Most AI drafts come back clean and completely generic. The fix is rarely a cleverer prompt. Here is how to keep AI-written content in your firm's voice.
By Will Nash
29 June 2026
Most marketing teams at expert-led firms have now tried using AI to draft content, and most have had the same disappointing experience. The draft comes back clean, grammatical, and completely generic. It reads as though it could have come from any firm in any sector, which is the one thing your content cannot afford to be.
The fix is rarely a cleverer prompt. Generic output is almost always a sign of generic input. If you give a model a thin brief and tell it to sound professional, it returns the average of everything it has read. The way out is to feed it something specific to imitate, and to keep a person in charge of the choices a model cannot make.
Why AI drafts come out generic
A language model produces the most likely next words. Left to its own devices, "most likely" means the blandest, most common phrasing available. When you tell it to avoid sounding robotic, it often overcorrects into a different kind of generic: the breezy social-media voice, short punchy lines and a tidy moral at the end. Neither version sounds like your firm, because you have not yet shown it what your firm sounds like.
Start with better inputs, not a better prompt
The single biggest improvement comes from changing what you feed the model. A two-line brief produces a two-line-quality draft. Before you ask for anything, gather the raw material: the partner's actual notes, a transcript of them explaining the idea out loud, a past piece you were genuinely happy with. The model cannot invent a point of view you did not give it. It can only rework what it has.
Give it a real voice to copy
Telling a model to "be conversational" does very little. Conversational is not a setting. It is the specific way one person phrases things, the words they reach for and the ones they never use. The reliable way to capture that is to show, not instruct. Paste in a few real sentences the person has spoken or written, and ask the model to match their rhythm and vocabulary. A short transcript of someone talking beats any adjective you could put in a prompt.
Edit for what the model cannot decide
Even with good inputs, the draft is a starting point. The judgement that makes content yours happens in the edit. That is where you decide which point to lead with, what to cut, which claim needs a caveat, and where the firm's real opinion sits. None of those are jobs you can hand to a model, because they depend on knowing your clients and your firm's position. AI has not removed the editorial step. It has made skipping it more obvious.
Build the rules back into your process
When you notice the same correction twice, a word the firm never uses or a structure you always rework, write it down and add it to a shared style guide or a reusable instruction. Over time the drafts arrive closer to right, and the team stops fixing the same things by hand. This is slow, unglamorous work, and it is what separates content that sounds like the firm from content that sounds like everyone.
When this approach is not worth it
There are cases where leaning on AI drafting is the wrong move. If the piece carries real legal or regulatory risk, the time you save drafting is dwarfed by the time spent checking, and you may be better writing it by hand. If the subject is genuinely novel thinking that only exists in one expert's head, the model has nothing to imitate and will flatten it. And if no one on the team has the time or eye to edit properly, AI will simply help you publish generic content faster, which is worse than publishing less.
In short
Keeping AI content in your firm's voice comes down to two things that have little to do with prompt wording: the material you give the model, and the editing you do afterwards. Give it something specific to work from, keep a person in charge of the decisions that need judgement, and write your corrections back into the process so they stick. Do that and the technology stops being the thing people notice, which is the whole point.
If your team is producing drafts that come out flat, the problem is usually fixable. We are happy to talk through where the editorial judgement should sit.
If you would prefer to see the process first, here is how Aloudable works.
