Almost every company says it “uses AI” now. Ask how the team is actually supposed to use it, though, and you’ll usually get a shrug. Nobody’s set any rules, so people paste things into chat windows and hope it works out.
That’s why this post caught my eye. The author leads a team and got tired of that same gap, so they wrote a memo for their team and made it a non-negotiable instruction. To me it reads like the AI onboarding doc most workplaces never wrote. I’ve laid it out as nine steps you can use with your own team, with the reason behind each one.
🔒 Step 1: Protect the data before you paste anything
The expert’s first rule kicks in before you even open a chat. Ask yourself one question: “Would I be fine seeing this in the company-wide channel, with my name on it?”
- If the answer is no, swap names for roles. “Sarah at Acme” becomes “the client.”
- Never click thumbs up or thumbs down on a work chat.
- Never click “Share” on a work chat.
Why it matters: The author warns that clicking the thumbs button sends that chat off to train the model. One small click can hand over client details you never meant to share. Swapping names for roles takes about five seconds and removes most of that risk.
🔗 Step 2: Open every link
Once your input is safe, check the output. The original poster is blunt here: check every number, every name and every source.
You shouldn’t be the person who shows fake data to the team and to clients just because you didn’t double-check the source.
Why it matters: AI tools can make up a statistic that sounds completely real. If it ends up in a client deck, the AI won’t take the blame. You will.
🧠 Step 3: Use your brain first, then prompt
This one surprised me the most, and I think it’s the smartest rule in the memo. Before you open any AI tool, write down three options of your own. Only after that do you bring in the AI, with a prompt like this:
“What did I miss? Compare all of them for our situation.”
One more detail: never end a prompt with “right?” or “maybe?”
Why it matters: When the AI starts from your ideas, it’s sharpening your thinking instead of replacing it. Endings like “right?” invite the model to just agree with you, and an AI that agrees with everything won’t catch your mistakes.
✍️ Step 4: Remember that Claude won’t be accountable, you will
From here, the memo moves from how you use AI to who owns the result. The rules are simple:
- If you send it, you wrote it.
- If you can’t explain a sentence yourself, it’s not worth sharing.
- If it’s a raw AI idea, label it: “Claude suggested this, worth exploring?”
Why it matters: The author says unedited AI output turns your work into their work, and that they’re not your proofreader. Any manager who has had to rewrite an AI-generated report will know that feeling.
🏋️ Step 5: Go to the brain gym
Owning your work also means keeping your skills sharp. This innovator suggests a small daily routine:
- Write one first draft yourself.
- Do one calculation by hand.
- Make one decision before asking.
Why it matters: We all did this before ChatGPT arrived three years ago. The author’s warning is clear: don’t trade your intelligence for a bit of short-term speed. Skills you stop using fade, and it happens faster than you’d think.
🔌 Step 6: Don’t connect everything
Next comes a risk a lot of people miss. A connector can read your whole inbox, your whole Drive and your whole Slack, and you’ll probably never use most of that access.
- Connect only what’s worth it, and use read-only access when you can.
- Never plug a work account into a personal AI account.
- Review what’s connected once a month.
Why it matters: Every connection you add is another way for data to leak. If you’re unsure about one, the memo’s advice is to ask the manager first.
👋 Step 7: Treat Claude like someone who joined yesterday
This is my favorite way of putting it. Claude doesn’t know your clients, your jargon or what happened in Tuesday’s meeting, so tell it.
- Send a screenshot instead of retyping the numbers.
- Tell it to look around first: “Before you start, open every email, doc, and tab that could be relevant.”
Why it matters: Most bad AI output comes from missing context, not a weak model. If Claude starts to seem dumb halfway through a conversation, the author’s fix is to open a new chat every time. Long threads collect clutter that drags the quality down.
✂️ Step 8: Keep it short
Better input leads to the next point, which is shorter output. You write something once, and your team reads it ten times. If a 2-line prompt produced a 2-page doc, just send the 2 lines.
- Don’t use AI for everything. You can handle small tasks faster than AI can.
- Strip out tired AI patterns like “It’s not X, it’s Y,” “delve,” “seamless,” a pile of em dashes and pep-talk endings.
Why it matters: People spot those patterns right away, and they stop trusting the writing once they do.
🎯 Step 9: Ask once, ask right
The last rule pulls the rest together. Every prompt should include three things:
- The goal
- The context
- What “done” looks like
The expert also suggests you delete “think step by step.” Modern models already reason on their own.
Why it matters: One clear prompt beats five rounds of “no, not like that.” You save time, and so does everyone who reads the result.
💡 How to roll this out with your team
If you lead a team, you could copy this framework almost word for word. Here’s how I’d use it:
- Share it as a one-pager: Pin the nine rules in your team channel so new hires see them on day one.
- Start with Steps 1 and 6: Data protection and connectors carry the biggest risk, so tackle them first.
- Model the labeling habit: When you share an AI idea yourself, tag it “Claude suggested this.” Your team will pick it up.
- Run a monthly check: Combine the connector review with a quick chat about what’s working.
Companies are rolling out AI faster than they’re writing rules for it. A short memo like this protects your data and your reputation, and it keeps your team’s thinking sharp.
The full LinkedIn post has the original wording of every rule, and it’s worth reading in full.