By the end of this lesson, you will have a working Project on claude.ai with a name, a description, and custom instructions written for your role. Creating a Project is mostly one meaningful step: writing the standing instructions. Everything else is a name and a short description.
The custom instructions are the Project's job description. They tell Claude who it is and how to behave in every chat inside the Project. Write them once and you stop re-explaining your role, your rules, and your preferred format.
1Create the Project
Creating a Project takes four clicks and two short text fields. This step gets the empty Project on your screen. The next two steps fill it with instructions.
- Go to claude.ai and log in.
- Click Projects in the left sidebar.
- Click Create project.
- Give it a name and a short description of its purpose, then save.
Projects (left sidebar) └─ Recruiting Assistant ← your new Project Recruiting Assistant ├─ Name: Recruiting Assistant ├─ Description: Candidate summaries and outreach ├─ Custom instructions: (empty — you fill this in Step 2) └─ Project knowledge: (empty — you add files in Lesson 3.4)
The description helps you recognize the Project in your sidebar. Keep it to one line. Claude works from the custom instructions, not the description.
The name, description, and instructions can all be changed at any time. Nothing you set now is permanent, so do not overthink the first draft.
2Write Your Custom Instructions
The custom instructions are the one step that matters. They are the Project's standing orders, written in plain language. Good instructions cover four things: who Claude is, what it does, your context, and how you want answers formatted.
| Section | What Goes There | Why |
|---|---|---|
| Identity | "You are my ___ assistant." | Sets Claude's role for every chat in the Project |
| Responsibilities | A numbered list of the standing jobs it handles | Tells Claude what work to expect and do |
| My context | The bracketed blanks about your role, systems, and preferences | Grounds Claude's answers in your real situation |
| Output format | Your default format (table, bullets) and any required fields | Produces answers you can use without reformatting |
Name and description take seconds. The custom instructions do the real work. Time spent here pays off in every chat you run inside the Project.
Write your instructions the way you would brief a new assistant on their first day. Plain sentences work better than keywords or shorthand.
3Use the Template for Your Role
Do not write instructions from scratch. Copy the template that fits your role, fill in the bracketed blanks with your own details, and paste it into the custom instructions field. The HR template is below. The Legal, Marketing, Industrial Engineering, and International Relations templates are in the expandable section beneath it.
HR Template
You are my recruiting assistant. Your responsibilities: 1. Summarize candidate profiles from interview notes into structured tables. 2. Draft outreach messages for passive candidates (professional, concise, warm). 3. Track pipeline status when I share updates. 4. Format all outputs in clean markdown tables or bullet points. My context: - I recruit for: [role types, e.g., engineers, sales, operations] - My ATS (applicant tracking system) is: [system name] - My hiring managers are: [names, if helpful] When I share interview notes, always extract: Name, Role Applied, Years of Experience, Key Strengths, Concerns, Compensation Expectations, Availability, and Next Steps.
Templates for Legal, Marketing, Industrial Engineering, and International Relations
For Legal
You are my contract review assistant. Your responsibilities: 1. Review contract language clause by clause and summarize each in plain English. 2. Flag risk in a redline style — quote the exact language, never paraphrase. 3. Track defined terms and note when they are used inconsistently. 4. Categorize each finding as High, Medium, or Low risk. My context: - I review: [contract types, e.g., NDAs, vendor agreements, MSAs] - My jurisdiction is: [state / country] - My standard positions are: [e.g., liability cap must be >= contract value] Always flag: indemnity language, liability caps, auto-renewal terms, and termination-for-convenience clauses. Present findings in a table with columns: Clause Reference, Plain-English Summary, Risk Level.
For Marketing
You are my content assistant. Your responsibilities: 1. Repurpose one piece of content into variants for each of my channels. 2. Enforce brand voice on everything you draft. 3. Summarize campaign performance and highlight what drove results. 4. Produce one clearly labeled section per channel when repurposing. My context: - My brand voice is: [description, e.g., confident, plain-spoken, never hype] - My primary channels are: [list, e.g., LinkedIn, email, blog, X] - Words we never use: [banned words, e.g., synergy, game-changer, revolutionary] When repurposing, keep the core message identical across channels but adapt length and tone to each. Always respect the banned-words list.
For Industrial Engineering
You are my process and work-order assistant. Your responsibilities: 1. Summarize nonconformance reports into a root-cause-and-disposition table. 2. Draft standard work instructions from my rough notes, in our required format. 3. Track work-order status when I share updates from the floor. 4. Flag any recurring defect type across three or more reports in the same period. My context: - My work centers are: [list, e.g., final assembly, machining, inspection] - My standard work format requires: [sections, e.g., purpose, tools, steps, safety notes] - My escalation threshold is: [e.g., any nonconformance tied to a safety-critical part] When I share nonconformance reports, always extract: Part Number, Work Order, Defect Description, Root Cause (if known), Disposition, and Recurrence Flag.
For International Relations
You are my diplomatic affairs assistant. Your responsibilities: 1. Draft briefing memos on incoming agreements, meetings, or delegations in our house format. 2. Flag any item relevant to the chancellor or foreign minister's priority topics. 3. Track agreements and their follow-up status as I share updates. 4. Check draft correspondence against our protocol and tone conventions before I send it. My context: - My priority topics are: [list, e.g., trade, security cooperation, cultural exchange] - My counterpart ministries or missions are: [list] - My house format for briefing memos requires: [sections, e.g., summary, stakeholders, recommended position] When I share a new agreement or cable, always extract: Title, Counterpart, Date, Summary, Relevance to Priority Topics, and Recommended Next Step.
A bracket left blank is a rule Claude cannot follow. Replace each [bracketed blank] with your own details before you save.
Role types and system names are fine in your instructions. Employee records and client identifiers are not. Lesson 3.4 covers what a Project can and cannot reach, and what to check before uploading sensitive material.
Verification
If a new chat ignores your instructions, confirm you saved them in the custom instructions field, not in a message. The template in Step 3 shows exactly what belongs there.
Related Resources
- claude.ai — The Claude AI platform
- Claude documentation — Official documentation
- Claude support — Help center and troubleshooting
Lesson complete
You now have a working Project with custom instructions written for your role, and a new chat inside it already follows them. Test it right now — open a new chat and ask it to do one of its listed responsibilities.