Brand checks that people do not learn to ignore

The failure mode for any automated writing check is not missing things. It is training the team to click past it. Linters that flag sixty trivial items per file end up ignored on the five that matter. A brand-voice check has to earn its place on every run, which means being deliberate about what it blocks on, what it merely warns about, and what it refuses to judge at all. The design constraint is attention, not coverage: a check that surfaces everything teaches readers nothing, while a check that surfaces the right tier teaches a habit.

Block on what is certainly wrong

A banned term is a fact, not an opinion. If your guide says never use "cutting-edge," then a draft containing "cutting-edge" has broken a written rule — the check can quote the rule, quote the sentence, and file it as a high-severity finding without any judgment call. The same goes for constructions a guide explicitly rules out: emoji in release notes, all-caps shouting, a competitor's old product name. These checks never produce false positives about whether the rule was broken, only about which rule you actually meant to enforce. Block here.

Warn on what is probably wrong

Patterns that good writing guides agree on, but that sensible brands sometimes break on purpose, belong in the middle tier. Sentence length is the classic one: plain-language authorities such as the U.S. government's plainlanguage.gov guidance push toward short, readable sentences, and readability formulas like the Flesch and Gunning Fog index operationalise that into a threshold. A 40-word sentence in a whitepaper may be perfect; the same sentence in an app-store listing is a problem. Passive voice behaves the same way — the Purdue OWL explains why guides recommend active voice for clarity, and why "mistakes were made" is occasionally exactly the sentence you want. Warn, show the count, let a human decide. Style guides written for engineering audiences, like Google's developer documentation style guide, encode hundreds of these preferences — useful as defaults, wrong as hard failures.

Leave alone what cannot be decided mechanically

Whether the joke lands. Whether the tone is right for this campaign. Whether the copy, technically clean, still sells. These are judgments against intent, and intent lives outside the text. A tool that pretends otherwise produces confident noise, which is worse than silence — it burns the credibility you need for the checks that genuinely know things.

A workload that stays sane

Teams that keep a voice check alive tend to run it on a fixed, boring cadence: one scan per draft, before the first human review, never after the schedule is committed. The scan is fast precisely because it is mechanical — parsing the guide's own never/avoid/prefer lines and matching them against the draft takes seconds and produces the same ranks every time. What the cadence protects is the reviewer's attention: if the block list is already cleared by the time a human reads the draft, the human spends their effort on the judgment calls the tool deliberately refused to make. That division — machine on rules, human on intent — is the only arrangement that has held up in practice. Teams that reverse it (humans hunting banned words, tools pronouncing on tone) produce exhausted reviewers and untrustworthy scores.

What to measure instead of compliance theatre

A percentage score is the enemy: it invites bargaining ("92% is probably fine") and hides the one finding that matters. Named, tiered findings survive disagreement — someone can argue with a rank, but nobody can argue with a quoted rule and a quoted sentence. If you must report a number, report the count of open HIGH findings, because that is the only figure with an obvious action attached: zero, or the draft does not ship. Severity distributions, trend lines, and voice dashboards sound like governance and function as decoration; the fix checklist, reviewed and closed line by line, is the actual control.

Tiering is the product

What teams actually want from a voice check is not more findings; it is a trustworthy ordering of them. Severity tiers — banned-term breaches first, measurable style slips second, everything else left to the editor — are what keep the check on the desk instead of on the mute list. CopyComply separates these tiers rather than collapsing them into one pass/fail score, because a single undifferentiated signal is what gets muted.

Decision-support only: findings quote your guide and your copy. You make the call before anything ships.