Marketing Mastery Workshop · Day 2 · Tuesday 27 October 2026 · 10:00 to 17:00

Putting AI to Work

Follow this page from top to bottom. Copy a prompt with one click, and tick each step as you finish it.

Hands-on · laptop open Demo · watch Talk · listen Game

Before you start

By 17:00 your team leaves with a working AI setup and three finished files from your own material.

Log in to Claude at claude.ai on your laptop.
Open ChatGPT and Gemini in two more tabs. You need all three for the AI Arena.
Have your material ready: a long report (brand tracker, industry or policy report), last month's campaign data, and your Day 1 brief and one-pager.

PrivacyNever paste patient names, photos, records or chats into any AI tool. The full rules are in session 2.

Your details

1 min

Fill these in once. Your answers go into each prompt below, ready to copy. They stay in this browser.

Hospital

City

Specialty in focus

Competitor 1

Competitor 2

Competitor 3

Your seat

The AI landscape, for marketers

Pick the right AI for each job. 10:00 to 11:15

AI guesses the next word

Talk

ChatGPT, Gemini and Claude work like well-read autocomplete. Like your phone keyboard, they pick the most likely next word and can't check if it's true. You'll see two results of that in the Arena: invented facts and yes-man answers.

The AI Arena: you judge ChatGPT, Gemini and Claude

Game

Your team runs each prompt in its own tool, on the same input as the other teams. You get four minutes per round, then the room scores the answers.

Round 1 · Summarise a long document

Attach the shared 30-page report, then send:

Watch for: numbers that are not in the report, points that fit any hospital, missing page numbers.

Round 2 · Write in Bahasa, to a brand standard

Watch for: broken rules, and Bahasa that reads like a translation.

Round 3 · The yes-man test

Read it like a confident pitch.

A good answer names the wrong tone for this department, the risk to patient dignity, and the wrong audience.

Scorecard

Rate each answer from 1 to 5. Score accuracy after round 1, Bahasa and the standard after round 2, and pushback after round 3.

Criterion	What we ask	ChatGPT	Gemini	Claude
Accuracy	Is every fact right? Anything invented?
Bahasa quality	Does it sound like a real Indonesian marketer?
Follows the standard	Did it obey the brand rules we gave it?
Honest pushback	Did it tell us the bad idea is bad?
Total		0	0	0

SycophancyAI tools learn to please you, so they agree with bad ideas. Ask "Headline ini bagus kan?" and you get a yes. Ask why it will fail, and you get real problems.

Our team ran all three rounds and filled the scorecard.

The method: PTCTFE, plus Check

Demo

Your prompt is your coffee order. Say "Mas, es kopi susu satu" and the barista makes the house default. Give the AI no details and it gives you the safe guess. This prompt leaves it guessing the hospital, the audience and the goal:

P · Persona

The role, e.g. a senior hospital marketer.

T · Task

The output you need, starting with a verb.

C · Context

Hospital, audience, situation, Day 1 brief.

T · Tone

The voice, e.g. warm and credible.

F · Format

Structure and length.

E · Exemplars

Paste a post you liked.

+ Check

Facts, numbers, medical claims, brand rules. Never ship an AI answer you have not read.

Build your prompt

Edit any line, then copy the finished prompt from the box below.

I ran the weak prompt and the PTCTFE prompt and compared the answers.

Meta-prompting: let the AI interview you first

Hands-on · 10 min

With PTCTFE you write all the context upfront. With meta-prompting the AI asks you for the context it needs. Use it for new tasks.

Half the table: rewrite your Round 2 prompt with PTCTFE.
Other half: use meta-prompting for the same caption.
Both: run the Check line, then compare. Which answer is better?

The first answer is a draft

01 Ask

PTCTFE, or let it interview you.

02 Check

Read it like an editor.

03 Say what is off

Name the problem: too long, too stiff.

04 Ask again

Repeat until it is right.

Four AI limits that hit hospital marketing

Talk

Hallucination

It invents facts with full confidence, including medical claims and numbers.

Bias

It repeats patterns from its training data. Ask it for a doctor and check who it draws.

Sycophancy

It agrees with you to please you. Ask it to disagree.

Copyright and disclosure

Know when to say AI was used. Medical claims need clinical sign-off.

Try to catch it guessing

Turn web search off, then send this. Search two of the titles yourself. If one does not exist, it guessed.

The rule for numbers and medical claimsOpen the source for each number. Get a doctor's sign-off on each medical claim. Cut anything you cannot check. If a Siloam chatbot invents a promo price, Siloam pays it.

I tried to catch the AI guessing and checked two of its sources.

Claude, set up for hospital marketing

You brief Claude once and use it for months. 11:15 to 12:30

Build your hospital project in 15 minutes

Hands-on

You give a new agency hire an onboarding pack before they touch a client. A Claude project is that pack: you brief Claude once, and each chat in the project starts with your brief.

Create the project. In Claude, open Projects, then create a new project. Name it after your hospital.
Add the profile. Paste the project instructions file above into the project instructions. Fill in the brackets: location, specialties, priority battle.
Add your voice. Upload three posts you are proud of, plus your brand rules.
Add the Day 1 brief. Get, Who, To, By from yesterday. No brief? Fill the template above.
Test it. Send the question below. A good answer repeats your Day 1 brief in its own words.

Privacy rules for any AI tool

Talk

RED · Never

Patient names, faces, photos
Medical records, diagnoses, lab results
ID numbers, phone numbers
Chats with patients

YELLOW · Remove details first

Internal revenue figures
Unreleased campaigns
Partner and vendor contracts
Staff data

GREEN · Go

Public websites and posts
Competitor content
Your published campaigns
Totals with no names

Yellow, made safe"Revenue Rp 4,2 M in September" becomes "revenue up 12% month on month". For an unreleased campaign, share the brief and remove launch dates and budget.

Vote: red, yellow or green?

Pick a colour for each item. You see the answer after you pick.

Install skills: your team's recipe cards

Hands-on

A skill is a recipe card pinned behind the bar. Write the house drink once and any barista makes it the same way. Claude picks up the right card when your request matches it.

Before you startUpload the zip as it is. Do not unzip it first. On a Mac, download with Chrome, because Safari unzips files by itself. Upload one zip at a time. Start with stop-slop.

Switch on code execution in Claude: Settings, then Capabilities. Without it, Claude can't run skills.
Download a skill zip from the table below.
Open Skills. In Claude: Customize, then Skills.
Upload. Click +, Create skill, Upload a skill. Pick one zip.
Switch it on. Check the toggle next to the skill is on.
Test it. Paste a caption and send the prompt below.

Stuck? Raise your hand. A helper will come to your table.

A starter pack of free skills, by seat

The skills for your seat show in green. Install the first three whatever your seat.

⬇ All skills in one pack (.zip, 0.6 MB) Unzip this pack once. On Windows, right-click it and choose Extract All. On a Mac, double-click it. Then upload the zips inside it as they are.

Skill	What it does	Seat

NoteCorey Haines wrote the marketing skills for US software marketing. Your hospital project adds the rules for Indonesian hospital marketing. Hardik Pandya wrote stop-slop and Matt Pocock wrote grilling. All use the MIT licence.

Read the card before you pin it

The barista follows each card to the letter, bad cards included. A skill file can hide instructions or a small program, so check any skill from outside this page before you upload it.

Found a skill with no zip? Ask Claude to pack it

Claude sends a zip back. Download it, then upload it the same way: Customize, Skills, Upload.

At least two skills are switched on in my Claude.

Build your own skill in 15 minutes

Hands-on

Pick a task your team does every week, the same way. For example: the weekly competitor promo check, a doctor profile for the website, or an event recap post.

Pick a task you do weekly, the same way.
Run the prompt above. If /skill-creator does not appear, send the same prompt without that first line.
Add an example of good output, without patient data.
Save and upload. Zip the whole skill folder. On Windows, right-click the folder and choose Compress to ZIP file (on Windows 10: Send to, then Compressed folder). On a Mac, right-click the folder and choose Compress. Then upload the zip under Customize, Skills.
Test it three times. If it does not start, rewrite the description.

Inside a skill

A skill is a folder with one file, SKILL.md. The description is the drink name on the card: Claude reads it to decide when to use the skill. Name the input and the trigger.

One workflow per skill

Start with plain written steps

Add anything fancy later.

No secrets on the card

No passwords or sensitive information.

First skill run: the morning scan

Hands-on · 20 min

Turn web search onEvery AI model stops learning on a certain date. Ask about anything newer and it guesses. With web search on, it looks up today's pages first and answers with links you can check.

Run this inside your hospital project.

I ran the morning scan for my hospital.
Check: I opened two sources and they say what Claude says they say.

A line to challenge: anything without a link or a date.

The productivity lab

You finish three weekly jobs before the break. 13:30 to 15:15

Lab rules

Talk

01 Watch

A 5-minute demo on screen.

02 Do

Same job, your own material.

03 Check

Facts, numbers, brand rules.

04 Log

Minutes saved, on the leaderboard.

RuleUse your own material, and keep red data out. Run every job inside your hospital project.

Job A · 40 pages into one page: what this means for us

Hands-on · 30 min

Input

A long report you never had time to read: brand tracker, industry or policy report.

Run

Attach the report and send the prompt. Then check every number.

Output

One page: three findings that matter to us, and what we do next.

My one-pager ends with what we do next.
Check: I opened the pages and confirmed every number.

Job B · a brief into a management-ready deck

Hands-on · 35 min

Input

Your Day 1 brief, or a data set for leadership.

Run

Ask for 6 to 8 slides. Check every chart against the data.

Output

A PowerPoint file you can edit and present tomorrow.

Agency habitPut the decision you need on slide 1 as well. Directors read the first slide first.

New message from the client

Client · 14:22"Mas, deck-nya bagus. Tapi bisa lebih singkat, lebih premium, dan logonya digedein dikit? Butuh sebelum jam 3 ya. Thanks!"
You have 2 minutes. Revise with Claude.

I downloaded the revised deck as a PowerPoint file.

Research on live websites, inside your browser

Demo

Claude in Chrome, or the browser in the Claude app, can open live websites and compare them for you. Use public websites only.

Check: open one competitor page yourself and confirm one price.

Job C · campaign data into the monthly one-pager

Hands-on · 30 min

Input

Last month's campaign data: spend, leads, new patients.

Run

Attach the data and the Day 1 one-pager template. Check every total.

Output

The one-pager leadership reads every month, ready to send.

The flag you want"60 new patients but only 12 bookings recorded. Check the CRM export." You want Claude to question the data before it writes the summary.

Claude flagged odd numbers, and I checked each one.
Check: every total matches the source data.

Minutes saved per month

Game

Enter how long each job takes by hand and with Claude, and how often you do it each month. Read your total out before the break.

Job	By hand (min)	With Claude (min)	Times per month	Hours saved
A · one-pager
B · deck
C · monthly report
Total per month

Row A shows an illustrative example: 3 hours by hand, 25 minutes with Claude, twice a month. Replace it with your own numbers.

Our team logged its total on the leaderboard.

BrandMind ID: the content engine

Strategy into content plans and campaigns. Watch only. 15:30 to 16:15

Your team and the content engine split the work

Talk

Your team, with Claude

Morning scan and research
Reports into one-pagers
Decks for leadership
The monthly report
Briefs and honest critique

The content engine

Positioning into content plans
Campaign content at volume
Brand consistency across five hospitals
Content that follows the Day 1 strategy

Our team agreed its split of the work.

Cheat sheet

Every prompt: PTCTFE + Check

Persona, Task, Context, Tone, Format, Exemplars. Then check facts, numbers, medical claims and brand rules.

Not sure what to give it?

"Before you write anything, ask me the questions you need answered, one at a time."

Want honesty?

"Sebutkan 3 alasan ini akan gagal." Ask for the criticism and you get it.

Anything recent?

Turn web search on. Ask for a link and a date on every line.

Never in any AI tool

Patient names, faces, records, ID or phone numbers, chats with patients.

Before you publish

Open the source for each number. Doctor's sign-off for each medical claim. Cut what you cannot check.