The Problem Isn't Writing. It's Deciding What to Say.

Most managers don't write badly. They write defensively. Updates hedge. Reviews recycle last quarter's adjectives. Hard messages get padded with so much cushion the recipient walks away unsure they were even criticized. AI makes this worse if you let it — it loves the passive voice and "I wanted to circle back."

Your job is to use AI to think clearly first, then write tightly. Never the other way around. If you can't say what the message is in one sentence, no model can rescue you.

Three formats eat 80% of a manager's writing time: status updates, performance reviews, and hard conversations. Get these right and the rest is easy.

Status Updates: Cut the Narrative, Keep the Signal

A good update answers three questions: What shipped? What's at risk? What do I need from you? Everything else is decoration.

Feed your AI the raw material — Slack threads, Jira exports, your own scratch notes — and force it into a structure. Don't ask for "an update." Ask for a specific shape.

You are helping me write a weekly update to my VP.
She reads it on her phone in 90 seconds.

Input: [paste raw notes, ticket dumps, metrics]

Output format:
- Shipped this week: 3 bullets, each one sentence, verb-first
- At risk: 1-2 items with the actual blocker, not "we're monitoring"
- Need from you: a specific ask or "nothing this week"

No adjectives. No "excited to share." No hedging verbs
like "started exploring" — say what you did or omit it.

Before (what the model gives you on a vague prompt):

This week the team made significant progress on the checkout redesign and continued to align stakeholders around the Q3 roadmap. We are actively monitoring some risks around vendor timelines and excited to share early results from the new onboarding flow.

After (with the structured prompt):

Shipped: Checkout v2 to 10% of traffic. Conversion +3.1% vs. control. Cut two abandoned experiments from the backlog. At risk: Vendor API for tax calc is slipping a week — pushes EU launch from Sept 12 to Sept 19. Need from you: Sign-off on the launch date change by Friday.

Same week. One is noise. The other is a decision-ready document.

Performance Reviews: AI as the Second Reader, Not the First Writer

Do not let AI draft a review from scratch. It will produce something plausible, generic, and slightly wrong — and your report will sense it instantly. Reviews are the moment your team checks whether you actually pay attention.

Write the review yourself in rough form. Then use AI as a sharper version of yourself reading it back.

You are a tough senior manager reviewing my draft review of [name].
Their level is [IC4 / senior PM / etc]. The review is for the past 6 months.

Draft:
[paste your draft]

Critique it for:
1. Specifics — flag every sentence that could apply to any employee.
2. Calibration — am I confusing "nice person" with "high performer"?
3. Actionable growth — is the development area something they can
   actually work on, or vague like "communication"?
4. Tone — does it sound like me, or like a textbook?

Don't rewrite it. Tell me where it's weak.

This works because you keep judgment in your hands. The model becomes the colleague who reads your draft over coffee and says "the third paragraph is vibes, not evidence."

Before:

Sarah is a strong collaborator and a positive presence on the team. She consistently delivers her work and is well-liked by peers. She should continue to develop her communication skills.

After (your rewrite, after AI critique):

Sarah owned the migration to the new payments service end-to-end — scoped it, ran the cross-team review, shipped it two weeks early with zero rollbacks. The gap: her design docs are dense and reviewers often ask the same clarifying questions. Next half, I want her to lead two RFCs where the discussion stays on the design, not the document.

The first version could be about anyone. The second couldn't be about anyone else.

Hard Conversations: Write It, Read It Aloud, Then Send

Difficult messages — performance concerns, declined promotions, layoff conversations — fail in two predictable ways. Either they're so blunt they read as cold, or so softened the person doesn't understand they're in trouble.

AI is genuinely useful here, but only if you brief it with the actual stakes.

I need to message [name] that they are not getting the promotion
they asked about. They've been pushing for it for two cycles.
They are a strong performer but not yet operating at the next level.

Tone: direct but warm. I respect them and want them to stay.
Length: short — this is the message before a 1:1, not a substitute for it.

Constraints:
- Be clear it's a "not yet," not a "maybe."
- Name one concrete thing they need to demonstrate, not a list.
- Do not say "I hear you" or "I appreciate."
- End with the time of the 1:1, not an apology.

Give me 2 versions with different opening sentences.

Then read it aloud. If you wouldn't say it that way to your friend, rewrite the line. Your voice is the thing the model can't fake — your job is to put it back in.

For nuance on framing growth conversations and giving structured feedback, the AI for Managers Playbook course goes deeper on the conversation itself, not just the writing.

A Voice Check You Can Run on Anything

Before you send any message a model helped you write, paste it back with one prompt:

Below is a message I'm about to send. Flag any phrase that sounds
like AI wrote it — clichés, corporate filler, hedging, or anything
a real manager wouldn't say in a 1:1. List them. Don't rewrite.

[paste message]

Common offenders: "I wanted to reach out," "circle back," "align on," "excited to share," "moving forward." Strike them. Replace with what you'd actually say. The model writes the scaffold; you put the human back in.

Write less. Mean more. Send it.