A house style holds across fifty pieces only if it is written down as rules

Consistency fails piece by piece, not all at once

A single article can be brought into line with a house style in one editing pass, and this makes the problem look solved. Across fifty pieces written over several months, by more than one person, with the instruction reworded slightly each time, the drift arrives gradually and nobody sees the moment it happens.

The symptoms are small and cumulative. Dates in two formats. Numbers spelled out in one piece and written as digits in the next. The organisation referred to as we in some articles and by name in others. Headings that are sentences in the early work and noun phrases in the recent work.

Individually none of these matter. Together they are the difference between a publication and a folder of documents, and a reader registers the difference without being able to name it. So does anyone assessing whether the whole thing was edited by a person.

Split the style into rules and taste

The useful division is between what can be checked mechanically and what requires judgement. Date format, number style, capitalisation of headings, whether lists take full stops, how the organisation refers to itself, which spellings are house — all of these have exactly one right answer and no interesting argument attached.

That half belongs in a short, blunt list. Not prose about our voice, but a column of decisions, each of which resolves a question that would otherwise be answered differently each time. Twenty lines usually covers it, and twenty lines is short enough that people actually read it.

The other half — rhythm, when to use a metaphor, how much to assume of the reader — cannot be reduced to rules without producing something stilted. It is transmitted by example, and that is a separate mechanism which the rule list should not pretend to cover.

Examples carry what the rules cannot

Keep two or three published pieces designated as reference. Not the best ones necessarily, but the most typical: the ones you would point at and say this is what we sound like when we are working normally. Supply them alongside the brief whenever a new piece is drafted.

A reference piece transmits register, sentence length, how much the reader is assumed to know and how directly claims are made. Those properties resist description and survive imitation quite well, which is the opposite of the situation with the mechanical rules.

Refresh the references when the publication genuinely changes, and not otherwise. A reference set that is quietly updated every few weeks provides no anchor at all, and you end up with a style that follows the drift instead of correcting it.

Enforce the mechanical half after writing, not during

Style rules stated in the drafting instruction are followed unevenly, particularly when there are many of them and the piece is long. A rule that was honoured in the first three sections and forgotten in the fifth is the normal outcome, and it isn’t a sign that the rule was badly worded.

Treat compliance as a separate pass over finished text with the list in front of you, or as a mechanical search where the rule permits it. Searching for the wrong date format finds every instance in seconds; asking for the right date format finds most of them, most of the time.

This also keeps the drafting instruction short, which improves everything else about it. A brief carrying thirty style rules has very little room left for saying what the piece is supposed to argue.

Audit backwards occasionally

Once a body of work exists, the most informative thing you can do is read four pieces from different months in one sitting. Drift that is invisible in sequence becomes obvious in comparison, and you will usually find two or three conventions that changed without a decision being taken.

At that point you either update the list to match what the work actually does, or correct the work. Both are legitimate. What isn’t legitimate is discovering the divergence and doing neither, because the rule list then becomes a document that describes an aspiration rather than the publication.

None of this is really about the tool. It is ordinary editorial practice, and it matters more when drafting is cheap, because a cheap draft is easy to accept in whatever shape it arrives — and a hundred pieces accepted in whatever shape they arrived is precisely how a body of work stops looking edited.