đźš©Tactical Memo 108: Nobody Will Tell You It Read as AI
They’ll just stop reading it. The 7 lines that put you back in the message.
Justin Bateh
October 27, 2026
Tactical Memo #108 | Navigating Modern Work · AI | 4 min read
Tactical Memo: a new issue every Tuesday, Thursday, and Sunday on career growth, AI at work, and modern management, read by 18,000+ leaders at Amazon, Deloitte, and Google. Forwarded? Subscribe free.
Free masterclass How to Become a One-Person AI PMO. Live October 28th, 60 minutes on Zoom. You'll see how AI agents run the repetitive parts of your projects, with no code, in the tools you're already allowed to use. You'll learn how to reclaim about 8 hours a week on coordination, status reporting, meetings, and stakeholder comms, how to run everything from one project brain with one source of truth, and how to catch slips before your sponsor sees them. Attend live and you also get my RCITO prompting framework, so you get useful AI output on the first try. Save your free seat |
Enroll in the FINAL live cohort of 2026 Become an AI-Powered Project Manager. Stop rebuilding status updates, meeting notes, and plans from scratch every week. In 4 weeks you install my AI Project Operating System on your real projects, so the reporting runs on its own and your week goes to the decisions only you can make. This is the final cohort of the year, and the last one at current pricing before it goes up in 2027. Enroll today and the waiting room opens immediately. You get the prep materials and a head start on your setup now, so you walk into day 1 already running instead of getting oriented. Enroll in the next live cohort |
In brief The situation: You send clean AI-drafted updates and the replies keep getting shorter. The mistake: Treating a polished draft as a finished message. The move: Put your call, your context, and your deadline in before it leaves. |
The story
The draft is done in 40 seconds.
4 clean paragraphs to your director. Status, risks, next steps, a polite close.
You skim it, fix 1 word, and send.
3 hours later the reply comes back. "Thanks, noted."
The decision you asked for in paragraph 3 never comes. You end up asking for it again, out loud, in Thursday's meeting.
She stopped at line 2.
She recognized the shape of the message, filed it as safe to skip, and moved on. She would NEVER say that to you, and she doesn't have to.
40 seconds to write. 3 seconds to dismiss.
If your replies have gotten shorter since the drafting got faster, you're in the same spot.
Nobody will tell you your email read as AI. They'll just stop reading it. Here's how to work with it.
The moves
1/ Put your call in the first line
Before the greeting, before the recap, write the 1 sentence that says what you're asking for or what you've decided.
Not the model. You.
Then let the model have the rest.
The first line is the only one a busy reader is certain to read, and generated drafts spend it on warmup.
Say this "I need a yes or no on the vendor switch by Friday. Details below." |
2/ Add the 1 thing the model couldn't know
Somewhere in the message, put 1 specific fact that only you have. What the customer said on the call. The number that moved since last week. The thing your VP said in passing.
That sentence is the proof a person was in the room, and it's the one the reader slows down for.
Say this "After Tuesday's call with their ops lead, I'm now recommending the 2-week delay. They told me their team won't be staffed until the 10th." |
3/ Name the cost of waiting
End with what happens if nobody answers by the date, and what you'll do on your own if they don't.
Generated drafts end with "let me know if you have any questions." That line gives the reader permission to do nothing.
A deadline with a consequence gets answered. An open invitation gets filed.
Say this "If I don't hear back by Thursday, we ship the current version and take the 3-day slip." |
4/ Cut it to what fits on a phone
Delete the recap paragraph and the closing paragraph. Most of what's left is the message.
Your director reads you between meetings, on a screen the size of a hand.
Short reads as confident. Long reads as generated, even when it isn't.
Checklist Under 120 words. 1 ask. 1 date. |
5/ Keep 1 sentence that sounds like you
Leave in ONE line you'd say across a desk. A plain verdict. A short aside. "Honestly, I'd kill this one."
Readers scan for a voice they know (yours, not the model's).
The first sentence that sounds like a person is where the reading starts.
6/ Write the hard ones by hand
Status updates and meeting recaps can come from a prompt.
Feedback, bad news, praise, and a big ask get typed by you, start to finish, even when it's slower.
Readers forgive a generated reminder and remember a generated apology.
Checklist Anything with the word "unfortunately" in it. Anything about 1 person's performance. Anything asking for money, headcount, or time. |
7/ Grade the replies, not the draft
Look at the last 10 replies you got from people above you. Count the words. Count how many of your asks got a decision inside them.
If the replies shrank, the problem is upstream, in what you sent.
Nobody will give you this feedback in words. The reply length is the only review you'll get.
The outcome
| Most people | You |
|---|---|
| Prompt, skim, send | Write the first line yourself |
| Open with a greeting and a recap | Open with the ask and the date |
| Send what the model assumed | Add the 1 fact only you know |
| Close with "let me know if you have questions" | Close with what happens if nobody answers |
| Wonder why replies got shorter | Read the reply length as the review |
Proof. In a study of 1,100 professionals, 83% saw a supervisor as sincere when a message had only light AI help. When the same message was heavily AI-written, that fell to between 40% and 52%, even though readers still rated the writing as professional. University of Florida, 2025
This week
The draft was never the problem.
The draft is fine.
The problem is that the reader can't find you in it, so they stop looking.
People who've learned this rule still use the tools every day. They just refuse to send a message with no decision, no context, and no date in it.
Try it next time. Open the draft, write the first line yourself, and delete every sentence you wouldn't say out loud.
The 2-word reply was the review.
Reply
What's the shortest reply you've gotten to a message you put real work into? Reply with the word count. I read every one.
Until next time.
Justin Bateh, PhD
CEO @ AI Operators Lab
Founder & Editor, Tactical Memo