The AI Email Problem: Everyone Can Tell

You have received one. Maybe this week. An email that is perfectly organized, grammatically flawless, warm in a way that means nothing, and says about as much as a blank page. You read it, felt a small drop in your regard for the sender, and moved on without quite putting your finger on why.

Your clients may be having that experience with your emails. Not because AI writing is bad, but because most people use it the laziest possible way: type a one line request, copy the output, hit send. The tool did its job. The person skipped theirs. We use AI for drafting every day and think you should too. The difference is what happens between the draft and the send button.

The Tells

People cannot always name what tipped them off, but they notice. These are the patterns that give it away.

Over-hedging. Every claim wrapped in a cushion. “It may be worth considering that this could potentially help in some situations.” Nobody talks like this. You either think it will help or you do not.
Unnatural balance. Two sides given identical weight, even when one is obviously right. Real advice picks a side. That is what makes it advice.
Empty enthusiasm. “Great question!” “I would be thrilled to help!” Warmth that costs the sender nothing and tells the reader nothing.
Vocabulary nobody says out loud. Leverage, robust, seamless, streamline, delve, holistic. If you would feel ridiculous saying it to somebody’s face, it does not belong in your email.
The same shape every time. Friendly opener, three bullet points, a paragraph that begins with “Ultimately,” and a closing offer to help further. Once your clients have seen it twice, they see it forever.
Length with no weight. Four paragraphs that could have been one sentence. The word count went up. The information did not.

Any one of these alone is just a stylistic habit. Stacked, they read as a form letter, and a form letter is what your reader concludes you thought they were worth.

Why This Costs You Trust

There is a real cost here, and researchers have measured it. Colleen Kirk of the New York Institute of Technology and Julian Givi of West Virginia University published work in the Journal of Business Research in 2024 finding that when consumers believed an emotionally charged message came from AI rather than a person, they responded with what the researchers described as moral disgust, and reported being less likely to recommend the business and more likely to switch brands. The same research found factual messages drew far less negative reaction than emotional ones.

Read that split carefully. Using AI to draft a straightforward service update is low risk. Using it for the apology, the condolence, or the note about a mistake your team made is where you lose people. Those are the moments a client decides whether you are real people who care.

The second cost is quieter. In research presented at the ACM Conference on Human Factors in Computing Systems in 2025, Cornell University researchers studying 118 writers found people using an AI writing assistant drifted toward a single dominant style, to the point that participants described their own festivals, food, and cultural references through a borrowed lens. Push that into a business context and you get the thing everybody feels but nobody says: the more your team leans on the same tool with the same prompts, the more your company sounds like every other company. Your voice was a competitive advantage, and you gave it away for free.

The Fix Is Your Own Writing

Harrison’s framing for AI is that you should treat it like a junior staff member. That framing solves this problem cleanly. A new hire who has never read anything you have written will produce generic work. A new hire who has read fifty of your emails will not.

So give it your writing. Pull five or ten emails you actually sent and were happy with, and paste them in before you ask for anything. Tell the tool to match how you write: your sentence length, your bluntness, your habit of skipping pleasantries. Save that set somewhere reusable. Most AI tools offer some way to store standing instructions or reference material, though the names and mechanics change constantly, so check what your tool currently supports.

Then edit hard. This is the step people skip, and it is the whole job. Cut the opening line. Cut every hedge. Delete any sentence that could appear in any company’s email. Put your actual opinion back in. A good draft plus five honest minutes of editing beats anything the tool or a rushed human produces alone. Check the facts too, since AI systems will state something confidently and be wrong, a failure mode the National Institute of Standards and Technology calls confabulation, and one that is dangerous precisely because the tone stays so assured.

Keep the Things No Model Has

Here is what actually makes an email sound like you. A model has read an enormous amount of writing. It has not been to your client’s warehouse. It does not know their server room floods when it rains hard, that their office manager retires in March, or what you told them in the parking lot last Tuesday.

Specifics beat adjectives. “We finished ahead of schedule” is a claim. “We finished Thursday instead of Monday” is a fact your reader can check.
Shared history is free credibility. Reference the actual conversation. Name the actual problem. Nobody can fake that but you.
Your opinion is the value. Clients are not paying for a summary of options. They are paying to be told which one you would pick and why.
A short imperfect note beats a polished long one. Three honest sentences from a person who knows the situation will always outperform six smooth paragraphs from nobody in particular.

The Rule That Catches Almost Everything

If you would not say it out loud, do not send it.

Read the draft aloud before it goes. Not silently, actually aloud, at your desk. You will hear the fake enthusiasm immediately. You will trip over the corporate vocabulary. You will notice that a sentence takes eleven seconds to say and communicates nothing. Every one of those moments is a cut.

It takes ninety seconds and it is the highest return habit we know of for anyone using AI to write. If your team is adopting these tools broadly, build the habit in from the start, which is the same argument we make in our guide to preparing your team for AI without the hype.

One related note. The same fluency that makes an AI draft sound plausible is what makes AI-assisted phishing harder to spot than the clumsy scams everyone learned to recognize, a shift we covered in our piece on AI-powered cyberattacks getting smarter. Polished writing is no longer evidence a message is legitimate, which is one more reason your real messages should sound unmistakably like you.

The Bottom Line

Use AI to draft. It is genuinely good at getting words on a page, and staring at a blank screen wastes your morning. Just do not confuse the draft with the message. The draft is raw material. You know the client, the history, and what you actually think.

Feed it your own writing, edit like you mean it, keep your specifics, and read it out loud before you send. Do that and nobody will be able to tell, mostly because at that point it really is your writing. Skip it and everyone will, and they will draw a conclusion about how much your business was worth their time.

If your team is using AI daily and you want a sensible set of guidelines that protect both your data and your reputation, that is exactly the kind of thing we set up for businesses across Denton County. Contact us today.

Sources:

AI Policy Artificial Intelligence Small Business IT

Comments are closed