Setting Up Claude AI Projects A Complete Guide for Teams

Claude Projects are shared AI workspaces that give teams a persistent context layer storing documents, custom instructions, and chat histories so every team member works with the same institutional knowledge. On the Team plan ($20–$125 per seat/month), they add permission controls, workplace connectors, SSO, and admin tools. Set up correctly, a Claude Project eliminates the “blank chat” problem: new members open a Project where Claude already knows your brand voice, your style guide, your policies, and your templates.

“The Claude Team plan is transforming our way of working. Claude has helped our team complete content creation and analysis tasks up to 5x faster than before turning what was once two weeks of writing and research into minutes of work.” Luka Anic, Senior Director at North Highland

Claude Plans: Feature Comparison Table

Feature	Pro ($20/mo)	Max ($100–$200/mo)	Team Standard ($20–$25/mo/seat)	Team Premium ($100–$125/mo/seat)	Enterprise
Projects	Yes	Yes	Yes (shared)	Yes (shared)	Yes (shared)
Context window	200K tokens	1M tokens	200K tokens	200K tokens	1M+ tokens
Usage per session	Baseline	5x–20x Pro	1.25x Pro	6.25x Pro	Custom
RAG on projects	Yes	Yes	Yes	Yes	Yes
Workplace connectors	No	Yes	Yes (Google Drive, Slack, GitHub, Microsoft 365, Gmail, Calendar)	Yes	Yes
SSO + domain capture	No	No	Yes	Yes	Yes
Admin controls	No	No	Yes	Yes	Yes (full)
Spend caps	No	No	Yes	Yes	Yes
API access	No	No	No	No	Yes
Minimum seats	1	1	5	5	Custom
Max seats	1	1	150	150	Unlimited

All pricing sourced from Anthropic’s Team plan help page (updated May 2026). Team plans do not train on your data.

1. What Claude Projects Actually Are

Claude Projects are self-contained workspaces each with its own chat history and knowledge base available to all Claude users including free accounts (with a 5-project cap). On Team and Enterprise plans, they become shared collaboration hubs.

Each Project has three layers that make it valuable:

• Project Knowledge: Uploaded documents (PDFs, text, code, spreadsheets) that Claude references across all chats inside that Project. Think style guides, product specs, pricing sheets, runbooks, and competitive intel. Claude does not automatically pull context from one chat to another only Project Knowledge persists across chats.
• Project Instructions: A system-level prompt that runs before every chat. Define tone, audience, role, formatting rules, guardrails, and what Claude must never do. Example: “You are helping the customer success team. Use concise language. Do not invent pricing or contract terms. If a policy is missing, say so.”
• Chats: Individual conversation threads within the Project, each benefiting from the shared Knowledge and Instructions layers above. Multiple team members can create chats simultaneously.

Retrieval Augmented Generation (RAG) kicks in automatically on paid plans when your Project Knowledge approaches context limits expanding capacity by up to 10x while keeping response quality high. RAG retrieves the most relevant chunks of your knowledge base for each query, but it does not exempt you from curating what you upload. Stale policies and current policies coexisting in a Project means retrieval can surface the wrong one.

2. Team Plan Pricing and Usage Limits

Anthropic’s Team plan requires a minimum of 5 seats and supports up to 150.

Standard seats:

• $25 per member/month (billed monthly)
• $20 per member/month (billed annually)
• 1.25x more usage per session than Pro
• One weekly usage limit across all models

Premium seats:

• $125 per member/month (billed monthly)
• $100 per member/month (billed annually)
• 6.25x more usage per session than Pro
• Two weekly limits: one across all models, one for Sonnet models only

Teams can mix seat types assign Premium to power users and Standard to occasional users. Usage limits are per-member, not team-wide. If one person hits their cap, others are unaffected. Organizations can also prepay for usage credits so work continues instead of waiting for limits to reset.

3. Workplace Connectors: Pull Context Instead of Uploading It

Team and Enterprise plans include workplace connectors native integrations that let Claude search and retrieve from the tools your team already uses:

• Google Drive, Gmail, and Google Calendar
• Microsoft 365 and Slack
• GitHub repositories

Claude fetches relevant context from connected documents, emails, calendars, and team channels without manual uploads. Enterprise search gives every team member a unified search bar across all connectors from one place. For teams drowning in scattered documentation, connectors eliminate the “where is that slide deck” friction.

4. Admin Controls and Security

Team plans include centralized administration missing from individual plans:

• Single Sign-On (SSO) and Domain Capture: Automatically authenticate users with your company email domain. Just-in-Time (JIT) provisioning streamlines onboarding.
• Role-based permissions: Customize who can access specific Projects. Permission levels are “Can use” (view, chat, cannot edit) and “Can edit” (modify instructions, knowledge, manage members).
• Spending controls: Set caps at both organization and individual levels.
• Centralized billing: One dashboard for all seats and usage.

For Enterprise plans, additional controls include audit logging, zero data retention options, and custom system prompts. Team and Enterprise plans do not train Anthropic’s generative models on your data without explicit consent.

5. How to Structure Projects for Your Team: The 5-Part Framework

Build a separate Project for each recurring deliverable your team produces. A law firm creates Projects for client memos, case summaries, demand letters, and internal reports. A marketing team creates Projects for campaign briefs, blog posts, competitive analysis, and social media calendars.

Step 1: Identify your deliverables. Use this prompt in one Claude chat:

I work at [company + industry]. My team helps [clients] [achieve goals].
You are helping me set up Claude for my team. We need to identify the 3-5 recurring deliverables my team produces, so we can create a separate Claude Project for each one. Interview me. Ask me one question at a time about:
1. What my team does day-to-day
2. What we deliver to clients, leadership, or each other
3. What tasks feel repetitive every week or month
4. What work someone always ends up redoing because the first version wasn't right
When you have enough context, stop and give me a numbered list with suggested Project names and exactly which documents to upload.

Step 2: Create Projects and upload the right context. For each Project, upload only what that deliverable needs. A sales proposal Project does not need your brand style guide. A meeting recap Project does not need your pricing sheet. Upload:

• One “gold standard” example of the deliverable
• Any relevant background documents (client roster for client updates, pricing for proposals)
• The template or brief your team currently follows

Step 3: Generate Project instructions. Claude’s help center is explicit: the Project name and description are not automatically sent as context. You must put your rules into Project Instructions. Use this prompt to generate them:

I'm setting up this Claude Project for one specific recurring deliverable my team produces: [DELIVERABLE NAME].
I've uploaded example outputs and background docs. Your job: Generate a Project instruction set I can paste into this Project's instructions field. Include:
- WHAT THIS DELIVERABLE IS: One sentence
- WHO IT'S FOR: Audience
- TONE & FORMAT: How it should read, structure, length
- QUALITY BAR: What separates a good version from a bad one
- GUARDRAILS: What to never do (never speculate on timelines, always include next steps, never exceed one page)
Format as a ready-to-paste instruction block.

Step 4: Create prompt templates. The #1 reason teams fail at AI adoption is the blank chat. Remove that friction with one-line templates stored in each Project’s Knowledge section. Use this prompt:

Based on the instructions in this Project, write me the shortest possible prompt template my teammates can copy-paste to produce this deliverable. One sentence max. One [INPUT] field. The template should rely on the Project instructions for everything else.

Step 5: Test and validate. Run each Project with a real scenario. Ask Claude to produce the output, then critique its own work. If tone or format is off, tweak instructions. Takes 2 minutes per Project.

6. Real-World Team Project Types

• Customer Support Project: Upload support policies, refund rules, escalation paths, approved macros, product limitations, and tone examples. Every agent gets consistent answers.
• Sales Project: Upload ICP criteria, qualification frameworks, objection handling, pricing guardrails, case studies, and legal handoff rules. Every proposal pulls from the same playbook.
• Marketing Project: Upload positioning docs, persona notes, brand voice guidelines, competitive messaging, campaign history, and approved proof points. Every piece of content stays on-brand.
• Engineering Project: Upload architecture summaries, coding conventions, runbooks, API references, and incident postmortems. Onboarding new developers becomes 5x faster.
• Executive Team Project: Upload board update formats, strategic priorities, OKRs, operating metrics, and decision memos.

7. The 5-Day Team Rollout Plan

Ruben Hassid’s team setup playbook (tested with 350K+ subscribers) provides a practical timeline:

Day 1 (45–60 min): Create 3–5 Projects using the framework in Section 5. Generate instructions, upload documents, validate outputs. Enable Extended Thinking for complex Projects.

Day 2 (15–25 min): Generate one-line prompt templates for each Project. Save them where teammates can find them.

Day 3 (20–30 min): Run one real task through the system. Screenshot the before/after comparison. This is your adoption proof.

Day 4 (15–20 min): Onboard one colleague the person who’s drowning, not the skeptic or the enthusiast. Use their actual work. Show the comparison from Day 3. Make them a co-owner of the Project.

Day 5 (60 min): Roll out to the full team. Send a Slack message that leads with the time saved. Include the prompt templates. DM 2–3 people individually asking them to try one template today.

8. Governance That Prevents Project Rot

Projects degrade without active maintenance. Three practices keep them healthy:

• Monthly reviews: Someone must review each Project for outdated content, stale documents, and unused conversations. Archive or delete what no longer serves.
• Designated ownership: Every Project needs one person accountable for knowledge quality. Diffusion of responsibility leads to neglected content.
• Contribution standards: Team members must understand what belongs in the Project and how to structure new contributions. Without conventions, Projects become dumping grounds.

9. Common Mistakes That Kill Team Projects

Three patterns destroy the value teams hope to create:

• Unbounded growth without structure: Adding everything to a Project eventually makes finding anything impossible. Invest in organization from Day 1.
• Treating Projects as dumping grounds: Not every chat belongs in a shared Project. Individual troubleshooting, performance reviews, and sensitive discussions stay private. Selectivity preserves value.
• Skipping onboarding: When someone joins the team, they need context about how the Project works, what knowledge exists, and what conventions the team follows. Without it, they either misuse the Project or ignore it entirely.

FAQ

Can I control who sees specific conversations within a Project? Project-level access typically extends to all Project content. For sensitive discussions, use separate conversations outside the shared Project. Chats have different sharing behavior than Project Knowledge do not assume everything is visible.

Can free users access Projects? Yes. Free accounts get up to 5 Projects with knowledge uploads and custom instructions. RAG, higher limits, and team sharing require a paid plan.

Does Claude train on Team plan data? No. Anthropic states that Team and Enterprise plan data is not used to train generative models without explicit consent covering Projects, chats, and uploaded documents.

How do I handle employee departures? Project knowledge persists as long as someone maintains ownership. Transfer ownership before departure. Admins can reassign Projects from the central dashboard.

Can multiple people use a Project simultaneously? Yes. Multiple team members can create chats, upload documents, and contribute concurrently within the same Project.

Which workplace connectors are available? Team plans include Google Drive, Gmail, Google Calendar, GitHub, Microsoft 365, and Slack. Custom connectors available on Enterprise plans.

Sources

• Anthropic Help Center: What are Projects? (updated March 2026)
• Anthropic Help Center: What is the Team plan? (updated May 2026)
• Anthropic Help Center: Project visibility and sharing
• Anthropic Help Center: RAG for Projects
• Anthropic Help Center: How large is the context window? (updated May 2026)
• Anthropic News: Collaborate with Claude on Projects (June 2024)
• Anthropic Trust Center
• Tactiq: Claude Team Pricing, Features & Benefits (updated April 2026)
• Ruben Hassid: Claude for Teams (March 2026)

Final word: The difference between teams that get value from Claude Projects and teams that fail isn’t intelligence. It’s intentionality. Spend 45 minutes structuring your first Project. The compounding returns shared context, reusable templates, consistent outputs make that hour the highest-leverage investment your team will make this quarter.