The style guide nobody opens, and what replaces it

Every organisation past about six people has a style guide, and in most of them it is a forty-page PDF that three people have read and nobody has open. The usual diagnosis is that people are careless. It is not that.

The failure is structural, not editorial

A style guide is a document you are meant to consult at the exact moment you have least capacity to consult a document.

The moments that produce off-brand work are not leisurely. Somebody is finishing a post at 6pm, replying to a comment on their phone, writing a headline while the deadline moves. The guide is somewhere else, it is long, and finding the relevant rule takes longer than writing the sentence did. So the rule does not lose an argument with the writer. It never enters the room.

That reframes what a good guide is. It is not the most complete one. It is the one that reaches the moment of writing.

Three things follow.

Rule one: length is a cost, not a virtue

A forty-page guide has a coverage advantage that is entirely theoretical, because coverage only counts when it is consulted. A one-page guide that people have actually read has real coverage of maybe six rules, and six enforced rules beat two hundred unenforced ones every time.

The instinct that produces the long version is reasonable — every rule in it was added because somebody got something wrong once. But a guide is not an incident log. Each addition costs a fraction of the attention available for the whole thing, so rule two hundred is not free, it is paid for by rules one through five getting skimmed.

Rule two: most of it is not a style guide at all

Open a typical one and sort its contents into two piles.

Mechanical rules. Oxford comma or not. Sentence case or title case in headings. Whether numbers under ten are spelled out. Which spelling of a product name is correct. These are checkable by a machine and have exactly one right answer.

Judgement rules. What we sound like when we are apologising. When to be funny. What we never claim. How much we hedge.

The first pile does not belong in a document at all. A rule with one right answer that a machine can check should be enforced by the machine — a linter, a checker, a template, a locked field. Putting it in a PDF converts a solved problem into a memory test that people fail while doing their real job.

What survives as a document is the second pile, and it is short. Usually three to six items, all of them things no tool can decide.

Rule three: the refusals are the part with load-bearing weight

The judgement pile splits again, and the half that gets written is the wrong half.

Almost every guide describes what the brand is: confident, human, clear, expert. Those adjectives are close to useless in the moment of writing, because they are compatible with almost any sentence. Nobody has ever deleted a draft for insufficient humanity.

What decides a sentence is what the brand refuses. We do not use exclamation marks. We do not say "excited to announce". We never claim a result we cannot evidence. We do not describe our own product as simple. A refusal is checkable in a second and it is falsifiable, which is why the twelve voice cards all print a refusal column beside the says column, and why the profession pages carry a notFor rather than only a target.

If your guide has no refusals in it, it is a values statement, and values statements do not change drafts.

So what replaces it

Three artefacts, in descending order of how much work they do.

The mechanical rules move into the thing people are already using. A template with the type locked. A checker that runs on the draft. A stored brand profile every generator reads before it produces anything. This is the whole argument for holding a brand as state rather than re-describing it per task, made at length in describing your brand once.

The judgement rules become a page, not a document. Short enough to read in a minute, refusals first, examples over adjectives. It is brand guidelines in the only form that survives contact with a busy week.

One question survives that neither of the above can answer. With the logo removed, would a regular reader attribute this to us? That is a judgement about a relationship, no scoring system has it, and it needs a second pair of eyes or a day's distance. Why a checklist catches drift and taste does not is about the split between the half a machine can hold and the half it cannot, and it is careful about which is which.

The uncomfortable part, since we sell one half of this

Repic is the machine in the middle artefact, so treat the argument accordingly: it is an argument in our favour and you should check it rather than take it.

The honest limit is that moving the mechanical rules into a tool does not raise the ceiling on the judgement rules at all. A brand with a thin description gets consistency around a thin idea, which is worse than an inconsistent brand with a sharp one, because it is harder to notice. Our consistency checker has a narrower version of the same hole: it reads writing and palette contrast, so a feed that drifted through imagery looks fine to it.

The one thing worth taking from this whether or not you use any tool: stop measuring your guide by how complete it is and start measuring it by how often it is open. If the answer is never, the length is the reason, and cutting it is not a compromise.