Stop Repeating Yourself: 10 Steps to a Claude That Knows You

Every morning, the same ritual. You open Claude, type your question, and spend the first three messages explaining who you are, what you do, and what you need. You are an ObGyn. You want evidence-based answers. You do not want a disclaimer telling you to consult your physician. You are the physician.

Then you close the tab. Tomorrow, you do it again. Claude has no memory of you. It is like seeing a new locum every shift who has never read a single chart.

It does not have to work this way. Claude has features that remember you, your files, and your preferences across every conversation. Here are 10 steps to set them up. Total time: about 30 minutes. You will never re-explain yourself again.

The 10 Steps

Step 1: Open a Project. In Claude, click “Projects” in the left sidebar and create a new one. Give it a name that describes its purpose. Mine is called “CDC Natality Analysis.” Another is “ObGyn Intelligence Writing.” A project is a workspace. Everything inside it—files, instructions, conversations—stays together and carries forward. Think of it as a dedicated chart for a specific type of work.

Step 2: Write your Project instructions. Every project has a text box at the top for instructions. This is where you tell Claude how to behave inside this project. Be specific. Mine says things like: “Always use Vancouver citation format. Write at 7th-grade readability. Never fabricate references. When I say Summ, produce a 400-word summary in three sections.” These instructions apply to every conversation you start inside this project. Write them once. They persist.

Step 3: Upload your key files. Drag your most-used documents into the project. Practice protocols. Your favorite review articles. Patient education handouts. Clinical guidelines you reference often. Claude reads these files before answering your questions. If your gestational diabetes protocol is in the project, you can ask “What does my protocol say about insulin initiation?” and Claude will answer from your document, not from a generic training set.

Step 4: Create your about-me file. Open any text editor. Create a file called about-me.md. Write a short description of yourself: your specialty, your clinical focus, your research interests, your communication preferences. Mine says I am an MFM specialist and ethics consultant, that I use CDC natality data, that I write for a general audience at obmd.com, and that I do not tolerate hedging language. Upload this file to your project. Claude now knows who it is working with.

Step 5: Be specific about what you do not want. This is the part most people skip, and it matters more than what you do want. Tell Claude what to avoid. “Do not add disclaimers about consulting a healthcare provider. Do not use the phrase ‘more research is needed.’ Do not soften conclusions when the evidence is clear.” Negative instructions are more powerful than positive ones. They prevent the generic AI voice from creeping back in.

Step 6: Start a conversation inside the project. Open your project and start a new chat. Ask a question you have asked before in basic chat. Notice the difference. Claude already knows your specialty, your preferences, and your files. It skips the preamble. It writes the way you told it to write. The answer is not generic—it is shaped by your context. This is the moment most physicians realize what they have been missing.

Step 7: Test it with a real task. Upload a study you recently read. Ask Claude to summarize it for a clinical audience. Or ask it to draft a patient counseling script based on your uploaded protocol. Or ask it to compare two approaches using the guidelines you provided. The more you test it with real work, the faster you learn what instructions to refine.

Step 8: Refine your instructions after the first week. After five or six conversations, you will notice patterns. Claude does something you did not want. Or it misses a preference. Go back to your project instructions and edit them. This is not a one-time setup—it is a living document. I update my instructions roughly once a week. The AI gets sharper each time because you get more specific each time.

Step 9: Create a second project for a different purpose. Once you see the value, create another project for a different part of your work. One for research. One for patient education. One for administrative tasks. Each project has its own instructions, its own files, its own memory. They do not bleed into each other. Your research project can be precise and technical. Your patient education project can be warm and simplified. Same AI, different context.

Step 10: Set Global Instructions for things that apply everywhere. Claude also has a setting called Global Instructions—sometimes called “Custom Instructions” in Settings. These apply across all conversations and all projects. Use this for universal preferences: your name, your credentials, your department affiliation, your default citation format. Project instructions handle the specific. Global instructions handle the constant. Together, they mean Claude never starts from zero again.

My Take

I spent 50 years building clinical knowledge. The idea that I should re-explain my expertise to an AI every single session is absurd. These features exist to solve that problem. They are not advanced. They are not technical. They are a text box and a file upload. If you can write admission orders, you can configure a Claude project.

The physicians who tell me AI is not useful are almost always the ones using basic chat with no context, no files, and no instructions. That is not AI failing. That is a misconfigured tool. Fix the configuration. The results follow.

Part 2 of the series “You Are Using 5% of Your AI.” Next: The Features That Work Without You.

Subscribe to ObGyn Intelligence at obmd.com