A text produced by an assistant comes back for review. It is correct, readable, free of mistakes. The reviewer rejects it without being able to say why. The problem is the sector’s codes, which nobody has written down.
Generative models produce average prose by default: warm, fast, explicit. In a constrained sector, those three traits are the ones the profession rules out.
Editorial line and AI come down to the same work: writing down what your team applies without ever having stated it. An editorial line, which we call an editorial remit, is that written decision.
What AI does not know: tone, tempo, compliance, structure
Tone. Too warm or too simplified, it breaks the professional distance readers expect in finance, law or health.
Tempo. The model writes fast and short. Some sectors expect the opposite: a slow rhythm, long texts, positions that do not change every quarter.
Compliance. A text produced by the model may state something inaccurate, cite a source that cannot be checked, or use a word your sector reserves for one precise use. In some sectors, imprecision is not an editorial weakness: it is a compliance risk. A vague promise, a risk played down or an essential piece of information pushed out of sight can amount to a misleading practice, a regulatory breach or professional misconduct.
Structure. The reader’s place, the share of what is left unsaid, the degree of assurance are markers of the profession. An ungoverned model shifts them.
None of these four points appears in a tone-of-voice instruction. All of them are passed on through practice, never through a document. Neither a tool nor a new recruit has access to them.
Why imprecision becomes a fault
The law does not stop at whether a sentence is literally false. Regulators also look at:
- what the public can reasonably understand;
- the important information that has been left out;
- where, how legibly and when the information is given;
- the imbalance between benefits put forward and limits, risks or conditions played down;
- whether the organisation can substantiate what it claims.
A careful, accurate wording can therefore still mislead through its overall effect.
The European directive on unfair commercial practices sets out this principle. A commercial practice is misleading where it “in any way, including overall presentation, deceives or is likely to deceive the average consumer, even if the information is factually correct” (Article 6(1)). A practice is also misleading where a trader “hides or provides in an unclear, unintelligible, ambiguous or untimely manner such material information” (Article 7(2)).
Source: Directive 2005/29/EC of the European Parliament and of the Council of 11 May 2005 concerning unfair business-to-consumer commercial practices in the internal market.
A generative model puts the benefit first, condenses the nuance and pushes the condition down the page. That is the imbalance regulators examine.
Three sectors, three systems of signals
Health. The tone stays neutral and informative, and the intent is to inform. An assistant suggests “discover the benefits of this treatment”: the wording promises a therapeutic result. Terms that promise such a result fall under a regulatory framework, and your legal department keeps the list. The editorial remit adopts that list, instead of rediscovering it with each text.
Finance. Content informs; it does not recommend. An assistant suggests “five reasons to invest right now”: the headline urges action where only information is allowed.
Luxury. Restraint comes first: few words, suggestion, ellipsis. An assistant suggests “a bag that combines elegance and practicality for every occasion”: the sentence explains, justifies and sells, where the house simply shows.
In all three cases, the gap is structural. The editorial remit sets a system of signals that readers in the sector recognise.
What happens without written rules
The scenario below is a typical situation, not a client case.
A company in a technical sector gives its marketing team an assistant to speed up article production. The texts come out quickly. At review, the subject-matter experts reject many of them: too general, or wrong on a point only a specialist can see. Production stops while the editorial remit is rewritten.
Three effects add up. Your positioning is diluted, because content with no grounding in the profession becomes interchangeable. Your voice becomes unstable, because the model shifts register from one text to the next. And you lose credibility, because in B2B your content is often the first contact anyone has with your expertise.
The tools are not at fault. It is the absence of written rules that leaves the model to decide.
Writing an editorial remit a tool can apply
Set down the foundations. Objectives, tone, formats, channels, frequency, measures: everything that used to be passed on by example gets written down.
Lock the language. A general tone of voice is not enough. You need a priority vocabulary, approved wordings and writing prohibitions, in other words an editorial charter your teams and your tools can both use.
Write down your sector’s markers. Models work from general corpora: they know nothing of your thresholds of caution or your professional norms. These markers belong in a review grid: timing, stance, emotional charge, relationship with the reader. The grid does two jobs: instructing the tool and reviewing what it produces.
Document the voice. That is what the Brand Voice Framework is for: nine pillars that set the brand’s voice, its registers and what it excludes.
Put a steering system in place. An editorial remit filed away does not apply itself: it needs roles, checkpoints and a dated review. This is what we call a Content Operating System, and it is the level at which editorial governance is exercised.
What a tool-ready editorial remit contains
- Foundations: objectives, formats, tone, frequency, measures.
- A language: your own vocabulary, excluded registers, approved wordings.
- Sector markers: timing, stance, the reader’s place.
- A documented voice: the Brand Voice Framework, kept up to date.
- A steering system: roles, approval points, a dated review.
A written editorial remit sets the model’s tone, tempo and structure. Without one, it chooses them for you.
Frequently asked questions about the editorial line and AI
How do you adapt your editorial line to generative AI?
Write down what was left unsaid: the tone, the tempo, the reserved words, the reader’s place. Add a review grid specific to your sector, which serves both to instruct the tool and to check what it produces.
Why does AI-produced content ring false in a constrained sector?
Because the model learned from general corpora. It produces average prose: warm, fast, explicit. A constrained sector expects the opposite, without ever having written it down.
Do you need a separate editorial line for AI?
No. It is the same editorial remit, made explicit and enforceable. What a team applied out of habit has to become a rule that can be checked at review, whether that review is human or assisted.
To put your sector’s rules in writing, contact us.