Framework · free to use

How to train your team on Claude.

Buying Claude seats does not change how your team works. Training does — if it's the right training. Most internal AI training fails the same way: a 60-minute generic demo, polite nodding, then no behavior change. This is the playbook for training that actually sticks.

The principle

Train on workflows, not features

Generic AI training ("here's what an LLM is, here's how prompting works") produces awareness, not adoption. Workflow-specific training ("here is exactly how you, the sales rep, will draft a follow-up email using this Project") produces adoption.

Rule: every training session should leave the participant with one specific workflow they can do that afternoon, faster than they did yesterday.

Week 1

Foundations (90 minutes total)

Session 1: What Claude is and how we use it (30 min)

What Claude is, in one slide.
What we use it for here (specific workflows we've already built).
What we do NOT use it for (data classifications, policy summary).
How to access (login, Projects you have access to).

Session 2: Prompting basics (30 min)

How to write a clear prompt (with 3 examples in the team's actual workflow context).
How to refine when the output is off ("more concise," "add an objection-handling section," "shorter").
How to use Claude Projects vs the standalone chat.

Session 3: First real workflow, hands-on (30 min)

Pick the most relevant Project for their role.
Walk through one full real-work use, step by step.
Have each person do their own example with a real input from their own work.
Office hour scheduled for 1 week later.

Week 2-4

Reinforcement & refinement

Weekly office hours with the AI lead. Anyone can come; bring real work that is not flowing.
Slack channel for prompt sharing. "This prompt worked great" or "Why is Claude refusing X?"
Workflow owners check in 1:1 with each team member to see what's actually being used.
Update the Project knowledge based on patterns from real use.

Role-specific exercises

What to have each function do

Role	First workflow	Time to fluency
Sales rep	Draft a follow-up email from real call notes	1-2 sessions
Marketer	Draft a content brief from a topic + target keyword	2-3 sessions
CSM	Build a QBR pack from usage data + account notes	2-3 sessions
Executive	Get a one-page brief from a board doc	1-2 sessions
Ops	Summarize meeting notes into action items + decisions	1 session
Support	Draft a reply to a real ticket	1-2 sessions

Common failure modes

Where adoption stalls

People use Claude once, get a meh output, give up. Fix: show them how to iterate. "More concise, shorter, add X" turns a meh draft into a good one in 30 seconds.
People use it for the wrong workflows. Fix: be explicit about the workflows where it shines vs the workflows where it does not yet.
People do not have time. Fix: assign protected calendar time for the first two weeks. "Use Claude for 30 min/day, every day, for the next 10 work days."
People do not trust the output. Fix: review some of their outputs together. Build trust calibration. Show what Claude is reliably good at and what needs human verification.
The Project is not loaded with the right knowledge. Fix: iterate. Most Projects need 3-4 rounds of knowledge updates before they hum.

Measuring adoption

How to know it's working

Usage frequency: are people using Claude weekly, daily, or never? Daily users adopt; weekly users sometimes; never-users will never absent intervention.
Workflow output: are proposals/content/replies actually shipping faster, with the AI tool?
Voluntary sharing: are people posting prompts and patterns in the Slack channel without being asked? That's the strongest signal of organic adoption.
1:1 check-ins: ask each person, "what's the workflow you use Claude for most often?" If they can name one, they're adopted. If they can't, they're not.