Brand Voice Checker

Ask a team what its brand voice is and you will hear adjectives: warm, clear, confident, never stuffy. Ask to see the voice guide and you will usually find something more useful than adjectives. You will find rules. Say “use”, not “utilize”. Write the product name with a capital in the middle. Keep sentences short. Talk to the reader as “you”. Do not shout with exclamation marks.

That gap between the feeling and the rules is the whole argument of this page. A feeling cannot be checked. A rule can.

What a voice guide actually contains

Public style guides show the pattern plainly. The GOV.UK A to Z style guide is a long alphabetical list of small decisions, most of them about single words: when to use a capital, how to write numbers, which contractions to avoid. Its advice on numbers is a good example of a rule a machine can apply: write numbers as numerals, including 2 to 9, but write out a number that starts a sentence and keep “one” in many ordinary uses. Its entry on contractions is just as concrete. It asks writers to avoid negative forms such as “can’t” and “don’t”, because many readers find them harder to read or misread them as the opposite of what they say.

The US National Archives puts its plain language advice into ten principles, and the last one is the interesting one here. It asks staff to promote a consistent voice by following the agency’s own style guide. Consistency is not a mood the writer is in; it is the habit of applying the same written choices every time.

Read enough of these guides and three families of rule appear. There are word rules: preferred terms, words to avoid and names spelled one exact way. There are shape rules: how long a sentence or a paragraph may run. And there are habit rules: whether the writer says “we” or “you”, whether contractions are welcome, how numbers and lists are punctuated. Every one of them can be counted in a draft, and every count can point at the exact words that caused it.

Why a list of findings beats a single number

A score from 0 to 100 feels efficient. It hides everything you need to act on. If a draft scores 62, which sentence lost the points? Was it the jargon, the length or the exclamation mark in the second paragraph? A writer cannot fix a number. A writer can fix “utilize” in line three, and an editor can disagree with a finding in a way nobody can disagree with a score.

That is why this tool shows the target beside the measured value for every rule and lists each finding with the rule that raised it. Nothing is averaged into a verdict. If your guide allows 25 words in a sentence and one sentence has 42, you see that sentence, the count and the limit, and you decide.

The rules in the example guide on the page come from published plain language advice rather than from taste. The US plain language guidance on digital.gov lists swaps such as “help” for “assist”, “begin” for “commence” and “use” for “utilize”, and it suggests “to” in place of “in order to”. It also argues for “must” over “shall”, because “shall” can signal either an obligation or a prediction. Those are exactly the kind of entries a word list holds.

Word lists and their traps

Word lists look trivial until you run one. The first trap is the substring. A check that looks for the letters “art” will fire inside “start”, “smart” and “party”, and a writer who sees three false alarms in a row stops trusting the fourth. Whole words matter. A match should only count when the characters on either side are not letters or digits, so that “art” matches “art” and “Art” but never the middle of another word. The checks behind this page run the matcher against a slow, letter by letter reference on 3,000 random texts full of punctuation, capitals and accents, and the two must agree on every match.

The second trap is the plural, and its cousins. A guide that bans “synergy” says nothing about “synergies” unless someone writes that down. The same is true of “utilize”, “utilizes” and “utilizing”. A prefix marker helps for words to avoid, but replacements are harder: “utilizing” should become “using”, not “use”. The honest answer is a separate entry for each form you care about, which is how the example guide handles it.

The third trap is the name. Brand and product names have a single correct spelling, and the usual mistakes are case mistakes: all capitals in a headline, a missing capital in the middle, a lower case first letter. A name rule therefore matches without regard to case and then compares the exact characters. That catches “Yourbrand” and “YOURBRAND” while leaving the correct form alone.

There is a quieter trap underneath all three. The same visible letter can be stored in more than one way. MDN’s page on String.prototype.normalize shows that an e with an acute accent can be stored as a single code point or as a plain e followed by a combining accent; the two look identical and compare as different strings until both are normalised. A word list that skips this step will miss a term that a reader can plainly see. Text segmentation has a similar wrinkle across languages: splitting on spaces fails for Japanese, Chinese, Thai and other scripts written without them, which is why browsers now offer Intl.Segmenter for words and sentences.

Sentence length, paragraphs and plain language

Shape rules have the best support of any in a voice guide. GOV.UK asks writers to split sentences over 25 words and to keep paragraphs to no more than 5 sentences. The digital.gov guidance suggests paragraphs of no more than 150 words in three to eight sentences, says a paragraph should never pass 250 words, and asks for one idea in each sentence. The two governments chose different numbers, which is a useful reminder that a limit is a decision a team makes, not a law of nature.

Finding a sentence is harder than it sounds. A period ends most sentences, but not the one after “Mr.” or “Dr.”, not the one inside “3.5” and not the one after an initial. Browser segmentation follows general Unicode rules, and in a test of 20 short texts for this page the segmenter split after titles such as “Mr.” where a reader would not. So the tool uses its own stated rule for English and tells you where it can go wrong.

Voice rules sit close to these. GOV.UK asks writers to address the reader as “you” and to write conversationally, as if talking to one person. US guidance prefers the active voice because it makes clear who should do what. It also gives a usable test for the passive: a form of the verb “to be” followed by a past participle, which usually ends in “-ed”. That test is good enough to raise a hint and too crude to deliver a verdict. “We are excited” passes the test and is not passive at all, so the tool shows passive voice as something to look at, never as a fault.

The serial comma gets the same treatment. Wikipedia’s article on it notes that it is common in American and formal writing, usually left out in journalism, and can either resolve or introduce ambiguity; a pattern match cannot tell a list from a clause, so it can only ask.

What readability formulas can and cannot tell you

Readability formulas are the closest thing writing has to a gauge, and they deserve a careful reading. The Flesch-Kincaid grade level was derived for the US Navy in 1975 from tests with enlisted personnel. In the report, the new grade formula is 0.39 times the average words per sentence, plus 11.8 times the average syllables per word, minus 15.59. The older automated readability index uses characters instead of syllables: 4.71 times characters per word plus 0.5 times words per sentence, minus 21.43, rounded up to a whole grade. It was designed for real time monitoring on electric typewriters, where counting keystrokes was easier than counting syllables.

Here is a worked example you can reproduce. Paste “The Australian platypus is seemingly a hybrid of a mammal and reptilian creature.” into the tool as a draft. Wikipedia counts it as 13 words and 24 syllables, which gives a grade of 11.3 and a reading ease of 37.5. The tool’s own syllable rule also finds 24 here, so with the example guide’s target of grade 9 or lower, that single sentence produces one readability finding with its value printed beside the target.

Now the limits, because they are large. The grade has no upper bound, and its floor is minus 3.40, reached by “Go. See. Stop. Rest.” A formula sees only lengths, so it rates nonsense made of short words as easy and a precise legal sentence as hard. Its syllable counts come from rules, not a dictionary; on a list of 76 hand-counted words, the rule in this tool matched 63 exactly. The constants are tied to English, and to one population of readers. Even the 1975 report gave a recalculated ARI with different constants from the original, so “the” readability score depends on which version you mean.

Used well, a grade is a smoke alarm. If a help page that usually sits near grade 7 jumps to grade 13, something changed, and the long sentence findings will usually show you what.

One shared file keeps a team consistent

The hardest part of a voice is not writing the guide. It is keeping five writers on the same version of it. A rule that lives in a slide deck drifts the moment someone copies it into a new document. A guide saved as a small file can be opened by anyone, and checking every draft with that same file produces the same findings for everyone.

That is also how you compare drafts fairly. Run the launch email, the help page and the social post through one guide and look at them side by side: which rules each one breaks, and how far apart their average sentence length, grade and “you” share sit. A spread of four grade levels between two pages written for the same readers is worth a conversation. A single “consistency score” would have hidden it.

Rules will never capture everything a reader feels about your writing. They do capture the decisions your team already agreed on, and they let anyone check them line by line.