In three years, the term has reached the boardroom. It has also gained an elasticity that weakens it: everyone uses it to name what they already do. The calendar becomes a governance. The brand guidelines become a governance. The shared review file becomes a governance. Meanwhile, the question the word was coined to designate goes unanswered: who decides what the brand says, under what rules, and what happens when that person leaves?
The answer has changed in nature since generative models started writing. Here is what editorial strategy, governance and steering cover today, what has moved, and where an organisation takes back control.
Three words we confuse, and what the confusion costs
These words designate three distinct levels, and the order matters.
Editorial strategy decides the destination: who the organisation speaks to, on which subjects it is legitimate, what it refuses to say. It is settled once, revised each year, and fits in a few pages.
Editorial governance installs the rules, the roles and the approvals that produce coherence without depending on individual judgement. It answers the thankless questions: who arbitrates a disagreement over tone, who has the last word on a headline, at what point a piece leaves the circuit. Governance is recognised by one thing: it survives the departure of the person who wrote it. We were already arguing the urgency of editorial governance before generative AI made it a conference subject; the urgency has since changed scale.
Editorial steering measures and corrects: what gets published, what gets read, what gets cited, and what you decide to stop.
An organisation that skips the middle level produces what we call authorial drift: as writers multiply, agency included, the voice drifts. Every piece holds up on its own. The whole no longer resembles anything. The cost appears on no budget line, which is why it is left to run: this is editorial debt, and it is paid every month.
What generative models have moved
For twenty years, the editorial bottleneck was production. Writing was expensive, so little was written, so coherence largely held by itself: few hands, little drift.
That bottleneck is gone. Producing costs almost nothing now, and the point of tension has moved to the decision. A marketing department can feed every channel without ever having settled what its brand is entitled to say. The result is mechanical: more volume, less signal.
A second shift commits more. Generative engines now return brands where they used to rank pages. When an executive asks a model what a given company does, the answer is built from everything that company has published elsewhere, and from the coherence of that whole. A brand whose language changes from one channel to the next collects the average answer of its sector, the one in which nothing distinguishes it. Citability by models has been added to visibility in search engines, and it answers to the source, not to the markup. It can be measured: that is the object of an AI footprint audit, and it extends the EEAT criteria Google installed well before generative models.
Hence the principle we hold in our training work: the tool amplifies what it finds. Without documented architecture, teams diverge, the voice varies from one medium to the next, and the model smooths over the rest. The blur you start with comes back amplified.
The debt is paid even by those who teach it
An example, and it is our own.
Auditing our own site in 2026, we counted twenty-six pages positioned on variants of “content strategy”. Three of them ran to more than three thousand words each and competed for the same query. None broke through. Traffic on the generic term went to the home page, through the brand, not through the subject. Years of serious work had built editorial capital that fought itself.
Nobody had worked badly. Every article, on its date, answered a sound intention. What was missing was the rule that would have said, at the tenth article: this one already exists, strengthen it instead of writing an eleventh. That is what governance is for, and that is what it costs when it arrives after the fact. We are now in the consolidation phase: one master page designated, the one covering brand equity and content strategy, with differentiated satellites around it. This is remedial work, and it recovers authority already paid for.
We tell this story on stage because it disarms the room. An editorial engineering firm that shows its own debt earns a hearing that ten best practices never earn.
The four grounds on which governance is judged
Editorial governance is judged by what it makes possible, not by the volume of documents it produces. Four grounds show it, and their order is an order of dependency.
- Clarity of the offer. An offer understood at first reading sells better than a brilliant, confusing one. Until an organisation has named and ranked what it sells, no editorial rule downstream will repair the ambiguity. This is the level that commands all the others, and the one our Clarity programme starts with.
- A documented voice. A brand voice is documented and governed like an asset: it is recognisable, it holds across every medium, it transmits to teams and to suppliers alike. A voice that exists only in the communications director’s head becomes a continuity risk the day they leave.
- A steered editorial system. The organisation moves from case-by-case production to a governed whole, where coherence outranks volume and every piece serves an intention. This is the level that absorbs AI without deforming, and the one our method for building a B2B editorial system describes.
- Authority. The position of reference that gets cited is built by peers, by the press, and now by generative models, which return a brand all the better when its speech is clear at the source. This is what the Hub documents, our knowledge centre on editorial engineering.
This last ground is best understood through a case. We audited the AI footprint of a major public reference institution, on subjects where it has held authority for decades. The finding surprised everyone, starting with the institution itself: its content is reproduced almost word for word by ChatGPT, Gemini, Perplexity and Copilot, but the source has disappeared from the answer. The institution is absent, or cited without attribution, while other bodies occupy the place that belongs to it.
The important point lies in what the audit did not find. The content remains accurate, current, better documented than that of the sites replacing it in the answers. The deficit concerns citability, not knowledge. An organisation can hold the reference knowledge in its field and disappear from the place where the public now looks for it. This is a loss of authority without a loss of competence, and it can be repaired: making content citable, structuring the data, holding the speech at the source. Muriel Vandermeulen presented this case at the All for Content trade show in Paris, on 1 July 2026.
Why field experience has become a criterion again
The subject attracts a great many commentators now that generative AI has made it a conference theme. Most speak of editorial governance as an idea. Few have installed one, and fewer still have watched one fail.
The gap shows in the room, at the first concrete question. A brand manager asks what to do with a piece produced by a model and approved by nobody. The theoretical answer describes the circuit that ought to be set up. The field answer describes the circuits that were set up, the one that held, and the precise moment the others gave way. Audiences never remember the first.
That is the material Muriel Vandermeulen brings to her talks. She founded WeAreTheWords in 2000, published in 2011 the first French-language book devoted to web content strategy, has since supported more than a hundred and fifty companies and trained more than fifteen hundred professionals. Twenty-six years of documented trial and error, which yield a rare repertoire: the bad practices you have tested yourself.
What field experience changes, participants say better than we do.
Finally, content that is concrete and workable, not full of fine phrases you don’t know what to do with afterwards. An example, what works and what doesn’t, then a good practice.
Gisèle Llados, CRM manager, La Redoute IntérieursGenuinely interesting, with substance, and very lively.
Bruno Lalanne-Claux, CRM officer, Musée du LouvreBoth testimonials translated from the original French.
The talks take place where professionals train and debate: chambers of commerce, the Belgian Association of Advertisers, WordCamp, the All for Content trade show, in France and in Belgium.
Bringing the subject into your organisation
Editorial governance always meets the same obstacle: it interests the teams before it interests the people who fund it. An editorial frame that no executive has understood does not survive the first budget arbitration.
An outside voice removes that obstacle. It puts everyone in front of the same scene, at the same moment, with the same words. The format follows the context: a plenary to open a convention, a workshop to equip a team, a seminar to frame an editorial transformation. When the stake bears specifically on AI, our training method takes over in a longer format.
A talk that works leaves behind a method, a frame, and one sentence the room takes away. Ours returns in every intervention we give: most content problems are frame problems.
Frequently asked questions
What is the difference between editorial strategy, governance and steering?
They are three distinct levels, and the order matters. Editorial strategy decides the destination: who you speak to, on which subjects you are legitimate, what you refuse to say. Governance installs the rules, the roles and the approvals that produce coherence without depending on individual judgement. Steering measures and corrects: what gets published, what gets read, what gets cited, and what you decide to stop. Skipping the middle level produces authorial drift, and its cost appears on no budget line.
What has generative AI changed for editorial governance?
Two things. The bottleneck has moved: for twenty years it sat in production, it now sits in the decision. A marketing department can feed every channel without having settled what its brand is entitled to say, which yields more volume and less signal. Second, generative engines return brands where they used to rank pages. A brand whose language changes from one channel to the next collects the average answer of its sector, the one in which nothing distinguishes it.
How do you know whether a governance holds?
By one test: it survives the departure of the person who wrote it. Beyond that, it is judged on four grounds, in this order of dependency: clarity of the offer, a documented voice, a steered editorial system, authority. Governance is judged by what it makes possible, not by the volume of documents it produces.