Free resource
Claude Starter Pack
An interactive guide for getting your team working from the same baseline on Claude. Pick your track, work the checklist, and check off steps as you go. Your progress saves in this browser.
Which track are you?
Two parts are for everyone whatever you pick: the safety section below and the prompting basics.
Before you connect anything
Your Gmail, Drive, and shared folders almost certainly hold sensitive information: personal data, financial records, donor or client details, and in regulated settings things like student records (FERPA) or health data (HIPAA). The moment someone connects Drive and points Claude at the wrong folder, that data is in play. This is the one part that isn't optional. A shared posture now prevents a hard conversation later.
- Know what's in the folder before you point Claude at it. In Cowork, Claude only sees the folders you give it. Folders with regulated or personally identifiable information (personnel files, financial or donor records, health or student data) stay off-limits unless your organization has cleared that use.
- Default to de-identified or aggregate data. Most drafting, analysis, and planning works fine with names stripped out. Make that the habit.
- Turn on "require approval" for tools that can change things. For connectors that can edit records, set Claude to ask before it acts. Read, don't write, until you trust the workflow.
- Vet plugins before installing them. A plugin bundles connectors and actions that widen what Claude can do in one click. Stick to verified plugins from the Claude directory and check what each requests.
- This is an org decision, not a personal one. Whoever owns data and compliance at your organization should set the line on sensitive data before the team goes wide. Put it in writing, even one paragraph.
On Claude Team and Enterprise plans, Anthropic does not use your conversations or connected data to train its models by default. That lowers the risk but does not erase your own data-protection obligations, which travel with the data regardless of vendor.
Use Claude Cowork safely · Anthropic Trust Center
A note from Jason: this is practical posture, not legal advice. Loop in whoever handles compliance to set the actual policy.
Your setup checklist
Steps below are grouped. The shared ones show for everyone; the rest follow your track. Check things off as you finish.
Add members, assign roles, manage billing in the admin console.
On Team plans an admin must turn on each connector before anyone can use it. Nothing in "Connect your tools" works until this happens. Set org-wide guardrails here too.
Decide what's in-bounds before the team goes wide. See "Before you connect anything."
A good prompt is clearer, not longer. Tell Claude four things: who it should be, what it's working with, what you want, and what the output should look like.
The habits that matter most
- Give it a role and context. "You're helping a nonprofit's development team" beats a cold ask.
- State the task and format. "Draft a 200-word funder update, friendly but professional, three short paragraphs."
- Show, don't just tell. Paste an example you liked. Claude matches patterns better than adjectives.
- Say what NOT to do. "No jargon, no exclamation points, don't invent numbers."
- Treat it as a conversation. The first answer is a draft. Push back: "tighter," "more concrete."
- Reuse what works. Save a good prompt as a Project instruction or a Skill so the team gets the same result.
Paste a real document and run the strong version. Feel the difference.
You're prepping me for a board meeting. Summarize the attached report in one page for trustees who haven't read it: three takeaways up top, then risks, then decisions they need to make. Plain language, no acronyms without spelling them out.You don't have to start over. Most of what you've built transfers in under an hour.
Paste one prompt into ChatGPT or Gemini, copy the result into Claude's memory settings. About five minutes.
What maps to what
- Custom Instructions → Settings, Personalization, or the memory import above
- Custom GPTs → Projects (paste the instructions, upload the files) or Skills
- Saved prompts → Project instructions, or keep a doc and paste
- Memory → Imports via the tool above
Ethan Mollick (Wharton) compares the major tools so you understand what's different, not just what's equivalent.
Then do the Track 2 steps. Claude 101 covers features ChatGPT has no equivalent for.
Three things, in order. They cover 80% of what most people need.
The basics: prompting, uploading documents, search, choosing models.
Free self-paced course on everyday work: writing, analyzing documents, organizing data. Includes a certificate, no technical background needed.
Pick a card from the section below and actually do it this week.
For people comfortable in Claude who want to work with real files and build reusable workflows.
Cowork is Claude's desktop mode for working directly with your files and folders. This is the structured on-ramp.
Projects are shared workspaces with their own instructions and knowledge files. Skills are reusable instructions for a specific task (your grant template, your tone, your reporting checklist). Admins can push Skills org-wide, which is how individual fluency becomes team consistency.
A plugin bundles Skills, connectors, and ready-made workflows into one installable package, so a team shares a whole way of working in a single click instead of rebuilding it person by person. Once your team lands on a workflow worth standardizing, a plugin is how you distribute it. Vet before installing (see the safety section).
Jeff Su (ex-Google, 1.2M subscribers) does tight, immediately applicable videos.
Learn 80% of Claude Cowork in Under 20 Minutes · Full playlist
Only works after your admin enables connectors at the org level. Re-read the safety section before granting Drive access.
Gmail, Calendar, and Drive. Then Claude can search email, check calendars, and read Drive files in a conversation.
A HubSpot Super Admin must connect it first. Turn on the setting that requires approval before Claude edits records.
QuickBooks has an official connector (finance and reporting), and the major note takers (Fathom, Granola, Fellow, Fireflies) all connect.
Try this first: five things to try this week
Concrete starting points from Anthropic's use-case library, each with the exact prompt to follow.
Work through grant options
See funders by odds, award size, deadline, and effort, then narrow to a shortlist. Development and leadership.
Open walkthrough →Draw your theory of change
Describe a program; Claude maps the causal chain, inputs through impact, with the assumption behind each arrow.
Open walkthrough →Adapt to every reading level
Turn one source page into versions for different reading levels plus a slide deck. Academic and instructional staff.
Open walkthrough →Turn meeting notes into a tracker
Pull decisions and action items out of your notes (and email or Slack if connected) into one view of who owns what and what's blocked.
Open walkthrough →Audit a folder of assets
Point Claude at a folder and your style guide; get back off-brand colors, old logos, and missing required copy.
Open walkthrough →Go broader
Sector examples and customer stories for your context.
Claude for Nonprofits →Claude for Education →
A simple rollout sequence
- Week 1: Admin finishes setup and enables connectors. Whoever owns compliance writes the sensitive-data policy. Switchers run the memory import; new users watch the video and start Claude 101.
- Week 2: Everyone reads the safety section, connects their own tools, tries one real task with each, and does one "Try this first" card.
- Week 3: Pick one recurring workflow as a team, build a Project for it, use it for real. Track 3 users lead.
- Ongoing: A 30-minute biweekly share-out where people demo one thing that worked. Pair a Track 3 user with a newer one.
PERISCOPE