3 Signs Your Employee Used AI to Write That Email - Digital Compliance Academy

I received an email last week from a supplier that started:

“I hope this email finds you well in these transformative times for the digital landscape. I am writing to unlock potential synergies…”

I stopped reading. I knew immediately it wasn’t written by a human. It was written by a lazy prompter using ChatGPT-3.5 defaults.

It felt sterile. It felt fake. And it made me trust the sender less.

As a manager, you need to spot this in your team’s work. Not to play “Gotcha!” or to punish them, but to maintain quality standards. Unedited AI copy damages your brand voice.

Here are the 3 huge giveaways that tell you someone hit Ctrl+C, Ctrl+V.

1. The “Tapestry” of Words (The Vocabulary Tell)

AI models—especially older ones like GPT-3.5 and even base GPT-4—have “favourite words.” These are words that are statistically safe, but humanly weird in a business context.

If you see these words popping up in your team’s work, be suspicious:

• “Delve”: (The #1 sign. Humans rarely say “Let’s delve into this.”)
• “Tapestry”: (“A rich tapestry of culture…”)
• “Landscape”: (“In the ever-evolving landscape…”)
• “Transformative” / “Game-changing”: (Hyperbole is the AI default).
• “Unlock” / “Unleash”: (Very common in AI marketing copy).
• “Foster”: (“To foster a culture of…”)

Unless your employee is a Victorian poet or a Silicon Valley caricature, they probably don’t talk like this.

2. The Structure is Too Perfect (The Rhythm Tell)

AI writes like it was taught to write in school.

• Introduction (re-stating the question).
• Point 1.
• Point 2.
• Point 3.
• Conclusion (“In conclusion…”).

It has perfect symmetry. The paragraphs are often exactly the same length.

Humans are messy. We use sentence fragments. We use one-line paragraphs for emphasis. We sometimes forget the conclusion. If an email looks visually perfect but feels “soul-less,” it’s AI. It lacks the “spiky” rhythm of human thought.

3. American Spelling (The Lazy Tell)

This is the big one for us in the UK. Most LLMs are trained on US data. Their default setting is American English.

If you see:

• “Honor” instead of Honour.
• “Analyze” instead of Analyse.
• “Program” instead of Programme (unless talking about code).
• “Dates written as Month/Day”.

…then your employee didn’t even bother to read the output before hitting send. This is the cardinal sin.

The Fix: Don’t Ban It, Raise the Bar

If you catch an employee doing this, your instinct might be to crack down. “No AI allowed!”

Wrong move. You want them using AI—it makes them faster. You just want them to use it well.

The Conversation to Have: “I noticed this report sounds a bit generic. I don’t mind if you use AI to draft the structure or generate ideas—in fact, I encourage it. But I mind if you send me unedited waffle. It needs to sound like YOU. Please edit the output next time.”

The ‘Human Polish’ Rule: Teach your team to spend 5 minutes “humanising” the text:

1. Deleting the intro preamble.
2. Changing “Delve” to “Look at”.
3. Injecting one personal opinion or anecdote.

Using AI isn’t cheating. Using AI badly is incompetence.