Module 5 of 7

Module 5: Decks, Docs, and Role-Specific Patterns

Why this matters

Documents and decks are where most knowledge work shows up, and where most people lose hours to blank-page paralysis. AI doesn’t replace your thinking; it gets you out of the blank-page state and into editing, which is where the real thinking happens. This module also gives you a short track for your specific role.

Outline first, prose second

The single biggest mistake people make with AI for docs and decks is asking for the finished thing immediately. The better workflow:

  1. Gather your reference material. Paste in or upload anything relevant, meeting notes, source docs, prior versions of the doc, examples of the voice you’re going for. AI does much better work with grounding than with a blank prompt.
  2. Ask for an outline (sections, bullets, structure)
  3. Edit the outline yourself, this is where your judgment goes
  4. Ask AI to expand the approved outline into prose
  5. Edit the prose
I need to write [a doc / a deck / a memo] about [topic] for [audience].
The goal is [what the reader should do or believe after].

Reference material below — use this as grounding, don't invent beyond it:
[paste notes, source docs, prior version, voice examples — or upload files]

Give me an outline only — section headings and 2–4 bullets per section.
Don't write any prose yet.

Then, after editing:

Here's the approved outline. Expand each section into prose.
Tone: [direct / warm / formal]. Length per section: [target].
Don't add any sections that aren't in the outline.
Stick to the reference material — don't invent facts or sources.

This pattern keeps you in control of structure, the part that actually carries the argument, and keeps AI grounded in real material rather than confidently making things up.

Edit-pass prompts

Once a draft exists, run a structured edit pass:

Run them one at a time, not all at once. Mixed instructions make weaker edits.

The “audience hat” trick

Before you send anything, ask AI to read it in the audience’s shoes:

Read the document below as if you are [the audience — e.g., "a CFO who hasn't seen the project before"].
Tell me:
1. What's confusing or unclear?
2. What questions you'd have walking out of this?
3. What seems like jargon or insider language?

[paste doc]

Catches the most expensive class of doc errors, the ones where the writer is too close to the material to see them. This is the “second perspective, not oracle” frame applied to documents, AI isn’t telling you the doc is good or bad, it’s surfacing perspectives you couldn’t see from inside your own draft.


Role tracks

Pick the one that matches your job. ~5 min each.

Project Management

The PM workday is full of AI-shaped tasks: status updates, risk logs, stakeholder comms, scope summaries.

Status update prompt:

Draft a weekly status update for [project name] covering:
- Progress this week: [bullets]
- Blockers: [bullets]
- Next week: [bullets]
- Risks: [bullets, optional]

Audience: [exec / team / cross-functional]. Length: [short — max 200 words].
Use clear section headers.

Risk register helper:

Given these risks: [paste list], for each one give me:
1. Likelihood (low/med/high)
2. Impact (low/med/high)
3. One mitigation in plain language
4. Trigger condition (when it stops being a risk and starts being a problem)

Data and Spreadsheets

If your job involves numbers, the trap is asking AI to do math. It’s bad at exactness. Use it for explanation and formula generation, not calculation.

Formula generator:

I have a spreadsheet with these columns: [list].
I want to [what you want to do, plain language].
Give me the formula for [Excel / Google Sheets].
Explain what each part of the formula does.

Explain-the-data prompt:

Below is a [table / paste of data].
Tell me:
1. What's the most surprising thing about this data?
2. Three trends or patterns I should look at more carefully
3. Three numbers I should double-check (because they look unusual)

Don't compute anything you can't be confident in — flag anything you'd want me to verify.

[paste data]

Customer-Facing Roles (Support, Sales, Service)

Your job is high-volume, high-context replies. The wins are in tone calibration and handling the awkward middle of a thread.

De-escalation reply prompt:

Below is a customer message that's frustrated/angry/confused.
Draft a reply that:
1. Acknowledges what they're feeling without being saccharine
2. States clearly what we can and can't do
3. Gives them one concrete next step
Tone: warm but professional. Length: short.

[paste customer message — redact identifying info first]

FAQ-from-conversation prompt:

Below is a [customer thread / call transcript].
Pull out the 3 most likely questions a future customer would ask in the same situation,
and a clear answer to each. Format as Q/A pairs.

[paste content — redact identifying info first]

Operations / Admin

The ops workday is process docs, SOPs, recurring comms, and spreadsheet hygiene. Your win is codifying tribal knowledge.

SOP-from-conversation prompt:

Below is a [Slack thread / email chain / set of notes] where we figured out how to do [task].
Turn it into a clean SOP. Format:
- Purpose (one sentence)
- When to use this
- Step-by-step (numbered, plain language)
- Things to watch for
- Who to ask if it goes wrong

[paste content]

Recurring-comms template prompt:

I send a [type of message] every [cadence]. Each one covers [content shape].
Build me a reusable template I can fill in each time. Mark fields with [BRACKETS].
Keep it short and skimmable.

Try It At Work: Outline-First on a Real Doc

Time: 15 min

You’ll need: A doc, deck, or memo you actually need to write this week, plus any reference material (notes, source docs, prior versions). Confirm safe to paste, or redact first.

Do this:

  1. Paste or upload your reference material.
  2. Run the outline prompt, outline only, no prose.
  3. Edit the outline. Reorder, cut, add. Make it yours.
  4. Run the expand-to-prose prompt with your edited outline.
  5. Edit the prose to your voice.

Done when: You have a draft you’d actually send, and you can point to which sections came from your outline edits vs. AI’s first pass.

Try It At Work: Run Your Role’s Prompt

Time: 10–15 min

You’ll need: A real artifact from your job that matches your role’s prompt.

Do this:

  1. Pick the prompt above for your role.
  2. Apply it to a real situation, your real status update, your real customer message, your real spreadsheet question.
  3. Compare AI’s output to what you would have written. Note where it’s better, worse, or different.

Done when: You’ve used the role-specific prompt on real work and decided whether to keep it as a regular tool.

Key takeaways

Quick Check

1. Why ask for an outline before prose?

2. The "audience hat" prompt is best used

3. For data work, what's AI bad at?

4. The right way to run multiple edit instructions