Here's what I was doing wrong. And if you've used AI for your email marketing, you've probably done it too.

I'd open ChatGPT, type one sentence, and expect it to somehow know who I was, who I was writing to, and what I actually needed to say.

"Write an email to a client who hasn't responded."

That's it. That was the whole prompt.

I might as well have handed a stranger a blank recipe card and asked them make my favorite dish for dinner.

The AI did exactly what I told it to do. It guessed.

And when you ask a model to guess, it reaches for the only thing it has: the statistical average of every email ever written on the internet.

That's the problem. Originality is by definition an anomoly that AI automatically rejects as a statistical outlier.

And that's why generic prompts give you generic copy. Each. And. Every. Time.

No voice. No specifics. No you. It sounds like a robot. Not a person.

I kept blaming the output. I never once considered my input was the problem.

Here's the part that actually stung.

I wasn't just wasting time. I was sending real emails to real prospects, and they were falling flat in a very specific way.

They weren't bad, exactly. They were grammatically fine. Polite. Professional.

And completely forgettable.

An attorney client of mine described it perfectly once. He said his AI-written emails read like "a form letter wearing a nice suit."

Looked put together from a distance. Fell apart the second you got close.

The real cost wasn't the wasted hour writing them. It was what happened after I hit send.

Nothing. No replies. No questions. Just quiet.

And here's the thing I didn't want to admit at the time.

When an email gets no response, the easy move is to blame the reader. They're busy. They're not ready. The timing was off.

Sometimes that's true.

But if every single email reads the same, sounds the same, and asks for the same thing regardless of who's on the other end, the problem isn't timing.

It's that you never told the AI enough for it to write like it actually knew your reader.

I remember sitting there one afternoon, staring at a batch of five emails I'd generated back to back for five completely different prospects.

Different names. Different industries.

Same email. Word for word, almost. Same opening line, same middle, same closing ask.

If I couldn't tell them apart, why would my readers feel like any of them were written for them specifically?

That was the moment it clicked.

I wasn't getting five personalized emails. I was getting one generic email, copied five times, with the name swapped out.

The AI hadn't failed me. I'd asked it for the same thing five times and called it customization.

Once I figured out the real issue, the fix wasn't complicated.

I didn't need a better AI tool. I needed a better habit.

Now, every prompt I write includes four specific pieces of information. I think of it as PTCC:

• Persona

• Task

• Context

• Constraints

Persona. Before I ask the AI to do anything, I tell it who to be. Not "write like a professional." That's not enough. I tell it the actual role, the years of experience, and how that person talks. A real estate agent I work with uses something like this: "You are a residential agent with ten years in a competitive market. You talk to clients the way you'd talk to a friend, warm and direct, never salesy." That single sentence changes everything downstream.

Task. Then I name the one specific thing I need. Not "write an email." Something closer to: "Write a three-sentence follow-up to a lead who attended a webinar three days ago and hasn't replied. The only goal is a reply, not a booked call." One job. One deliverable. No guessing.

Context. This is the piece almost everyone skips, and it's the one that matters most. I tell the AI what's actually going on. What's already happened. What the reader probably feels right now. An estate planning attorney I know includes something like: "This prospect came to a free seminar two weeks ago and asked a question about blended families. They haven't booked a consultation." That's not filler. That's the difference between an email that could go to anyone and one that could only go to this person.

Constraints. Last, I set the boundaries. Word count. Tone. Words to avoid. How many calls to action, which is always one. A financial planner I know puts it this way: "Under 150 words. No mention of guaranteed returns. One reply-based question only." Without a constraint like that, the model will happily write something that sounds great and gets a compliance officer very upset.

Four pieces. Every prompt. That's the whole shift.

Let me show you the difference in practice.

Reading about the four pieces and seeing them stacked together are two different things.

Say you're a health and wellness coach. Someone signed up for your list two weeks ago after downloading a free meal-planning guide. They haven't bought anything.

A lazy prompt would be "write an email inviting them to join my program."

Here's what a full PTCC prompt looks like instead:

Here's where it gets even more useful.

That third piece, Context, is where you can build in something specific to your relationship with the reader. Not just what happened, but how much trust actually exists between you two right now.

Someone who just downloaded a free guide is in a very different place than someone who's bought from you twice.

If your prompt tells the AI "this person just found me, trusts me a little, and hasn't spent a dime," the email that comes back will ask for a lot less than one written for someone who already knows and likes you.

Skip that detail, and you'll get an email that asks a total stranger to book a call.

That kind of mismatch is the number one reason a well-written email still gets no response.

The writing wasn't the problem. The ask was too big for the relationship.

I won't pretend every email is perfect now. It isn't.

But the difference between a one-line prompt and a four-piece prompt is night and day.

The emails sound like they came from an actual person who knows something about the reader. They ask for something reasonable.

And people started writing back.

The AI never changed. I did.

If you've been blaming the tool the way I did, try this instead.

Next time you sit down to write an email with AI, don't type a sentence.

Give it a persona, a task, some context, and a few constraints. Watch what comes back.

And if you want help figuring out exactly what to send, when, and why, that's precisely what I built Connor the Conversation Builder for.

Connor asks the strategy questions before a single word gets written: what trust level is this reader at, what's the most natural next conversation, and what should this email actually ask for.

It's a $17 custom GPT, and it requires a ChatGPT account to use. And it works with any subscription level -- even the free one.

You don't need a smarter AI. You need a better conversation with the one you've already got.