Where copy stops sounding like the brand

Off-brand copy is not usually a writing problem. It is an ownership problem with a clock attached. The style guide was written in March, the campaign ships in November, and in between the voice was maintained by memory — each writer, each model, each freelancer holding a slightly different version of "how we sound." Nobody decided to break the guide. The guide simply stopped being checked.

The four places it starts

• Hand-offs to people outside the core team. Freelancers, agencies, and AI assistants inherit the brief, not the ten years of tone decisions behind it. The output reads fluently, which is exactly why it passes a casual eye — and it is fluent in someone else's voice.
• Deadlines around launches. Under time pressure, copywriters drop from the guide's careful register to the market's default register. "Disruptive game-changer" appears, not because anyone forgot the banned-word list, but because nobody had four minutes to consult it.
• Recycled boilerplate. A landing page from the seed-stage brand gets cloned for the enterprise-tier product. Every sentence is grammatical. None of it sounds like the company you are now.
• Channel drift. The same sentence has to work in a release note and a tweet. Writers compress, add an exclamation, swap "recommended" for "insane" — and the voice moves a little at a time, below the threshold where anyone would raise it in review.

Why a written guide does not hold on its own

Style guides are good at naming a voice ("knowledgeable, not condescending") and bad at policing one. Research on how tone actually decomposes — Nielsen Norman Group's well-known four dimensions of tone of voice — describes voice as a position on several continuous scales rather than a single label. That is the practical problem: a label in a document cannot be checked, but a position on scales can be violated over and over without anyone noticing. Corporate guides such as the Microsoft Writing Style Guide work because they translate values into concrete rules — banned constructions, preferred terms, a fixed point of view. Rules can be checked. Vibes cannot.

Why normal review misses it

Editorial review reads for correctness: facts, grammar, legal risk, typos. Voice failure is none of those. A sentence with a banned word, a passive construction your guide forbids, and three exclamation marks in a row is not a typo — it is a mismatch against a written rule, and the reviewer has to hold the whole guide in their head to notice it. Almost nobody does. That is why the slip is usually surfaced by an outsider: a client who says "this doesn't sound like you," or a support agent who sees the confusion downstream.

The cost lands later than the mistake

Off-brand copy almost never fails loudly at the moment it ships. It fails quietly: engagement sags and nobody can say why; an agency deliverable gets rejected after the schedule was built around it; two landing pages for the same product read like two different companies, and the reader feels the inconsistency without being able to name it. Plain-language research on document usability makes the same point from the reader's side — comprehension and trust drop when text does not behave the way the reader expects (see the U.S. Plain Language guidelines at plainlanguage.gov). A brand voice is a set of reader expectations. Every unchecked slip charges interest against them, and the bill arrives after publish, when fixing is ten times more expensive than reviewing.

What a mechanical check should report

Three things, and only three, to be useful before the deadline: which rule broke — quoted from your own guide, not from a generic style opinion; which sentence broke it — quoted from your copy, so the fix is obvious; and how badly — a severity rank that tells you whether to stop the press or file it for next week. Anything more elaborate (rewrite suggestions, tone scores, personality assessments) belongs downstream of a human decision, because the moment a tool silently rewrites copy, you no longer know whose voice shipped.

The mechanical answer is to check every draft against the guide itself — parse the guide's own never/avoid/prefer rules, run them against the copy, and rank the hits by how badly they breach it. CopyComply does exactly that: it reads your copy against the style guide you paste and reports the lines that no longer hold.

It reports candidates for review. It does not rewrite your copy, does not publish for you, and a clean run is not proof the copy is on-brand — only that it passed the rules that were checked.