AI Is Everywhere. Here’s How We Use It Without Sounding Like It.

You can tell. That is the whole problem, and it is the reason I have opinions about this.

You are scrolling, you hit a caption, and something in your brain goes hm before you can articulate why. Maybe it is the phrase “in today’s fast-paced digital landscape.” Maybe it is that every sentence is exactly the same length. Maybe it is that the post is about a plumbing company in Millsboro and it reads like a keynote.

Whatever the tell is, you clocked it in under a second, and now you trust that business slightly less than you did two seconds ago. That is a real cost, and most businesses paying it have no idea they are.

So: yes, we use AI. Every day, on client work. And a big part of my job is making sure you cannot tell. Here is exactly how that works, because I think the how is more useful than the philosophy.

First, what actually gives it away

I have read a truly upsetting volume of AI-written marketing copy at this point. The tells are consistent.

• Throat-clearing. Two sentences of preamble before the point. Real people start in the middle.
• Rhythmic sameness. Every sentence lands in the same 14-to-18-word pocket. Human writing lurches. Short sentence. Then a longer one that wanders a little before it gets where it is going.
• Vocabulary you would never say out loud. Leverage, seamless, elevate, robust, unlock, navigate the landscape. Nobody has ever said “we elevate your brand” in a truck.
• Enthusiasm with no object. “We are passionate about excellence.” About what, specifically?
• The triple. Three parallel items, always three, forever.
• No friction. Nothing is hard, nothing is a tradeoff, nobody has an opinion. Real businesses have opinions.

Every one of those is a default. Which is the key insight, honestly — sounding like AI is what happens when nobody overrides the defaults.

Our process, and why the order matters

Everybody says “a human reviews it.” That is not a process, that is a disclaimer. Here is what we actually do.

Step one: we know the client before we generate anything

Before a single post gets drafted for a new client, we build a voice profile. Not adjectives — evidence. How does the owner actually talk? What do they call their customers? What words do they refuse to use? What are they proud of that they would never brag about? What is the one thing they always say on the phone?

One of our clients says “we’ll take care of you” constantly. That phrase is worth more than any headline I could write. It goes in the profile and it shows up in the work.

Without this step, everything downstream is generic, because generic is the only thing available.

Step two: the draft is raw material, not a draft

I do not treat the first output as something to polish. I treat it as a pile of parts. Usually one sentence in it is genuinely good and the rest is scaffolding. I take the sentence and build around it.

This is the mental shift that matters most. If you think of it as “editing,” you will keep the structure and swap words, and the structure is the tell. If you think of it as “mining,” you will keep the one good idea and throw the shape away.

Step three: we break the rhythm on purpose

Read it out loud. If it sounds like a person talking, ship it. If it sounds like a press release, cut every third sentence in half and delete the first one entirely. The first sentence is almost always throat-clearing.

Fragments are fine. Starting with “and” is fine. One-word paragraphs are fine. Actually.

Step four: something real has to be in it

This is the rule that does the most work. Every piece of content has to contain at least one thing that could only be true of this business.

A name. A town. A specific job. A number. Something that happened on Tuesday. A photo taken on an actual site with actual mud in it. The machine cannot invent those, which means their presence is proof a human was involved and paying attention.

Generic content is not just boring, it is unfalsifiable. It could be about anybody. So it is about nobody.

Step five: somebody who knows the client reads it last

Not a proofreader. Somebody who would notice if the tone were wrong for that particular owner. Half the catches at this stage are not errors — they are “he would never say it like that.”

Where we do not use it at all

Some things are off limits and I want to be clear about them.

Responses to real people. Reviews, comments, DMs, complaints. If a customer took time to write something and gets a generated reply, you have insulted them efficiently. Everything with a human on the other end gets a human on this end.

Anything about a hard moment. Condolences, apologies, a business closing, a service failure. If you have ever seen a company post a generated statement after something went wrong, you know exactly why.

Photos of people or places that do not exist. We use real photography of real clients doing real work. A generated photo of a “happy customer” is a lie with a smile on it, and the AI-sheen on those images is getting easier to spot every month, not harder.

The honest bit

People sometimes ask whether it is dishonest to use AI and not announce it. I do not think so, for the same reason it is not dishonest that I used spell-check. What would be dishonest is publishing something under a client’s name that they do not mean, or that is not true.

And if a client asks us what was AI-drafted, we tell them. Straight answer, every time. That policy is not hard to keep, because we are not doing anything we would be embarrassed to describe.

Why any of this matters

The reason people hire a local business is that it is run by people they can find. The second your marketing sounds like everyone else’s, you have thrown away the only advantage a national competitor cannot buy from you.

AI made it much cheaper to produce content. It did not make it cheaper to produce content worth reading. If anything, the flood makes the human stuff stand out more — which is a pretty good deal for anyone willing to do the extra pass.

That extra pass is the entire product. It is why our social media work takes as long as it does and why it does not sound like a robot on a treadmill.

Come say hi. Ask for Shelby. I will absolutely tell you which of your competitors’ posts are generated.