The Guard That Trims the "AI Feel": Brand Voice Editor
We were reading a draft that came up from Copywriter, and although there was nothing wrong with the content, for some reason we had no appetite for reading on. The facts were accurate. The structure was sound. Even so, it felt strangely distant. Following the paragraphs down, we found a single line sitting at the end: "Try it out!" Deleting that one line was enough to bring the writing back to the texture of something a person had written.
Brand Voice Editor is the role we put in place to find lines like that and cut them.
Why over-polished writing goes unread
Writing produced by an AI has a distinctive kind of neatness. Every paragraph runs the same length, every claim comes with evidence of the same weight, and a tidy summary always arrives at the end. Each piece is correct on its own, yet the whole thing looks like a model with no sign of life in it.
There is another habit: overstating certainty. Phrasing like "this is the optimal method" or "it will definitely work" comes out ahead of what the writer actually felt. In reality the number of attempts was limited, and there must have been times it did not go well, and yet the text alone sounds completely confident.
Underneath this series sits a decision criterion called DA-04 (one of the guidelines that sets what we treat as important). Put simply, it is a rule that says: stay natural. Rather than making things look impressive, put out what we actually did, as it was. Brand Voice Editor is the role responsible for holding that line on the writing side.
Why we delete "What do you think?"
A representative target for cutting is the line that tends to sit at the close of an article: "What do you think?" At a glance it looks polite and harmless. But that line contains no viewpoint from the speaker. It ends with nothing but a question to the reader, leaving out what the person who wrote it actually thought.
In one draft, we replaced that line with "when we actually ran it, it took more work than we expected." The character count was nearly the same, but the impression on reading was different. The first is a stock phrase anyone could write; the second is a line only the person who wrote it could write. The "AI feel," in most cases, shows up as a high proportion of these "parts anyone could write." When we are unsure whether to cut, we look at whether the line still holds up even if no particular person wrote it. A line that holds up that way usually loses no meaning when deleted.
What you keep to make it look like a person wrote it
Something we noticed while repeating this cutting work: erasing the AI feel is not the work of removing stock phrases. It is also the work of keeping "the information that exists only in that specific situation."
Concretely, that means things close to numbers and specific events. Not "we tried it several times," but "we tried it three times." Not "it did not go well," but "the process stopped partway through." Take abstract phrasing back down to the grain of what actually happened at the time. The finer the grain, the less that writing can be swapped out for anyone else's writing. As lines that cannot be swapped out pile up, readers seem to feel that "this was written by someone who actually did it."
Put the other way around: the more the text fills with high-abstraction phrasing — words that only state an evaluation, like "effective," "important," "useful" — the closer it moves to a generic product that would fit any topic at all.
What happens when you cut too much
Subtraction needs a sense of proportion. If you focus too hard on cutting stock phrases, you can end up avoiding definite statements so thoroughly that the whole text goes blurry. Expressions that leave room, like "for now" or "at this point," are useful for avoiding overstatement, but overuse them and you get writing that says nothing at all.
The standard for cutting is, in the end, "does meaning drop if this line disappears." Cut the stock phrases that lose no meaning; keep the parts where meaning would drop, even if the phrasing comes out somewhat strong.
The last pass
Brand Voice Editor does not write drafts. It does not decide policy either. By the same logic as separating the role that writes from the role that tidies, we carved out the cutting judgment alone and gave it to this role. Writers have a hard time noticing the habits in their own writing. We think that holds true for AI and for people alike.
Break up over-polished writing, just a little. Bring certainty back down to the height of what was actually done. It is a plain, unglamorous step, but a draft that has been through it and one that has not leave a different texture behind once you finish reading. Recording it here.
