Quick answer: it depends on the tool's context window, which varies by model, app and subscription plan. As a rule of thumb, a 128,000-token window holds about 170 single-spaced pages of English text, a 200,000-token window about 270 pages, and a 1 million-token window about 1,350 pages, after leaving 10% free for the reply.
Why there's no single answer
ChatGPT, Claude and Gemini each come in several models, and each company offers different limits in its free app, paid plans and developer API. Limits also change regularly as new models launch. A number quoted in an article can be out of date within months, so check your own tool's current limit in its help pages or plan details.
Pages per context window size
Assumes 1.33 tokens per word of English, 500 words per single-spaced page, and 10% of the window kept free for the reply.
| Context window |
About this many words |
Single-spaced pages |
Double-spaced pages |
| 8,000 tokens |
6,000 |
11 |
22 |
| 32,000 tokens |
24,000 |
43 |
87 |
| 128,000 tokens |
96,000 |
173 |
346 |
| 200,000 tokens |
150,000 |
271 |
541 |
| 400,000 tokens |
300,000 |
541 |
1,083 |
| 1,000,000 tokens |
750,000 |
1,353 |
2,707 |
| 2,000,000 tokens |
1,500,000 |
2,707 |
5,414 |
To check a specific document, paste it into the context window calculator. It estimates the tokens and shows how much of each window size the document uses.
Things that reduce the real number
The conversation itself. Earlier messages, your instructions and any other files all share the window.
PDF formatting. Tables, headers, footnotes and repeated page furniture add tokens. Scanned PDFs may need text recognition first, and some tools process images of pages differently.
Non-English text and code. Both typically use more tokens per word than English prose.
App limits on uploads. Some apps cap file size or the number of files separately from the context window.
Fitting is not the same as reading well
Even when a long document fits, models can miss details buried in the middle of very long inputs. For important work:
- Paste only the relevant sections where you can.
- Ask the AI to quote the passage it's relying on, so you can check it.
- For very long documents, summarize in sections and then combine the summaries.
For what happens when you go over the limit, see what happens when you exceed the context window.
Common documents and what they need
Rough token estimates for typical English documents, using 1.33 tokens per word:
| Document |
Typical length |
About this many tokens |
| Two-page cover letter |
600 words |
800 |
| 20-page report |
10,000 words |
13,300 |
| Research paper |
8,000 words |
10,600 |
| Employee handbook |
30,000 words |
40,000 |
| Novel |
90,000 words |
120,000 |
| Year of meeting transcripts |
300,000 words |
400,000 |
A typical report or paper fits comfortably in almost any current AI tool. Books and large collections are where context limits start to matter.
- Check the help centre or plan comparison page for the app you use. Limits are usually listed per plan.
- For developer access, check the provider's model documentation, which lists the context window per model.
- Note whether the limit includes the reply. In most tools, input and output share the same window.
Because these numbers change as new models launch, check them when you start a big project rather than relying on an old article.
Getting good answers from long documents
Ask specific questions. "What does section 4 say about refunds?" gets better answers than "Summarize this."
Ask for quotes. Request the exact sentence the answer is based on, so you can verify it.
Put instructions first. State what you want before pasting the document, and repeat the key instruction after it for very long inputs.
Split by section. For documents near the limit, process chapters or sections separately and combine the results.
Remove clutter. Page numbers, headers, footers and tables of contents add tokens without adding meaning.
Scanned PDFs and images
A scanned PDF is a picture of text. Some tools read it with built-in text recognition, others can't, and image-based pages can use far more of the context window than the same text pasted directly. If a scanned document doesn't work well, convert it to text first and paste that.
Quick checklist for long documents
- Estimate the tokens with the context window calculator.
- Check your plan's limit in the tool's help pages.
- Leave room for the reply and follow-up questions.
- Remove clutter such as headers, footers and page numbers.
- Split anything that uses more than about 70% of the window.
- Ask for quotes so you can check the answers.
Questions people ask
How many pages can ChatGPT read at once?
It depends on the model and plan. A 128,000-token context window holds roughly 170 single-spaced pages of English, leaving room for a reply. Check your plan’s current limit.
How many pages is 200,000 tokens?
About 150,000 words, or roughly 270 single-spaced pages after leaving 10% for the reply.
Why can’t the AI read my whole PDF?
The PDF may exceed the context window, contain scanned images, or hit a separate file-size limit in the app.
Can ChatGPT read a whole book?
It depends on your plan’s context window. A 90,000-word novel is about 120,000 tokens, which fits in larger windows but not smaller ones.
Why does a scanned PDF use more of the context window?
Scanned pages are images, which some tools process differently from text and which can take more space. Converting to text first usually helps.
Do images and charts in a PDF count toward the context window?
Often yes, if the tool processes them. Images can take more space than the same information written as text.
Is pages a good way to measure what fits?
It’s a useful rough guide, but tokens are what count. Dense text, tables and code use more tokens per page than ordinary prose.
How many words fit in a 128K context window?
About 96,000 English words in total, or roughly 86,000 if you keep 10% free for the reply.
Sources and further reading
Figures in this guide were calculated with our own calculators and checked against these sources.