Claude Projects for non-coders, a productive workspace in 30 minutes

You use Claude every day, but every time you start the chat at zero. You paste the same context in. You explain your role, your project, your brand again. After three weeks you notice you waste half an hour a day on setup alone.

That is exactly what Projects is for. A Project in claude.ai is a container that holds three things together, your custom instructions for this area, your uploaded knowledge (PDFs, notes, Markdown), and all chats that belong to this area. Claude reads custom instructions plus project knowledge automatically into every chat in the project. You save yourself the re-briefing.

Important upfront, Projects is a feature in claude.ai (the web UI). It is not a coder feature. You do not need a terminal, no API, no MCP server. A Pro or Team account is enough. This playbook is for job seekers, solo founders, marketers, coaches and anyone who works with Claude without programming.

Step 1, one project per life area

The most common mistake is one mega project for everything. Do not do that. Split by area. I run four projects in parallel, "Job hunt 2026", "Newsletter StudioMeyer", "Client project Halim", "Personal learning". Each has its own instructions and its own knowledge.

Rule of thumb, if you need very different tonality or very different knowledge for two tasks, make two projects. If you only flip the tonality switch, one project with a custom instructions section for both modes is enough.

Step 2, write custom instructions

Custom instructions are what Claude additionally receives on every chat in this project. They are not a system prompt in the strict sense, but they behave similarly. The limit is several thousand characters, enough for a compact briefing page.

What belongs in there, your role ("I am a job seeker for junior frontend roles, focus React"), your style ("answers in Du form, short, no marketing speak"), recurring preferences ("when you write code, always with comments in German"), and anti-patterns ("do not write me cover letters with In today's fast-paced job market").

Write it in bullet form. Claude reads this every time, so keep it sharp and without fluff. If you read it two weeks later and think "I do not need this anymore", out with it.

Step 3, upload knowledge with a system

Project knowledge is the storage area. You can upload PDFs, Word, Markdown, text files. Claude has access to the content in every chat in the project.

Do not upload everything at once. Go strategically. For the job hunt project, my resume as PDF, three examples of my best work, a list of jobs I am applying for with a note why. For a client project, the brief, three earlier work samples, a list of agreed terms ("the client is not called company but Halim").

Rule of thumb, anything you had to explain more than twice in one chat belongs in here. Anything that changes weekly rather not, it goes stale.

Step 4, refresh knowledge

Project knowledge does not stay fresh forever. If you upload your resume in April and apply in November, that is outdated. Schedule a 5-minute routine every 4 weeks, open every project, scroll through the knowledge panel, remove what no longer holds, upload fresh.

For me that sits on the first Sunday of each month in the calendar. 20 minutes, four projects walked through, done. Without this routine Claude tips into wrong assumptions at some point because the knowledge is old.

Step 5, one starter prompt per project

Every project should have a clear first prompt that you or Claude use on every new chat in the project. For job hunt, "Here is a job posting, give me the top 5 requirements + which of my experiences match, short." For newsletter, "Here are three topic ideas, which is the strongest and why."

Write this starter into the custom instructions. Example line, "When I just write go, ask me what my task for today is and then offer a short list of fitting sub-tasks." That makes every chat sharp.

Step 6, organise chats in the project

In the project you see all chats that belong to this area. Default naming is mostly bad, "New Chat" or an auto title from Claude. Rename them as soon as you know what they are about.

In my job hunt project every chat reads as "Cover letter Company X 2026-05-08" or "Interview prep Company Y 2026-05-11". Date upfront matters, you find your way back faster later.

Older than three months without reuse, archive. You do not need them in your line of sight.

Step 7, use artifacts on purpose

When Claude in a project gives answers that are a document (cover letter, table, program plan), explicitly ask "put that in an artifact". Artifacts are documents in the right panel that you can edit on purpose and develop across multiple chats.

Practical for job hunt, one artifact per company. You can improve the text iteratively, Claude refers to the last version next time instead of starting from scratch.

Step 8, avoid tonality drift

After 20 chats in the same project, Claude often starts to adapt to patterns you did not want. If you said "write it more casual" three times, at some point he writes everything like a stand-up comedian.

Antidote, every 2 weeks reread your custom instructions, sharpen them, and on every new chat strictly check the first output. If the tone drifts, fix it in the custom instructions, not just in the chat. Otherwise Claude makes the same mistake at the next chat.

Step 9, data hygiene and sensitive content

Project knowledge is, according to Anthropic, not used for model training when you have a paid tier. Still, do not upload passwords, no internal contracts without NDA clearance, no personally identifiable info about third parties. If you use client material, ask first.

Rule of thumb I use, would I be able to email this content to a stand-in, yes, then it can go into the project. No, then not. Rather use an anonymised excerpt.

Step 10, share projects with the team

On the Team tier you can share projects with colleagues. Custom instructions and knowledge are then the same for everyone, every person has their own chats. Practical for marketing teams that maintain the same brand, or coaching practices with standardised tonality rules.

When you share, define a "keep shared knowledge fresh" role. Otherwise the knowledge rots because nobody feels responsible.

What's next

When this is in place, you know what a project looks like and how to maintain it. Next step is the lesson L2 System Prompts which shows what sits behind the concept of custom instructions, and the playbook Build a prompt library that helps you store your favourite prompts reusably. Whoever wants to start working with Claude Code in the terminal goes on to the playbook From ChatGPT to Claude.

Official Projects docs, https://www.anthropic.com/news/projects.