# How to turn a PDF into a course

> Most PDFs get opened once and then rot in a downloads folder. Here is how to make one into something you work through.

https://learnfold.app/guides/turn-a-pdf-into-a-course/ · Updated 2026-10-30

To turn a PDF into a course, give an AI tutor a clean, focused file, check the outline before anything gets written, then work through it in short sections with questions and self-checks. The short version comes first.

1. Check that the PDF has real text rather than pictures of pages.
2. Cut it down to the part you actually need.
3. Know what you want from it before you start.
4. Read the plan and make sure it fits before you approve it.
5. Read one section at a time and ask about the sentences that lose you.
6. Test yourself without looking, then check numbers and quotes against the original.

The rest of this guide explains each step and where it tends to break.

## Step 1. Find out what kind of PDF you have

There are two kinds of PDF and they behave very differently. A PDF exported from a word processor or a website contains real text. A scanned PDF is, in the words of the University of Washington's records office, "little more than a high-definition photograph" of each page. You cannot search it or select its words until someone runs optical character recognition (OCR) on it.

The ten-second test is to try selecting a sentence or searching for a word you can see on the page. If nothing highlights, it is a scan.

This matters because of how AI models read documents. Anthropic's documentation says that when you send a PDF to Claude, each page is converted into an image and the page's text is extracted and provided alongside it. Google's Gemini docs describe the same idea, with embedded text extracted and pages processed visually. For a born-digital PDF, the model gets clean text plus a picture. For a scan, it mostly has the picture, and Anthropic warns that its model "might hallucinate or make mistakes when interpreting low-quality, rotated, or very small images." Google's advice for its own model is to rotate pages to the correct orientation and avoid blurry pages.

If your PDF is a scan, run OCR first if you can (many PDF editors have a "recognize text" option), rotate any sideways pages, and expect to double-check names and numbers.

## Step 2. Cut it to the part you need

Long documents fail quietly. Anthropic notes that dense PDFs, with small fonts, complex tables or heavy graphics, "can fill the context window before reaching the page limit," and recommends splitting large files into sections. Page caps exist too and differ by tool. Treat any figure you read today as temporary.

There is a second reason to cut. A course built from a 300-page report will be broad and thin. A course built from the 25 pages on the method you need to understand can go deep. If the PDF is a textbook or a long report, save the chapter you are working on as its own file.

## Step 3. Know what you want from it

Before you start, write one line for yourself. Who are you, what part of the file matters, and what should you be able to do at the end? For example, "I work in marketing, I never took statistics, and I want to read a research summary without nodding along." That line is your yardstick for the plan in step 4 and for every question you ask later.

In a general chatbot you can paste that line in with the file. In Learnfold, you attach the PDF on the New Course screen and the course builds on what you brought. You steer it after that. Review the plan, then use "Where next?" at the end of each lesson, questions about the page, and your own questions to put the weight where your goal needs it. If the PDF route is not working, you can also start from a link, notes, Markdown or a photo.

## Step 4. Read the plan before you approve it

Learnfold shows a course plan and waits for your approval before it writes any lessons. Use that moment. Hold the plan up against the PDF's table of contents and ask three questions of it.

Play the approved example, or move the slider through each step. In your own course, stop at the plan and check it before you approve.

Does it cover the sections you care about? If chapter 3 has a section on sample size and the plan skips it, you have two choices. Approve it and chase sample size later with your own question through "Where next?", or, if the gap is serious, don't approve it and start again with a sharper question or a file cut down to that chapter.

Is anything in the plan that is not in your document? An AI may fill gaps from general knowledge. Sometimes that is welcome background. Sometimes it means the plan drifted.

Is the order sensible for you? If the plan is not right, don't approve it. Start again with a sharper question or a tighter file.

## Step 5. Read in sections and ask about the exact passage

UNC Chapel Hill's Learning Center suggests tackling "no more than 5-10 pages at a time" and pausing after each section to summarize it from memory in your own words. That advice works just as well for lessons generated from a PDF. Learnfold writes lessons section by section, so you can start reading the first part while the rest is being written.

When a sentence loses you, ask about that sentence rather than the whole topic. In Learnfold you can select a passage and ask "Why?", "How?", or for an example. Typing your own question works on the whole page and through "Where next?". A few questions that work well on PDF-based lessons.

> Where in the PDF does this come from? Quote the sentence.

> The PDF uses "margin of error" and this page says "confidence interval." Are they the same thing here?

> Give me an example with real-looking numbers, then show me which part of the PDF it matches.

The first question is a good habit for any long document. Anthropic's own guidance for reducing made-up answers on long documents is to have the model pull word-for-word quotes first and work from those. You can do the same thing as a reader, then find the quote in the original yourself.

## Step 6. Check what stuck

Reading a clear explanation feels like understanding. Testing tells you whether it is. In his summary of the learning research for teachers, John Dunlosky writes that students "will benefit most from tests that require recall from memory," which is a good reason to choose a written answer over picking from four options.

Learnfold has optional checks at the end of a page, with multiple choice, a written answer or a guided discussion, and feedback on what you got right. A wrong answer can turn into a conversation about why. Practice pages help you revisit lessons after missed answers. If you want more on building a study routine around this, see [how to study a textbook chapter with an AI tutor](/guides/study-a-textbook-chapter-with-ai/).

## When this goes wrong

**The numbers look slightly off.** Tables and scanned pages are where small errors hide. Find every number you plan to reuse in the original PDF. If you cannot find it, do not use it.

**The course talks about things the PDF never says.** Ask about the page which parts came from the document. Then decide whether you want the extra context or a course that sticks to the text.

**The course is broad and shallow.** The file was probably too big. Split it and start again with one chapter.

**The PDF will not open or reads as nothing.** Password-protected and encrypted files can be refused. Anthropic's docs list "no passwords/encryption" as a requirement for its API, for example. Save a copy without the password, or paste the text in as notes.

**An explanation is confidently wrong.** It happens. Learnfold's own note is plain about it. It uses AI to write explanations, AI makes mistakes, and you should treat it as a tutor, not a textbook, and check important facts against a reliable source. For a PDF course, the PDF itself is often that source. If you find a bad sentence, you can edit the page yourself or ask the AI to rewrite that section. For a full routine, read [how to check an AI explanation for mistakes](/guides/check-ai-explanations-for-mistakes/).

## A few details we cannot promise

Learnfold's published material says you can attach a PDF and the course builds on it. It does not publish page limits or say how it treats scanned pages, so the safe defaults above apply. A clean, text-based file of a sensible length will give any tool its best shot. If your PDF is really a set of notes, [turning study notes into a course](/use-cases/study-notes-to-course/) covers that version, and [better questions for an AI tutor](/guides/write-better-questions-for-an-ai-tutor/) will help in every lesson.

## Quick answers

**Can AI read a scanned PDF?**

Often, but less reliably. A scanned page is a picture of text, so the AI has to read the image. Blurry, rotated or low-contrast scans cause more mistakes. Running OCR first, or checking names and numbers against the page, helps.

**How long a PDF can I turn into a course?**

It depends on the tool and model, and the limits change. Even when a long file is accepted, a course on 40 pages you care about is usually better than a course on 400 pages you skimmed. Split long documents by chapter.

**Can AI read tables and charts in a PDF?**

Some models read page images as well as text, so they can describe tables and charts. Treat any number pulled from a table as unverified until you find it in the original.

**Will the course only use what is in my PDF?**

Do not assume so. AI models also draw on what they learned in training. If it matters, ask which parts of an explanation come from the document and ask for the exact sentence it relied on.

## Sources

1. [PDF support](https://platform.claude.com/docs/en/build-with-claude/pdf-support), Anthropic, Claude Platform docs
2. [Vision (limitations section)](https://platform.claude.com/docs/en/build-with-claude/vision), Anthropic, Claude Platform docs
3. [Reduce hallucinations](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations), Anthropic, Claude Platform docs
4. [Document understanding](https://ai.google.dev/gemini-api/docs/document-processing), Google AI for Developers, Gemini API docs
5. [Optical character recognition](https://finance.uw.edu/recmgt/resources/optical-character-recognition), University of Washington Records Management Services
6. [Reading textbooks effectively](https://learningcenter.unc.edu/tips-and-tools/reading-textbooks-effectively/), UNC Chapel Hill Learning Center
7. [Strengthening the student toolbox: study strategies to boost learning](https://www.aft.org/ae/fall2013/dunlosky), John Dunlosky, American Educator (AFT), Fall 2013