Do not ban it, and do not pretend it is not happening. Set four rules. The request comes first. The person who sends it owns it. Every AI draft gets checked against the request. Facts get a source or get cut. Then make the check automatic so it actually happens.
It is already happening
Whether or not you have a policy, your team is using AI. Emails, reports, posts, summaries, code. Some of it is great. Some of it is smooth and wrong. You find out which when a client replies.
Banning it does not work. People just hide it. Reading everything does not work. You do not have the time. What works is a short set of rules and a way to check that does not depend on you.
Rule 1: The request comes first
Before anyone opens an AI tool, they should be able to say what the work must contain. Who is it for? What must it include? What must it not say? This is a brief, and it is the same brief you would write for a person. Here is what goes in it.
A clear request makes the AI output better. It also gives you something to check the output against. Without it, you are judging drafts on how they sound.
Rule 2: The sender owns it
Whoever sends the work is responsible for it. Not the tool. "The AI wrote that part" is not an answer a client will accept, so it is not an answer your team should offer.
Say this plainly and kindly. It is not a threat. It is clarity about where the job ends.
Rule 3: Every AI draft gets checked against the request
Not read. Checked. Against the list from rule 1. Did it answer every part? Did it add anything that was not asked for? Did it skip the hard part and fill the gap with nice sentences?
This takes five minutes with the AI work checker. Paste the request and the draft. It shows the facts to verify, the details that look invented, the parts of the request that were skipped, and the filler.
Rule 4: Facts get a source or get cut
Every number, date, name, and quote in an AI draft needs a source the sender can point to. If they cannot, it goes. AI tools invent statistics that sound real. This rule is the one that prevents the embarrassing email.
Write the rules down
One page. Share it. Here is a starting point you can adapt: team AI use policy template. Keep it short enough that people actually read it.
What good looks like after a month
- Drafts arrive with sources on the numbers.
- "The AI said" disappears from conversations.
- Fewer things come back from clients.
- People use AI more, not less, because they trust the output more.
That last one surprises managers. Good checking does not slow AI use down. It makes it safe to speed up.
Where M-Tracker fits
M-Tracker is rule 3 built in. Each task holds the request. When the work is marked done, M-Tracker compares it with the request and tells the person what matches, what is missing, and what to change, in one plain paragraph. It works the same whether a person or an AI tool wrote the draft. It does not try to detect AI. It checks whether the work is right, which is the thing that matters.
Make the check automatic
Rules that depend on people remembering them fade in a month. The check in rule 3 needs to happen without anyone deciding to do it.
That means the check lives where the work is marked done. When someone finishes a task, the check runs, shows them what is missing, and they fix it before it moves on. Nobody has to remember. Nobody has to be the AI police.
Try M-Tracker free14-day trial. No credit card. No employee surveillance.
Work you can trust, without redoing it.
Your team gets feedback before work reaches you, or your client.
Copyright © 2020 – 2026 M-Tracker. All Rights Reserved. Privacy Policy