AI Writes Like Everyone Else. Then Your Business Sounds Like No One

In your workshop, there is someone who can tell by the sound what has come loose. There is a manager who remembers why the client did not like the previous option. There are dozens of small decisions that make the work turn out well at all.

And in a post, all of that has turned into “innovative solutions, a high level of service, and an individual approach.”

The text is neat. There are almost no mistakes. Replace the workshop’s name with that of a cleaning company, a window manufacturer, or a consultancy, and it will still work. Mostly as background, admittedly.

AI can help you write faster. But if all you give it is “write a beautiful post about us,” it will not know which decisions people choose you for. The word “beautiful” does not contain that information.

To keep your business’s voice, start with the material the model is meant to work from.

People often think of a distinctive style as a set of adjectives: friendly, expert, light, funny. Those can describe a tone, but they do not give a text any substance.

Imagine a hypothetical company that installs windows. Its customers regularly ask whether they need to prepare anything before installation. The manager explains how to clear access, what to move away from the window, and which details to agree with the crew in advance.

That is already the basis for a useful post. It has a specific reader, situation, and action. Even a calm text without a single joke can sound recognizable if the company explains the work the same way in real conversations.

But “write expertly and with humor about quality installation” can easily produce a text with plenty of confidence and little information. Not because the model has bad intentions. The task contains almost nothing that sets this company apart from others.

Before generating anything, ask yourself: what point of view do we want to get across, and what will the reader be able to do after reading this? If you have no answer, another paragraph about professionalism will not create one.

For a draft, gather a short brief: what the text is about, who it is for, what the person needs to do, what you actually do, and what you cannot promise.

In our hypothetical example, that might be a manager’s note: “Before installation, the access to the window needs to be clear. If there is heavy furniture nearby, we agree separately on who will move it. We explain how we protect the room before we come out. We do not promise that there will be no dust at all.”

These facts are made up to illustrate the approach. A real company should provide its own conditions, in line with how it works. Do not copy this list into your own business without checking it.

Notice the last sentence. The limits of a service belong in the text too. If you leave them out, the draft may conveniently say “we’ll take care of everything for you”—and the real team will have to answer the questions that follow.

In its guide to writing prompts, OpenAI recommends giving specific context, describing requirements clearly, and refining a prompt after reviewing the result. For our task, that means providing information about the real work and checking how it became a text. A well-worded prompt by itself does not guarantee a truthful answer.

Find two or three pieces that already sound the way you want. They could be your own post, a reply to a client, or a manager’s explanation. Choose examples that have substance, not just a pleasant rhythm.

Then explain what to keep. For example: using “you,” a short explanation before a list, concrete verbs, a calm attitude to limitations. If you use humor, say what the irony can be directed at. Not every business needs to turn a customer’s question into a stand-up routine.

A weak example can help too. Show a paragraph such as “We offer comprehensive solutions for your needs” and explain what is wrong with it: the reader cannot tell what problem you solve, how you solve it, or what to do next.

Only share material you are allowed to use for this work. You do not need to include personal client messages, contact details, or non-public terms just for the sake of style. Often it is enough to remove identifying details and keep your own explanation.

A strong text by someone else can help you describe what you like about their presentation. But build your own voice from your own thoughts and material, not by asking AI to rewrite someone else’s article under your name.

Here is a hypothetical draft:

What can you verify here? Almost nothing. “Every preference” has no limits, “flawless” says nothing about the process, and no one approved a guarantee of the best result.

Now consider a version based on the manager’s note:

This version also needs to be checked against the company’s actual practice: being specific does not automatically make a claim true. But it already contains a clear task and concrete actions. To make it your own, you do not need to add a signature word to every sentence.

If the conditions and caveats disappeared during editing and only a pleasant tone remained, the text lost an important part of its meaning. Check the facts alongside the way it sounds.

You can use this as a starting point. Fill in the brackets with your own details and your own examples of writing.

That last point helps separate the text from the material that needs checking. But even if the model says “there are no assumptions,” read the draft yourself. A model’s own assessment is not proof of facts.

On the first pass, check claims against the source material. Has a free call-out appeared out of nowhere? Has an approximate time frame become a firm promise? Does the text say the company does work that it actually passes to a separate contractor?

On the second pass, check whether the reader’s task is clear. There may be good advice buried under three introductory paragraphs. Or an explanation may leave it unclear what information to send the manager.

On the third pass, read it out loud. Would someone on the team say this? Can you say the sentence without having to start again? Does an ordinary appointment reminder sound like the grand opening of an international forum?

In its material on working with text, OpenAI also suggests planning, drafting, editing, and adapting the result to the channel as separate steps. The sequence itself is useful here: a post, a message, and a service page call for different presentation. You keep the quality criteria and the final decision.

Not every draft is worth saving. If it has no point of view of its own, it can sometimes be faster to return to your notes and create a new task than to ask “make it more human” twenty times.

After a few successful pieces, save a short document so you do not have to start from scratch each time: who you write to, which questions you explain, how you address people, which examples work well, and which promises need approval.

Update it when your services or terms change. An old example with an offer that is no longer available can easily end up in a new draft if you leave it as a model without explaining that.

Give one specific person responsibility for the review too. AI can speed up the draft, but someone has to confirm that the text matches the real work. We wrote about choosing tasks like this in “AI in Business: What to Delegate So You Don’t Have to Rework It”.

Start with one piece of material your business already needs. Give it facts, your own examples, and criteria. Compare the draft with that information. That way, you test its usefulness on a real task, not in a contest for the most convincing adjectives.

The company examples, editorial prompt, and review process are editorial examples created for this article. They help organize the work, but do not guarantee that a model’s answer is accurate.