which is enough for an editorial summary of a report or a paper.

Who it suits. People who summarise a PDF now and then and want a summary in plain, well-written English.

The catch. Free usage resets every five hours with a weekly cap, so a long PDF and a few follow-up questions can use it up. There is no Opus model on Free.

And ChatGPT?

ChatGPT is a good PDF reader on paid plans, and ChatGPT for Word now works on every plan, including Free. But the free plan has a catch that explains many disappointing summaries: Free uses GPT-6 Luna with an Instant context window of 27,000 tokens, roughly 12 pages of input. Feed it a 60-page report and it cannot hold the whole thing at once, so the ending may simply be missing from the summary. Free also has limited uploads. If you use ChatGPT for long PDFs, use a paid plan, and switch off Improve the model for everyone under Settings, Data controls if the document is confidential.

How to choose an AI for PDF summaries: three questions

  1. How long is the document? For anything beyond a couple of dozen pages, use a tool that is built to hold the whole file, such as Gemini Notebook, rather than a free chat plan with a small context window.
  2. Do you need proof or an opinion? If someone will check your summary against the original, pick the tool that cites passages (Gemini Notebook). If you need a view on what the document means, Claude is the better reader.
  3. How confidential is it? Contracts, HR files and unpublished results belong in a work account that excludes training, such as Gemini Notebook with a work or school account, Claude Team or Enterprise, or ChatGPT Business or Enterprise.

A PDF summary workflow, end to end

Take a 90-page supplier tender you need to summarise for your manager by tomorrow.

  1. Check the PDF is readable. If you cannot select the text in your PDF viewer, it is a scan; run it through text recognition first or expect gaps.
  2. Upload it to a new notebook. In Gemini Notebook, create a notebook and add the PDF as a source. Add the earlier tender or the requirements document too if you want comparisons.
  3. Ask for structure before summary. Request an outline of the sections with page references. This shows you what is there before you decide what matters.
  4. Ask the questions your reader will ask. Price, deadlines, obligations, penalties, exclusions. Click each citation and read the passage for anything with a number or a date.
  5. Get the judgement. Paste the key passages, or upload the PDF, into Claude and ask what is unusual, risky or missing compared with what you would expect.
  6. Write the one-pager. Ask Claude for a summary in your manager's format: the recommendation first, then the five facts that drive it, each with its page number.
  7. Spot-check. Pick three claims at random and find them in the original. If one is wrong, check them all.

Our upcoming guide on how to summarise a PDF with AI covers the menus step by step, and the Gemini Notebook lessons let you practise on your own files.

Why AI summaries of PDFs go wrong

  • The file is too long for the plan. A small context window means the tool reads part of the document and summarises that part confidently.
  • Tables and charts. Numbers in complex tables and figures are where extraction errors happen most. Check every number against the page.
  • "Summarise this" with no purpose. A summary for a lawyer and one for a sales lead should be different. Always say who it is for and what they will decide.
  • Trusting the summary over the source. Quote the document, not the AI, when it matters.

Prompts to copy

Outline before you summarise

Give me an outline of this document: every main section with its page number and one line on what it covers. Then list any appendices, tables or figures that contain numbers.

You get: A map of the document so you know what is in it before you trust a summary

Summary for a specific reader

Summarise this document for [reader, e.g. my finance director] who needs to decide [decision]. Lead with what matters most for that decision, then the five key facts with page numbers, then anything that is unclear or missing.

You get: A decision-focused summary with page references

Pull out obligations and dates

List every obligation, deadline, penalty and condition in this document in a table with columns: what, who is responsible, date or trigger, page number. Quote the exact wording for each.

You get: A checkable table of commitments with quotes

Compare two versions

Compare [document A] with [document B]. List every substantive change: what was added, removed or reworded, why it might matter, and the page in each version.

You get: A clear list of what changed between two PDFs

Ask for the risky parts

Read this [contract, tender, report] as a cautious [role]. What is unusual, one-sided, risky or missing compared with what you would normally expect? Quote the passage and give the page for each point.

You get: A reasoned list of red flags, tied to the text

Check numbers against the source

List every number in your summary above and the exact sentence and page in the PDF it comes from. Mark any number you calculated or inferred rather than copied.

You get: A trail from each figure back to the page

Start the NotebookLM lessons Start the Claude lessons