There are three rules for using AI at work. Don’t deceive anyone with it, don’t waste people’s time with output you never read, and don’t hand off responsibility for work that carries your name.
At TJ Digital we use AI in every workflow across roughly 42 clients. That’s what lets us deliver about four times the work at standard agency rates. We tell clients so openly, which means these three rules get tested here every day.
AI is here to stay, and I think that’s a good thing. Plenty of people would disagree with me on that. What we’re all still working out is when it’s okay to reach for AI and how to use it without annoying everyone around you.
I still get just as irritated by the misuse of these tools as anyone else does.
Is It Ever Not Okay to Use AI at Work?
Yes. Using AI to make it look like something happened that did not happen is the one line nobody should cross.
That covers a narrower set of things than people assume. Drafting an email with AI is fine. Generating a photo of a meeting that never happened crosses the line.
A few examples of what crossing that line looks like:
- Generating a customer quote or testimonial nobody actually said
- Producing an image or video of an event that never happened
- Reporting analysis you told AI to invent rather than run
- Sending an AI reply that implies you personally reviewed something you didn’t
Everything else lives in a gray area, and the gray area is where most of the real questions are.
Do You Have to Disclose Every Time You Use AI?
No. AI is becoming common enough in daily work that disclosing every use would get tedious fast, and nobody wants a footnote on every email.
The test I use is whether the recipient would feel misled if they found out. If knowing a machine drafted it would change how they read it, say so. If it wouldn’t, save everyone the disclaimer.
| Situation | Would I disclose it? |
| AI fixed grammar or tightened a paragraph I wrote | No |
| AI drafted an internal email that I then rewrote | No |
| AI produced most of a client-facing report | Yes |
| AI generated an image used as if it were a real photo | Yes |
| AI recommended a technical fix I can’t personally evaluate | Yes |
| A policy, contract, or law requires disclosure | Yes |
What decides each answer is how much of your own judgment stands behind the output. The further the work travels without you, the more the other person deserves to know where it came from.
What Is AI Slop at Work?
AI slop is output that looks like finished work but doesn’t move the task forward. You’re halfway through an email or a post before you realize what you’re reading, and by then the time is already gone.
This is currently the most common frustration people have with AI, and the numbers back that up. Researchers at BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 US full-time workers and found that around 40% had received this kind of output in the past month.
Each incident took an average of one hour and 56 minutes to sort out. The workslop research puts the total drag at roughly $186 per worker per month.
The trust damage lands harder than the time cost. In that same survey, 42% of the people who received it rated the sender as less trustworthy afterward.
How Do You Keep From Sending AI Slop?
Read the whole thing yourself before you send it. That one habit prevents almost all of it.
What bothers me is that I can tell when you didn’t. It’s worst with long emails, long Slack messages, and long reports. I’m partway through one of those when it hits me that you’re asking me to spend more time reading this than you spent writing it.
That’s the actual insult. AI can absolutely help us communicate better and faster, and I use it that way constantly. Have enough respect for the person on the other end to make sure what you send is worth their time.
Before anything goes out:
- Check that the names, numbers, and facts in it are right
- Cut it down, because AI almost always writes long
- Confirm it contains a decision, an answer, or a next step
- Delete any paragraph you’d skim past yourself
That last one catches most of it. If you wouldn’t read it, don’t send it.
Who Is Responsible When AI Gets Something Wrong?
You are. Even when the work came out of a model, you’re the one responsible for checking it and confirming you agree with it.
I use AI in essentially all of my work now, and I encourage my team to do the same. None of that transfers the accountability anywhere. The output has your name on it the moment you pass it along.
This matters more as tools get more autonomous. We’re moving toward AI agents doing real work across multi-step tasks, and the temptation to stop reading the output grows with every capability they add.
What If You Can’t Verify the AI’s Answer?
Say so, and pass it along with a disclaimer attached. Sometimes you’re not in a position to take responsibility for what the AI produced, and that’s fine. Take responsibility for that instead.
Suppose you’re not especially technical and the AI has recommended a technical solution. It’s fine to hand that solution to whoever needs it. Attach the note that it came from AI and that you can’t personally vouch for it.
The rule underneath all of this is short. If you can’t personally defend the output, you either fix it or you disclose the source.
How Do You Know If You’re Using AI Responsibly?
Run three questions against the work before it leaves your hands. Each rule has one question attached to it.
| Rule | The question to ask | What breaking it looks like |
| Don’t deceive | Does this imply something happened that didn’t? | Invented quotes, fake events, fabricated analysis |
| Don’t waste time | Did I read this, and is it worth their time? | Long, padded output nobody proofread |
| Don’t abdicate | Can I defend every claim in here? | Passing along answers you can’t evaluate |
None of these are hard. Most people skip them anyway.
Should You Tell Clients That You Use AI?
Yes. At TJ Digital we’re open about using AI in every workflow, because the efficiency gain belongs to the client.
Hiding it creates a problem you can’t solve later. Once a client finds out on their own, every previous deliverable becomes suspect. Being upfront also forces a healthier standard internally, since work you’re willing to attribute is work you actually reviewed.
That’s the same principle behind how we build a content machine for clients. AI handles volume and the human handles judgment, and everybody knows which is which.
Is It Cheating to Use AI to Write an Email?
No. Using AI to draft an email is about the same as using a template or a grammar checker.
The problem starts when you send the draft without reading it. The voice, the judgment, and the final call still have to be yours, and the fastest way to prove they were is to edit the thing before it goes out.
What Other Rules Should Apply to AI at Work?
Those are my three commandments, and I’d bet there are more. I left off the obvious ones, like don’t hack my infrastructure and don’t build a bioweapon that ends humanity, since I assume we agree on those. If you think I’m missing one that applies to day-to-day use, I want to hear it.
Our team applies these same rules to every piece of client work we produce. Reach out through our contact page for an audit of your website and marketing, and I’ll walk you through the biggest opportunities to get more leads.