How to make AI writing sound like you

The short answer

To make AI writing sound like you, hand the tool the decisions you make without thinking when you write: who you are talking to, what you believe that competitors don't say, how long your sentences run, what you would never say, and how you open and close a piece. Write those down once, supply them every time, then edit the draft as you would a colleague's.

Most people work the other way round. They give a generic prompt, get a generic draft, and spend longer fixing it than writing it would have taken. That fixing is real work, and it has to be repeated on every piece because nothing about the next prompt has changed.

Why does AI writing sound like everyone else?

A language model writes by predicting what usually comes next. Ask it for a LinkedIn post about customer service with no other context and it will produce something close to the middle of every customer service post it has seen. That middle is fluent, polite and forgettable, because it was assembled from everyone and belongs to no one.

The familiar habits follow from that: every choice is the safe one at that step, and a piece built entirely from safe choices reads like a template.

Your own writing sounds like you because at dozens of small points you made a less safe choice. You used a word your industry uses and outsiders don't. You started with the complaint a customer made last week, and left a sentence short because padding an obvious point would have been slightly insulting to the reader. A model has no reason to make any of those choices unless you give it the reasons.

What does "sounding like you" actually consist of?

Ask most people to describe their voice and they reach for adjectives. Put those in a prompt and very little changes, because the model already believes it is being friendly and professional.

A voice is easier to pass on as a set of decisions, and they fall into three groups.

Behaviour. What you do on the page. Do you lead with an opinion or with the evidence? Do you admit what you don't know? Is there humour, and is it dry or warm? Do you name the approach you disagree with, or leave it unmentioned?

Audience. Who is reading and what they need from you. A practice manager and a sole trader may both buy the same service. One wants the cost case up front; the other wants to know nothing will go wrong.

Structure. How you build a piece. Some writers open with a story and some with the answer. Some break every few paragraphs with a subhead, others write in long unbroken runs. Plenty always end on a next step and never on a summary.

Here is the difference in practice. Take an invented business: a two-person joinery workshop that builds fitted furniture. A bare prompt asking why people should choose bespoke furniture produces something like this:

"Bespoke furniture offers a unique blend of craftsmanship and functionality, tailored to your individual needs and style."

Now give the tool three decisions. The owners talk like tradespeople and avoid sales language. Their readers are homeowners who have already tried flat-pack shelving in an alcove and watched it fail. They always open with a measuring problem. The same request comes back closer to this:

"The last alcove we fitted was 14mm out of square from floor to ceiling. A catalogue unit would have left a gap at the top you could post a letter through."

The second version sounds like somebody in particular.

Why fixing it afterwards keeps failing

The usual advice is to humanise the draft: vary sentence length, cut stock phrases, add a personal touch. Each helps a little. Together they produce something smoother and still anonymous, because the editing works on the surface while the generic choices sit underneath, in what the piece argues and in what order.

If the draft made the obvious point, rewording it won't make it argue yours. If it was written for nobody in particular, swapping words won't aim it at your customer. You end up rewriting the substance by hand, which is the work you hoped to avoid.

Edits also don't carry over. Every new draft starts from the same average, so the same corrections have to be made again, and in a busy week they won't be.

Who are you writing for?

Voice stays fairly constant. Tone moves with the reader. A good writer is recognisably the same person in a proposal and in a social post, but adjusts how much to explain, how quickly to reach the point and how much reassurance to offer.

This is where most AI prompts are thinnest. "Write for small business owners" covers millions of people with opposite needs. A useful description says what this reader is trying to get done, what worries them, what they would count as a win, and how they prefer to take in information. Some readers want the headline first. Others won't believe the headline until they've seen the proof. A cautious buyer wants reassurance that the risk is small, while an ambitious one wants to see what becomes possible.

Write that down for each audience you serve, and keep the descriptions separate. Merge a sceptical buyer and an enthusiastic one into a single paragraph and you have described a reader who doesn't exist. The model will write for the average of the two, and neither will feel addressed.

Structure and channel are part of the voice

The same idea written as a LinkedIn post, an email and a web page should share a voice and differ in nearly everything else: length, opening, the amount of context, the call to action.

AI tools default to one shape for everything, usually an introduction, a handful of points and a summary. Telling the tool which structure to use moves the result a long way, and copywriting frameworks are a quick way to do it. PAS (problem, agitate, solution) suits a reader who already feels the problem. When someone needs to picture the change before they care about the method, BAB (before, after, bridge) works better. A buyer comparing suppliers line by line is usually best served by FAB, which runs from features to advantages to benefits.

Choose one on purpose and the piece stops reading like a template, because its shape now belongs to its reader.

Write in layers, then edit

Put those pieces together and you have an order of work. Start with the business: what you do and for whom. Add the brand: the personality, values and behaviours that make up your voice. Then the audience for this particular piece, the framework that suits them, and the channel it is going to. Each layer narrows the choices the model makes, until the safe average stops being its most likely answer.

You can run that by hand with a well-kept document and some discipline. BraVo runs the same order of work for you. Each piece passes through five layers in sequence: your business, your brand (an archetype blend from which the voice and tone are derived), one of up to three audiences, a copywriting framework such as PAS or BAB, and the channel it is going to. Everything it produces is editable, so you finish the draft as you would any other. The difference is the starting point.

Whichever route you take, try one exercise this week. Find the best thing you have written in the last year and list five decisions it makes that a generic draft would not. That list is the start of your voice, and the first thing any tool should see.