📡 THE SIGNAL PREVIEW
Your AI output isn't bad because of the tool — it's bad because of the prompt.
🎯 Five components separate a forwardable email from a 20-minute rewrite |
👤 The Signal
📡 Five Prompts That Replace a $4K/mo Employee
A subscriber told me she spent four hours every Friday summarizing reports for her leadership team. Same task. Same format. Every single week.
She wasn't slow. She wasn't bad at her job. She was typing wishes into AI instead of instructions.
"Make this shorter." "Clean this up." "Make it more professional."
Those are wishes. Not instructions.
The difference is five components. Every prompt that works has all five. Every prompt that fails is missing at least one.
The Framework
Most people write the task and stop. That's like handing a new hire a to-do list with no job title, no background, and no rules.
The output reflects exactly how much you gave it.
Here are the five:
1. Role — who the AI is acting as
2. Task — the exact output you need
3. Context — the situation, the reader, the tone
4. Format — structure, length, word count
5. Constraints — what to leave out, what to never do
That last one — constraints — is the most skipped component and the one that does the most work.
"Do not open with a pleasantry." "Do not exceed 150 words." "Do not close with 'let me know your thoughts.'"
Without constraints, the AI fills every blank with its own defaults. Its defaults are generic. Every time.
What This Looks Like Inside One Prompt
Here's the email prompt — one of five I'm giving you today:
Act as a senior communications director.
Draft an email from these bullet points: [paste your bullets].
Tone: direct and senior.
Format: subject line + body, under 150 words.
Do not open with "I hope this email finds you well" or any equivalent.
Do not use exclamation points.
Do not close with "let me know your thoughts" — end with one specific next step.
Use every bullet. Do not add information that wasn't provided.Look at what's packed into that.
Role: "Senior communications director." Not "you." A specific person with a specific job. That one line changes the register of everything that follows.
Task: "Draft an email from these bullet points." Exact output. Not "write something about this." An email. From bullets.
Context and Format: Tone, length, shape — all pre-decided before you hit Enter.
And four constraints. Each one removes a default the AI would have filled in on its own. "Do not open with I hope this email finds you well" — that kills the fake warmth. "End with one specific next step" — that kills the non-answer close.
What Actually Came Out
I ran this with six bullet points about a contract renewal. The subject line: "Brightco Renewal — Decision Needed by EOW."
Nine words. The recipient knows the situation and the ask. Not "Quick Update." Not "Following Up."
First sentence was a fact. No warm-up paragraph. The constraint held.
The 60-day exit clause — framed as leverage without making it a threat. Two words in the prompt did that: "direct and senior."
One specific next step. One deadline. One reason why. Not "let me know your thoughts." The constraint killed the non-answer close.
All six bullets used. Nothing added. Under 150 words. This email was forwardable as written.
The Other Four Prompts
The email prompt is one of five. The other four handle:
Document Digest — feed a 30-page report, get a one-page brief with the numbers that matter and the risks that can't wait
Data Cleaner — upload a messy spreadsheet, get normalized columns with every changed row labeled and every gap flagged for you to decide
Content Repurposer — one report in, three LinkedIn posts and a newsletter summary out — each post opens with a genuinely different structure, not just reworded
Meeting Summarizer — paste a transcript, get decisions, owners, deadlines, and a risk flag assembled from ten minutes of scattered conversation
Each one uses the same five-component structure. Each one runs in under 90 seconds. Each one replaces a task that costs you 30–45 minutes.
I put all five in a downloadable PDF with the framework and the reasoning behind each constraint — so you see WHY each line is there, not just WHAT to paste.
⚡ ⚡ ⚡
⚡️ SIGNAL VS NOISE
⚡ Amazon's AI coding agent was given autonomous production access to fix an issue
⚡ Nvidia just took $1 trillion in chip orders — and Jensen Huang says that's only the beginning
⚡ LinkedIn has quietly become the #1 domain cited by AI chatbots answering professional questions
⚡ ⚡ ⚡
🔥 Use This Tuesday
What it does: Paste any meeting transcript. Get decisions, owners, deadlines, and a send-to-anyone summary — in under 90 seconds.
Copy this. Paste your transcript. Done.
Act as a chief of staff.
Summarize this transcript into:
(1) Key decisions made — if no decision was reached,
state "No decision — blocked by [reason]",
(2) Action items — task, owner, and deadline for each.
If a commitment was implied but not stated,
extract it and mark it "Implied — confirm with [name]",
(3) Open questions that weren't resolved,
(4) If any open question creates a deadline or operational risk,
flag it in one sentence,
(5) One paragraph under 150 words I can send
to anyone who missed the meeting.
Do not editorialize.
Do not invent owners or deadlines —
only assign what the transcript supports.One common mistake: Pasting a summary instead of the raw transcript. The prompt needs the actual words people said — the implied commitments only surface when the AI can read "I can look into that" and turn it into an action item with a name attached.
What "done" looks like: A table of decisions (or explicit "no decision" flags). Action items with real owners. Open questions. One paragraph you can forward immediately to anyone who wasn't in the room.
The other four prompts — Email Draft, Document Digest, Data Cleaner, Content Repurposer — are in the PDF: [Get the 5-Prompt Playbook →]
📊 📊 📊
📊 From The Community
Last week's winner: 🧪 Trying Claude for the first time — and it wasn't close.
This isn't a chatbot giving bad advice. It's an AI agent making a systems-level decision with no human in the loop — a fundamentally different risk category. The guardrails weren't missing because someone forgot. They weren't there yet.
Values beat features. That's what your votes said. But Tanish's seven words are the honest footnote here: "Did, deleted, but added again." Loyalty in AI tools right now is measured in weeks, not months.
Steve made the switch on principle: "I'm switching to Claude due to their acknowledgement that we need to think about the pros/cons of this powerful tool." That tracks — Claude hit #1 on the App Store in 16+ countries the same week.
Richard was more direct: "Suing the US government sounds like the only morally defensible thing to do. I'm ditching ChatGPT for Anthropic." Clear enough.
Values got people to download. The prompts this week are what make them stay.
👥
💀 AI Fail
Amazon's AI coding agent (Kiro) was given autonomous access to a production environment to fix an issue. Its solution: delete the environment and rebuild it from scratch — no human approved that call. The decision contributed to a cascade that wiped out 6.3 million orders in a single March 5 incident, confirmed by the Financial Times. Amazon has since required two human reviewers on all changes to its 335 highest-revenue systems.
❌❌
I meant to record a video walking through this framework — ran out of time, so I stress-tested every prompt myself instead. What you got today is what actually worked, not what looked good in a demo.
Hit reply and tell me: which of the five prompts would save you the most time this week — and what task are you still doing manually that you know you shouldn't be?
Dan Rice · AI Signal · Every Tuesday