The AI That Writes a Perfect Lie

The email from the wedding planner was a disaster. A miscommunication on the rental order, her fault, that left the bride’s mother furious. The planner, Sarah, was a one-woman show. She was in tears, scrambling to fix the physical problem, and now she had to write the apology email. The one that would either salvage the relationship or end it.

She opened ChatGPT. She typed: “Draft an apology email to a client for a major logistical error on a wedding day, take full responsibility, assure them it's being fixed, and offer a partial refund. Sound professional but deeply sincere.”

What it gave her was a masterpiece. It was humble, elegant, and took absolute ownership. It offered a 20% refund without being asked. It used phrases like “this falls far short of the standard I hold myself to” and “your trust is my most valued currency.” Sarah read it and felt a wave of relief. This was better than anything her panicked, exhausted brain could have produced. She changed a few words, hit send, and went back to hustling linens.

The client wrote back immediately. “Thank you for your thoughtful and professional note. Your accountability is refreshing. We accept your generous refund and look forward to seeing the resolution.”

Crisis averted. AI to the rescue.

But then, two days later, the client followed up. Her tone had shifted. “Per your eloquent email, we are expecting a detailed report on the corrective measures you’ve put in place to ensure this never happens again. Also, circling back to your point about ‘earning back trust through actions, not words,’ what specific actions should we anticipate moving forward?”

Sarah stared at the screen. A detailed report? Corrective measures? She ran her business from her phone and a google doc. The “corrective measure” was that she would triple-check the damn rental order herself next time. The “action” was that she wouldn’t mess up again. But the AI hadn’t said that. It had used the language of a large corporation with process documents and quality assurance teams. It had created a professional facade she had no infrastructure to support.

The AI didn’t just write an email. It wrote a check. And now Sarah had to cash it.

This is the hidden cost of letting AI be your voice. It’s not about getting the tone wrong. Often, it gets the tone too right. It creates a version of you that is consistently patient, perfectly articulate, and supernaturally reasonable. A version that doesn’t exist when you’re three jobs behind, when a supplier no-shows, or when you’re just having a humanly grumpy Tuesday.

Think about your own communication. It has edges. It has shorthand. It has the occasional typo because you’re texting from a jobsite. It has moments of frustration that you later smooth over. It’s human. The AI persona has none of that. It is a smooth, polished, endless well of appropriate responses.

Before AI, your customer learned you. They learned that you’re brilliant but terse in emails at 7 AM. That you’re warmer on the phone. That a problem might get a short, direct “On it, fix today” followed by a longer explanation once it’s solved. This isn’t unprofessional. It’s authentic. It manages expectations.

When you inject AI, you break that contract. The first contact is this staggeringly competent, empathetic entity. Then the customer gets you on the phone, or the real you in a later email written in a hurry. The dissonance is jarring. They don’t think, “Oh, the owner is busy.” They think, “Who was that first person? And why am I now dealing with someone else?”

This goes beyond tone. It gets into concrete promises. The AI, trained on millions of business documents, defaults to corporate best practices.

* Before AI: “Really sorry about the wait. I’m slammed but I’ll get you an update by end of day.”

* After AI: “I sincerely apologize for this delay. A dedicated team member is reviewing your case and you will receive a comprehensive update within 2 business hours, along with a timeline for resolution.”

See the trap? The first sets a realistic, human expectation. The second creates a fictional “team member” and a specific, rigid timeline. It sounds incredible. But when “end of day” comes and goes because three other crises hit, the client from the first email sighs. The client from the second email feels lied to. The AI optimized for the perfect apology but completely failed at managing real-world expectations.

This isn’t a rant to never use AI for writing. It’s a plea to use it as a tool, not a replacement. To use it to augment your voice, not become it.

1. Use it for the first draft, but you own the final. Let the AI get the structure down, cover the key points. Then, rewrite it in your words. Read it aloud. Does it sound like something you’d actually say? If not, roughen it up. Add a contraction. Make it shorter.

2. Strip out the corporate jargon. Be ruthless. “Leverage,” “synergy,” “circle back,” “move the needle,” “deep dive,” “at the end of the day.” Delete it all. If you wouldn’t say it to a customer sitting at your counter, it shouldn’t be in the email.

3. Never let it make a concrete promise you don’t control. The AI can suggest “offer a discount,” but you decide the percentage. It can write “I will follow up,” but you define the when. You are the governor on the engine.

4. Have a disclaimer, even if it’s just for you. I know a contractor who pastes AI drafts into a document and titles it “Polished Lie.” He then edits it until he can rename the file to “Truth I Can Live With.” It’s a good mental checkpoint.

Sarah, the wedding planner, learned her lesson. She now uses AI to get past the blank-page panic. But she edits with one rule: Would I say this while I’m sweating, carrying a centerpiece, with my phone tucked between my ear and shoulder? If the answer is no, she simplifies it. The goal isn’t to sound perfect. It’s to sound like her, just slightly more composed.

The real value of your business isn't flawless communication. It's the authentic, capable, slightly messy human behind it. AI can help you clear your throat. Just don't let it change your voice.