Your AI copy doesn't sound like you. Here's the real reason.

You've done this. You open the thing, you describe your business as well as you can in a paragraph, you ask for a week of posts. What comes back is fine. It's competent, it's grammatical, and it sounds like it could belong to any one of four hundred other businesses in your category.

So you rewrite it. And by the time you've finished rewriting it you've spent roughly the hour you were trying to save, except now you're also slightly annoyed.

Then you assume the problem was your prompt, so you write a longer one next time, and the same thing happens again.

I've started calling that AI amnesia, because it isn't a quality problem. The model is perfectly capable. It just doesn't know you, it has no way of coming to know you, and every conversation resets it to nothing.

What's actually missing is context, and nobody hands it over well

Now, there's a related idea I've written about separately, which is that every industry has its own hidden language that the people inside it know and nobody outside does. That piece is about why context matters. This one's about how you actually get it out of someone and keep it, because that turns out to be the hard part.

Here's what most marketers do. They get on a call and they ask what makes you different from your competitors. And what they get back is a brochure sentence. *We care about quality. We've been going twelve years. Our service is second to none.* Every business on earth says a version of that, which is exactly why it produces content that sounds like every business on earth.

What I ask for instead is stories. Wins, losses, the funny ones, the job that went sideways, the customer who came back three years later. Just tell me things that happened.

The funny thing is that a story about a delivery going wrong will teach me more about how your industry really works than an hour of direct questions ever could, because it's full of detail the person doesn't know is interesting. That's where the gold sits. Nobody hides it from you deliberately. They just don't think to mention it, because to them it's Tuesday.

An example, because that's abstract

Take pre-workout in Australia. If you compare the formulas across the shelf, most of them are nearly identical, and there's a reason for it. The regulator caps which ingredients are permitted and at what doses, so the actual room to differentiate on the panel is small. I have written up what that constraint does to your content in AI marketing for supplement brands.

So what happens is the variable moves. Brands play with the serving size and the scoop-to-water ratio, which is why you'll pick up a tub and read something like one scoop per two litres of water. And every single person who actually uses supplements knows that's not how anybody uses it. There's a gap between what the label has to say and how the product gets used in real life, and the gap exists because of how the category is regulated, not because anyone's being sneaky.

Now, everybody inside that industry knows this. Almost nobody outside it does.

If you hand that brand to a marketer who doesn't know, one of two things happens. Either they write something that runs straight into a compliance problem, or they write something so cautious it says nothing at all and doesn't sell a single tub.

That's one nuance, in one category. Your business has its own, and there are probably a dozen of them.

Where it has to live

Getting the context out is half the job. Holding onto it is the other half, and it's the half generic tools don't do.

What I build for a client is a portfolio. One structured document with the business in it, the industry, the small awkward details that don't fit anywhere else, and the stories. That document is what feeds the system, so the same context turns up on post forty as it did on post one.

Then there's a second layer, and this is the one that actually separates a system from a clever prompt. It also learns from the posts that have already gone out, and from what happened to them.

Because here's a thing that took me a while to accept. You can have a genuinely great story, one you love telling, and it goes nowhere. It's interesting and it doesn't convert. That's not a failure of the story, it's information, and you only get it by putting things out and watching.

Think of it like a tackle box. You turn up with fifteen or twenty different baits, and the point isn't that you own fifteen baits. The point is that after enough casts you know which one these particular fish take. A system that doesn't record which bait worked is just a very fast way of casting the same lure over and over.

When you reject a draft, the same thing applies. We talk about why, and that conversation goes into the file. Most tools throw that away, which is why they can produce a thousand words in six seconds and still be no closer to sounding like you in March than they were in January.

How long before it sounds like you

Look, how long is a ball of string. It depends entirely on what goes in.

But for someone who's engaged, who's actually looking at the output and giving real feedback rather than nodding it through, you're looking at roughly three weeks before the voice is genuinely recognisable. Around a month before it's producing consistent results you'd be happy to have out there under your name.

The caveat I'd give anyone is that being that person is itself a job. Understanding the system, feeding it properly, editing with intent — that's real work, and it's the work most people don't have room for. Which is usually the reason they came looking in the first place.

The part that's still broken

I'm not going to sit here and tell you it's flawless, because it isn't. Even with a well-built system there are days where the model does something strange for no reason I can identify, and a human has to sit down and fix it by hand.

AI has a long way to go yet. Anyone telling you otherwise is selling you something, or hasn't built enough with it to have been let down by it.

What I can tell you is the direction of travel. Cycle one is rough and you'll rewrite plenty. By cycle ten, if the system remembered everything that happened in between, it's carrying your voice and your market's response and a year of things you learned the hard way.

A system that doesn't remember is a treadmill. One that does becomes an asset, and it's worth something even on the day you stop paying for it.

That's how we built Alpinist from a generic product into something that actually looked worth buying. Not one clever prompt. A lot of specific detail, held onto.

Find out where the voice is breaking

If your content doesn't sound like you and you can't work out where it's going wrong, that's a fixable thing and it starts with someone reading it the way a stranger does.

Three hundred dollars for the audit, your site and your socials gone through properly, and a list of what's leaking in the order it's costing you. Credited to your first month if you carry on.

And if you want the bigger argument for why any of this is structured the way it is, that's here.