Mail, Teams and short messages, hands-on with Claude and ChatGPT
In this module you'll make four real work pieces: a short internal email, a polite refusal, an onboarding email to a new staff member in plain language, and a tone-shift. For each: the same prompt in ChatGPT and Claude, and you pick which one you keep.
Time: ~60 minutes. Grab a notebook or a Word document; write down every usable prompt and sentence you encounter. That becomes your library later.
Lesson 2.1: Short internal email (planning an extra staff meeting)
Goal: know how to put a question on the table in three lines that people actually answer.
People in a hospital receive lots of email. What helps: a subject line that reveals the content, question first, explicit deadline, concrete response option (“reply with yes/no”). An AI can build this in consistently if you ask.
Rule of thumb
A good internal email is scannable: a colleague should know within 5 seconds what you need from him or her.
Live exercise 2.1: Two versions, pick one
Open ChatGPT and Claude in two tabs. Paste the prompt in both. Read the answers side by side. Keep the version you would send.
Do a quick second round: ask the tool you picked: “Make it a touch shorter and more direct; remove all ‘please’ and ‘could you’ softeners.” Do you feel how the second version is tighter?
Question for lesson 2.1
1. What's the smartest way to work with AI on an email?
Lesson 2.2: Polite refusal (without being clumsy)
Goal: an email that refuses without leaving the recipient feeling slighted.
An email saying “no, unfortunately not” is needed more often in a hospital than we think: a request that doesn't fit, an appointment to cancel, an external party for which you don't have extra capacity. Claude is known to write calmly and neatly, that's the tool we pick here.
Live exercise 2.2: Refusal email in Claude
Open Claude. Paste the prompt below. Iterate a few times (“too formal”, “too warm”) to feel where your final version lands.
Iteration tips: “Make it 20% shorter.” “Remove all assuming sentences (‘I understand’, ‘I can imagine’).” “Add one sentence saying we're open to revisit in six months.”
Lesson 2.3: Onboarding email to a new staff member in plain language
Goal: write a welcome email that is immediately clear to someone who doesn't yet know the hospital.
Every time a new resident, fellow or nurse practitioner starts, the secretariat has a short list of practical things ready: badge, parking, IT access, first-week schedule, date of the first annual review. A new staff member reads that email between 30 other messages, often for the first time in the hospital, so: short sentences, no jargon, one clear action per paragraph. ChatGPT and Claude can both do it, with different tendencies: ChatGPT sometimes comes across as slightly too brisk, Claude is more formal and patient. Feel it yourself.
Live exercise 2.3: Onboarding email new resident
Paste in both tools and pick which one you keep.
Compare: which tool makes the email least overwhelming for a first day? Stick to one version, then ask: “Read again; is there still a sentence with more than 12 words? Split it.” A handy final polish step.
Questions for lesson 2.3
2. What's the difference between “plain language” for a new staff member and “simplifying until you leave something out”?
3. May an AI-generated email going out on behalf of the department head or training supervisor be sent immediately after generation?
Lesson 2.4: Tone shift (from warm to business and back)
Goal: quickly take an existing text to a different tone without rewriting it from scratch.
In practice you often write something quickly from memory and then realise: too formal, or too casual. AI is strong here. The trick is to paste the text first, then clearly state which tone you want and what must stay.
Live exercise 2.4: Three tone versions of the same email
Open Claude and paste. We'll let Claude make three versions: business-short, warm-helpful, and very short & informal.
Important: the line “Don't invent anything; use only what's in my text” is a great habit. Without it, models sometimes invent a fictional date, location or agenda item, classic hallucination behaviour.
Lesson 2.5: Good subject lines (that people open)
Goal: a subject line that, in 30 characters, says what the email is about and what's expected of the recipient.
Three quarters of an hour on the clinic floor, phone beeping, and you get an email with subject “Quick question”. Do you open it? Maybe later. A good subject line does three things:
- Tells what the content is (not how the sender feels about it).
- Contains the action expected of the recipient (FYI, to decide, deadline).
- Fits in 50–60 characters, in most inboxes the rest gets cut off.
AI is surprisingly strong at this if you give it the right task. Let's practise.
Bad
“Question”
“Quick something”
“Update on the meeting”
“FW: FW: FW: Re:”
Better
“Availability MDT 12/3, reply before Mon 17:00”
“Draft minutes last Fri, OK?”
“To decide: new clinic slots Tuesday”
“Reminder: agenda input needed before 15:00”
Live exercise 2.5: Ten subject lines in one prompt
Open Claude. Paste. Save the three best for your library.
Bonus: Ask as a follow-up: “Make the three best ones a touch shorter, max 50 characters.” Feel how sharp it can get.
Question for lesson 2.5
What does most of the work in a subject line?
Lesson 2.6: Responding to a difficult email to the department head
Goal: write a first response to a tough email with understanding, clarity, and without the department head committing to anything that still needs investigating.
Staff secretaries often manage the inbox of the department head or training supervisor. That inbox gets pointed emails: a resident unhappy about supervision, a neighbouring specialty complaining about MDT collaboration, an external party saying promises weren't kept. Your first response is always: a friendly acknowledgement of receipt, routing to the right person (department head, training supervisor or HR), and stating a deadline.
This is exactly where Claude shines. Friendly, not over-formal, no admission of fault on the department head's behalf, but acknowledgement of inconvenience. Important: this email does not replace the substantive reply, it is the first response from which the department head can take it further.
What you never put in a first response (especially not on behalf of a manager)
No admission of fault (“you are completely right that it went wrong”). No blame shifting (“the training supervisor should have…”). No commitment the department head hasn't personally approved (“we'll schedule a meeting right away”). Limit yourself to: receipt acknowledged, who's picking it up, within what timeframe someone will come back.
Live exercise 2.6: First response to a complaint to the department head
Open Claude. Paste the prompt with an idealised, fictional complaint. Don't use real names.
Iteration: “Make it 10% shorter without losing the acknowledgement.” “Add one sentence suggesting a phone call before Monday, only if you think that's appropriate.”
Question for lesson 2.6
What does a good first response to a difficult email to the department head do?
Lesson 2.7: Batch email: the same email personalised for 10 people
Goal: send ten personalised reminders without typing ten times.
Classic situation: ten external referrers each receive a tailored reminder for the quarterly meeting. Everyone has a different background (some attended last time, others didn't). This is exactly where AI saves time.
Live exercise 2.7: Ten reminders in one prompt
Open Claude. Paste.
Time: 10 tailored emails in one prompt. Typing manually would take 25-30 minutes; with this technique about 5. Important: you do read them all and tweak one sentence per piece before sending. AI provides the first draft; you remain the editor.
Lesson 2.8: Template prompt for mail (for your library)
Goal: you have one fixed template that you use for 80% of your mail work.
A template isn't a finished email; it's an empty prompt where you fill in the four things: role, context, task, format. Below is the template you may take for free. Save it in a text file, OneNote, Notion or inside a Claude Project. Looking ahead: in module 3 we'll actually start working with those Claude Projects, you'll pin this template in a fixed place so every new chat already knows it.
Live exercise 2.8: Test the template in Claude
Paste in Claude. Fill in where ‘[ ... ]’ appears. This is the standard prompt you use from now on for 80% of your AI mail work.
Save this template. It works in all three tools. You never have to phrase from scratch again, you fill in.
Take-home of module 2
Iterative working
First output is draft. One or two iteration rounds (“shorter”, “more direct”) make the difference.
Don't invent
The line “use only what I tell you” in your prompt prevents invented data, names and agenda items.
Claude for tone & onboarding language
For polite refusals and onboarding mail (new staff member), Claude often lands calmer and more patient; ChatGPT is faster but sometimes more direct.
Subject lines work
Content + action/deadline in <60 characters. Higher open rate, faster reply. Three sentences in one prompt yields ten.
Complaint to manager: timeframe, no blame
Confirm receipt + refer to department head / training supervisor / HR + name the timeframe. No blame, no commitments on someone else's behalf.
Batch with personal layer
Ten personalised emails in one prompt, you read through, tweak one sentence each, send.
Template now in library
Save the mail template from lesson 2.8. In module 3 you'll put it in a Claude Project; module 5 closes with it.