Our marketing site is 72 pages, all of it written by hand into page files and rendered by one component. There is a house style, some of it written down and some of it not. I wanted to know which parts of it the pages actually obey, so I fetched every URL in the sitemap, extracted the text inside <main> with a parser rather than a regex, and counted.
56,021 words. 2,031 paragraphs. Here is what a grep can see about a voice.
The absolute zeros
exclamation marks 0
em dashes 0
en dashes 0
hype adjectives (18 searched) 0
The eighteen were: seamless, effortless, powerful, robust, leverage, utilize, amazing, revolutionary, cutting-edge, game-changing, unleash, supercharge, delightful, magical, innovative, world-class, best-in-class, stunning. Not one appears anywhere on the site. Neither does a single exclamation mark across 56,021 words, on a site selling a product whose whole subject is a fun night out.
The em dash zero is the one I am most confident is real policy, because it is the rule the project writes down and enforces on every new line of copy. Open any page and search for one. The host guide is about 1,500 words of prose and does not contain a single one.
The rule that is not written down
Then there is easy, which appears 66 times across 25 pages, plus simply 8, just 16, simple 4, quick and quickly 12.
Before calling that a failure I read them, which is the step a grep tempts you to skip. Most are ordinary English rather than a claim about our software:
A round where teams draw something, list ten things in any order, or write a sentence is easy on paper and hard or impossible on a phone.
Ten is the default because it is easy to total and it fits the attention span of a table between drinks.
That is "easy" as a description of a quiz round, not as a promise about a product, and I would not want it gone. The difference between the zeros and the 66 is not that one set of words is worse. It is that one rule got written into the project and the other one never left somebody's head, so it survives in some paragraphs and not others. Counting is how you find out which of your style rules are real.
Person and distance
you, your, yours 855
we, our, us, ours 135
they, their 251
Six times as much second person as first. Every one of the 72 pages uses "you" at least once. The highest "we" counts are, predictably, the two legal pages, where "we" is the data controller and has to be.
That ratio was not a decision anybody recorded either, but it is consistent enough across 72 independently written pages that it must be a habit rather than an accident. The copy describes the reader's quiz night and mentions our software as the thing that would help with it, which is also why the most frequent content words on the site are quiz, answer, round, night, questions, room, table, team and pub, and the product's own name is nowhere near the top.
How formal is it, measurably
contractions (don't, it's, you're) 120
expanded forms (do not, it is, cannot) 580
Nearly five to one in favour of the expanded form. I did not know that until I counted, and it explains something I had vaguely noticed about the site's tone: it reads as calm and a bit stiff, like a trade magazine rather than a startup landing page. Sentence shape agrees:
sentences in paragraphs 2,923
mean words per sentence 12.8
median 10
longest 51
sentences of 30+ words 198
paragraphs that are one sentence 1,307 of 2,031
Two thirds of the paragraphs on the site are a single sentence. That is the house style doing something unintentional: when a page is a list of sections, and each section is a heading plus a few paragraphs, writers reach for one idea per paragraph and stop.
The 477 question marks, and where they are not
477 question marks sounds like a site built on FAQ. It is not. They sit on 23 of 72 pages, and 13 of those are the free question pages, carrying 30 to 34 each because the page is literally a list of quiz questions. The FAQ has 18. Pricing has 7. The homepage has 3.
So the site asks almost no rhetorical questions in its sales copy, which is another habit nobody wrote down.
The most repeated sentence on the site
Every content page ends with a call to action, and 68 of the 72 pages end with the same five words on the button: "Start free, no card needed". The phrase "no card" appears 113 times on 68 pages.
That is either admirable consistency or wallpaper, depending on how many pages a visitor reads. The headings and bodies above those buttons are all different, 67 distinct headings and 67 distinct bodies, so the repetition is deliberately confined to the one string that is a promise rather than a pitch.
The other repeated strings are smaller and more interesting: 23 pages carry the same "Last reviewed 14 September 2026" line, and 6 carry the same sentence explaining that the competitor facts on that page were read off the competitor's own site in September 2026.
What the count cannot see
Everything that matters most. Whether the sentences are true, whether the guides are actually useful to somebody running a quiz on Tuesday, whether the tone fits a landlord rather than a developer. A word count cannot read.
What it can do is tell you which of your rules exist. Mine came back as four exact zeros and one 66, and now I know which of those five rules was ever really written down.
If you want to check any of it, every page is public and the whole corpus is the sitemap. Search any page for an exclamation mark, then search it for the word "easy".
Top comments (0)