How to Use Claude Projects to Study, Class by Class

To use Claude Projects to study, make one project per class, upload that course’s syllabus, lecture notes and readings into the project knowledge base, then write project instructions once telling Claude how to answer from that material. Every chat you start inside the project already has all of it loaded, so you stop re-pasting context.

This guide assumes you already know what Claude is and how to prompt it. If that part is new, the companion guide on how to use Claude to study covers the general technique. What follows is about one specific feature, and about running it across a whole term rather than a single session.

What is a Claude Project, and who can use one?

Anthropic’s own help page puts it in a sentence. Its article What are projects? says that “projects allow you to create self-contained workspaces with their own chat histories and knowledge bases”, and that “within each project, you can upload documents, provide context, and have focused chats with Claude”.

On availability the same page is unambiguous: “Projects are available to all users, including those with free Claude accounts. Free users can create a maximum of five projects.” So a student on a free account can run five courses in five workspaces today. Team and Enterprise plans add sharing and permissions on top, and every paid plan adds one capacity feature covered further down.

A project has two moving parts, and they are worth keeping straight.

• Project knowledge is the material you upload. Anthropic describes it as the ability “to provide context for your chats with Claude”, where you “upload relevant documents, text, code, or other files to a project’s knowledge base, which Claude will use to better understand the context and background for your individual chats within that project”.
• Project instructions are standing directions about behaviour. The same page treats them as a separate thing: “In addition, you can define project instructions for each project to further tailor Claude’s responses. For example, instructing Claude to use a more formal tone or answer questions from the perspective of a specific role or industry.”

One caveat on timing. That help page currently carries a note that a new version of projects is rolling out in stages, in beta to selected Pro and Max subscribers who use Claude Code, and that existing projects in chat keep working as they do today. Everything below describes the current version, which is the one a student on the web will meet.

How do you set up a Claude project for a class?

Four steps, and Anthropic documents them in How can I create and manage projects? The one detail people miss is in step one: when you give a project a name and a description, the help page notes that “Claude will not have access to these details”. Naming a project Organic Chemistry 210 tells the model nothing. The course name belongs in the instructions or in a file.

You can also fold in work you have already done. Anthropic documents moving a standalone chat into a project through the dropdown next to the chat name, so a conversation you started before setting the project up does not have to be abandoned.

What can go into project knowledge, and what cannot

Anthropic publishes the file limits, and the ones for project files differ from the ones for a single chat. Its article Upload files to Claude lists project files at “30MB per file”, with the number of files “unlimited, but total content must fit within Claude’s context window”, and content handled as “text extraction only (except for multimodal PDFs)”. It adds that “additional token limits may apply based on the length of extracted content”. There is no published file-count number to plan around, so the working constraint is total extracted text.

The accepted document types listed there are PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX, with XLSX requiring code execution and file creation to be enabled on your account. PDFs get their own rule: Claude reads both text and visual elements in PDFs of 100 pages or fewer, processes text only from 101 to 1000 pages, and refuses anything longer.

Two consequences for a student. A 400-page textbook PDF will load, and its diagrams will be invisible to Claude, so a figure that carries the argument needs cropping into its own shorter file. And for non-PDF documents Anthropic says Claude extracts text only, and “if they contain embedded images, Claude won’t be able to read or interpret them”, so a DOCX of photographed whiteboards adds nothing to the knowledge base.

Capacity is where the plans diverge. Anthropic’s article on retrieval augmented generation for projects opens with “RAG for projects is available on paid plans (Pro, Max, Team, and Enterprise)”, and explains that “when your project knowledge approaches the context window limit, Claude will automatically enable RAG mode to expand your project’s capacity by up to 10x while maintaining quality responses”. It activates on its own with no setup, and Claude then searches the uploads instead of holding all of them in context at once. A free project has no such expansion, which is the real difference between a free and a paid project once a course pile grows.

Before you upload a term of coursework anywhere, it is worth knowing what happens to it. The guide on whether it is safe to upload your notes to an AI chat tool walks through the questions to ask about retention and training, and they apply to a project knowledge base as much as to a one-off chat.

Write the project instructions once

Project instructions are where a study project stops being a folder. Anthropic offers tone and professional role as examples, and for a course the useful instructions are about sourcing and vocabulary. A block along these lines covers most of it:

Answer from the files in this project first. When something is missing from them, say so plainly before you draw on general knowledge. Name the lecture or reading behind each claim. Use the terminology from the course textbook, even where a more common word exists. When I ask you to check my understanding, quiz me one question at a time and wait for my answer before telling me how I did.

Each line does a specific job. Sourcing first keeps answers anchored to the material you are actually examined on. The attribution line turns every claim into something you can go and verify in ten seconds, which matters because an instruction is a steer rather than a guarantee: models still state wrong things with complete confidence, as the guide to AI hallucinations when studying sets out. The terminology line stops Claude quietly translating your course into its own vocabulary, which is how a fluent explanation ends up using words your marker has never seen. And the one-question-at-a-time line makes the default interaction a test.

Why a fact you established in one chat vanishes in the next

Here is the detail that catches people out, and Anthropic states it plainly in its create-and-manage article: “Context is not shared across chats within a project unless the information is added into the project knowledge base.”

Read that against the two things a project does carry. Of the knowledge base, the same page says “anything you upload to this space will be used across all of your chats within that project”. Of the instructions, it says “Claude will use these instructions for all the chats within the project”. Files and instructions travel. Conversations stay where they happened.

Across a semester that adds up. The clean explanation you and Claude arrived at in week three, the exam date you mentioned once, the correction you made when it misread a formula: all of it sits inside a conversation that week seven never sees. The workaround costs about two minutes a week. Keep one plain text file called something like course context, put the exam dates, the marking criteria, the definitions your lecturer insists on and your known weak topics in it, and re-upload it to project knowledge whenever it grows. Anything you want a future chat to start from has to be in a file.

Claude’s memory feature is a separate mechanism from project knowledge, and Anthropic notes that “memory is on by default for Free, Pro, and Max plans on the web, Claude Desktop, and Claude Mobile”. The instruction above is still the reliable one, because a knowledge base is a thing you can open and check.

What a project is worth across a semester

The payoff compounds. In week one a project is a folder with a syllabus in it and Claude can tell you roughly what a chat could tell you anyway. By week eight it holds forty pages of your own lecture notes, the reading list, three problem sets and the marking rubric, and the questions you can ask change shape.

• Cross-referencing. Which of my readings covers this concept, and where do two of them disagree.
• Notation continuity. Explain this week’s derivation using the notation from lecture four, since the textbook uses a different one.
• Gap finding. Compare my week two notes against the syllabus for that week and tell me what I never wrote down.
• Rubric checking. Here is my draft answer. Against the marking criteria in the project, what is missing.

None of those are possible in a fresh chat without pasting the material in first, which is the friction that stops people asking. If the grounded-workspace idea appeals and you also use Google tools, the guide to Gemini Notebook for students covers the closest equivalent, which takes a stricter line on answering only from your sources.

One honest caution. A project is only as current as the files in it, and it will answer confidently from an outdated picture of the course if you stop uploading. Put the upload step on the end of your existing after-class routine so it happens without a decision.

Where a project stops and testing yourself starts

A project is built for asking and discussing. You pose a question, Claude answers from your material, and you read the answer. That is genuinely useful for understanding, and it is a different cognitive act from recalling something with the source closed.

The evidence on which of those moves information into memory is not close. The review by Dunlosky and colleagues (2013) assessed ten common study techniques and rated exactly two as high utility: practice testing, meaning retrieving an answer from memory, and distributed practice, meaning spreading that testing across days. Reading a fluent explanation of your own notes feels like progress while it is happening, which is precisely the effect described in the guide to the illusion of competence.

GeniusPal exists for the other half. Upload the same lecture notes or chapter you put in your project, and it turns them into a set of questions you answer with the source shut: quiz, flashcards, or active recall, all drawn from your own material. It holds no conversation, and it browses nothing. Claude Projects and GeniusPal are answering two different questions about the same file, which is why using both beats choosing between them.

What that costs. Two study sets are free for the lifetime of a GeniusPal account, which is enough to try it on one week of a course. Those free sets hold ten questions each, every one multiple choice, and can be sat twice. Student, at $14.99 a month, raises the ceiling to 100 sets a month, 30 questions in a set, and the flashcard and written-recall modes on the same material. Genius costs $59.99 a year. The daily review, assembled from questions you have already missed, costs nothing on any plan.

Set the project up in the first week of term, when the syllabus is the only file you have and the whole job takes ten minutes. Everything after that is one upload a week, and the alternative is re-explaining your course to a blank chat box every time you have a question.