Claude Projects
Stop starting over. Give Claude a project with memory, instructions, and files.
You spent 90 minutes last Wednesday explaining your dissertation topic to Claude, pasting in your outline, and iterating on chapter 3. Today you start a new chat and discover the model knows none of it. Again.
Why this tool matters
Claude Projects are persistent workspaces inside claude.ai that remember your context — custom instructions, uploaded documents, style references, terminology — across every conversation you have inside them. They turn Claude from a one-shot chatbot into a research or writing environment you can return to for months.
The mental model: a Project is a room. The room holds 20-ish documents you have uploaded (your syllabus, a style guide, three source papers, an outline), and a custom instruction that tells Claude who it is for the work done here (“You are helping me write my PhD dissertation on X; maintain a scholarly voice; when I paste text, assume it is my writing and offer revisions, not praise.”). Every chat you start inside the Project inherits all of that.
For anyone doing a long-running project — a book, a thesis, a consulting engagement, a research program, a family-history investigation — Projects are the feature that makes Claude actually usable as a recurring tool. Without them, you are re-explaining yourself every time.
Setup
Account: Claude Pro ($20/month) or Claude Team. Projects are not available on the free tier as of this writing. The Pro tier includes everything a single learner or researcher needs.
Good first Project: a work project you will return to weekly for at least a month. Projects pay off proportional to their lifespan.
Walkthrough
Step 1: Create a new Project and name it precisely
In claude.ai, click Projects in the left sidebar, then Create a project. Name it as narrowly as you can: Dissertation Chapter 3 — Methodology is better than Dissertation. Narrow scope = better model behavior.
Step 2: Write the custom instructions
This is the single most important step. In about 200 words, tell Claude: who you are, what you are working on, what voice or style you want, and one or two things you do not want it to do. A good pattern: You are a collaborator on my dissertation in applied statistics. My advisor is skeptical of AI-assisted writing, so keep my voice. Use American Psychological Association style. When I paste my writing, suggest edits inline with a brief rationale. Do not rewrite unless I ask.
Step 3: Upload the foundational documents
Drag in 5–15 documents that give Claude the context it will need: a style guide, your outline, three key source papers, a glossary of terms specific to your field. Every chat in the Project will have these in context.
Step 4: Start a chat and test the context
Start a new chat inside the Project. Ask a question that would be impossible without the uploaded context: Given the APA style guide you have and my chapter outline, what section do I need to write next? If the answer is specific and grounded, your setup worked.
Step 5: Add instructions as you learn
Every time you catch yourself correcting Claude on the same thing twice, go back to the Project instructions and add a line. The instructions are a living document. After two weeks of use, yours will be much better than the first version.
Step 6: Branch conversations, keep the Project
Inside a Project, each chat is scoped to a task (one chat per chapter section, say). The Project persists; chats come and go. You end up with a clean archive of the work by sub-task.
Your turn
Basic: Set up your first Project
Create a Project for a real ongoing work item — a book you are drafting, a course you are preparing, a consulting engagement, a coding project. Write the custom instructions. Upload 5–8 documents. Start one chat and do 10 minutes of actual work inside it.
Notice what Claude gets right that it would have needed re-explaining in a regular chat. That difference is the ROI of Projects.
Advanced: A three-Project operating system
Set up three Projects that cover your main work modes:
- Research / reading — uploaded papers, your reading notes, instruction set: help me read and connect ideas.
- Writing — style guide, voice samples, outline, instruction set: help me draft and revise in my voice.
- Correspondence & communication — team bios, recent threads, instruction set: help me draft professional replies in a warm-but-concise voice.
Use each at least once in the coming week. After a week, refine each Project's instructions based on the corrections you found yourself making. This is how you build a personal AI workspace that is meaningfully yours.
Pitfalls and pro tips
Over-stuffed Projects. A Project with 40 documents and 800 words of instructions will behave worse than one with 8 documents and 200 words. More context is not always better; more relevant context is.
Context drift in long chats. Within a single chat, Claude's attention to the Project context can fade after many turns. Start a new chat for each new sub-task rather than running one 60-turn conversation.
Privacy. Anything uploaded to a Project is accessible to Claude during generation. For highly sensitive materials, use a Team plan (which has different data-handling policies) or anonymize before uploading.
How it compares
Claude Projects competes with ChatGPT Projects (similar pattern, slightly different UX), Gemini Gems (Google's customizable assistants), and the Custom GPTs marketplace on ChatGPT. Functionally they are close. Claude's differentiator is the quality of the model on long-context writing and reading tasks — if you are doing sustained writing or analysis, Claude Projects tends to out-perform the equivalents. For quick-start productivity (“I want a customer-email assistant in 5 minutes”), Custom GPTs are lower-friction.
When to use — and when not to
Use Claude Projects when you have a project you will return to for weeks or months — writing a book, a thesis, running a research program, managing an ongoing client. Projects pay off with repetition.
Do not use Claude Projects when you have a one-off question (regular Claude is faster), when you need web browsing (Perplexity is better), or when you are still figuring out what the project is — Projects reward commitment.