How to use Claude Projects and project knowledge
A chat forgets. A project remembers. Projects are the feature that turns Claude from a clever conversation into a research assistant that has already read your material. Here is how they work, what to put in them, and the habits that make them pay off.
What a Claude Project is
A project on claude.ai is a workspace with three parts: a set of instructions Claude follows in every chat inside it, a knowledge base of documents Claude can draw on, and the chats themselves. Open a new chat inside the project and Claude starts already knowing the brief and the material. You stop re-explaining who you are and re-uploading the same PDF.
Projects are available on free and paid plans. Paid plans get more usage and larger knowledge capacity, but the mechanics are the same.
Project knowledge versus uploading to a chat
You can attach a file to any chat. The difference is where it lives afterwards.
- Chat upload: the file exists in that one conversation. Ask about it next week and you upload it again. Two files in two chats can never be compared.
- Project knowledge: the file is available to every conversation in the project, now and later. Ask a question on Monday that spans a report you added in March and a transcript you added yesterday, and Claude answers from both, with citations.
The rule of thumb: if you will ask about something more than once, it belongs in project knowledge.
Set up a project in four steps
- Create it. In the claude.ai sidebar, open Projects and click New project. Name it after the topic ("Pricing research", "Podcast notes: AI infra"), not the task, so it stays useful after today.
- Write the instructions. A few sentences is enough: who you are, what the project is for, what good output looks like. For a research project: "I am a founder studying how developer tools price. Answer with specifics and cite the source document and timestamp. Flag claims that are unsupported."
- Add knowledge. Upload the documents Claude should know: PDFs, notes, spreadsheets exported as CSV, code, and transcripts. Name files descriptively; Claude cites filenames.
- Ask across the library. Start a chat inside the project and ask the question you actually have. Claude reads or retrieves from everything you added.
What to put in project knowledge
Anything you would otherwise re-explain. Some libraries that work well:
- Research on a topic: papers, reports, competitor pages saved as PDF, and the talks and podcasts where the interesting people actually said what they think.
- A course or a book club: the syllabus, your notes, and the lecture transcripts. Ask for a study guide the night before the exam.
- A product or a client: the spec, the brand voice document, past deliverables, and every customer interview.
- A "watch later" list: videos you meant to watch and never will. Claude has read them; ask which ones deserve your time.
Adding YouTube videos to a project
Video is where most project libraries stall, because Claude cannot watch YouTube videos. What it can read is the transcript, and that is what you add. The manual route is to copy the transcript from YouTube's transcript panel into a text file and upload it. The one-click route is the YouTube Transcript to Claude extension: it adds an Add to Claude button under every video, and a click files the transcript into the project you pick as a markdown document with the title, channel, link, chapter headings, and a timestamp every 30 seconds.
The extension remembers your default project, so a research library grows one click at a time. Step by step: add a YouTube video to a Claude project.
How much a project can hold
More than you think. A three hour podcast transcript is about 35,000 words. Claude's context window fits several of those at once. When a project grows past what fits in context, Claude switches to retrieval: for each question it pulls the relevant passages rather than reading every document end to end. Two consequences follow. Specific questions ("what did the second speaker say about churn?") work better than sweeping ones ("summarize everything"). And one project per topic beats one giant project for everything, because retrieval has less to sift through.
Habits that make projects pay off
- Name files for citation. "Q2 pricing interview, Maria.pdf" beats "doc (3).pdf". Transcripts added by the extension are named after the video title automatically.
- Keep timestamps and page numbers. A citation you can check is worth ten you cannot. Ask Claude to include them and it will.
- Put standing questions in the instructions. If you always want a "what would a skeptic say" section, say so once, in the instructions, not in every chat.
- Prune. A project is a library, not a landfill. Remove documents that no longer matter and the answers get sharper.
- Replace, do not duplicate. When a document gets updated, delete the old copy. The extension does this for you when you re-add a video.
Questions people ask
Are Claude Projects free?
Yes, projects are available on the free plan. Paid plans raise the usage and knowledge limits.
Can I share a project?
On team plans, projects can be shared with your organization so everyone works from the same knowledge. Personal plans keep projects private.
Can I add a whole YouTube playlist?
One video at a time, with a click each. Because the extension remembers your project, a twenty video playlist takes about a minute.
Does project knowledge count against my usage?
Reading knowledge uses context, so very large projects can consume usage faster on a free plan. Keeping projects focused helps.
Fill a project with videos in one click each
YouTube Transcript to Claude is a free, open source Chrome extension. No API key, no signup, it uses your existing claude.ai session.
Add to Chrome, it's free