We help marketing and communications teams structure their editorial thinking before they structure their AI. That is what we stand for in our training programmes. Client work since 1991. 195 companies. Dozens of briefs unpacked; validation committees run and challenged; offer narratives stress-tested in commercial meetings; brand voices deployed across five channels simultaneously — and lost along the way because no one had set the shared rules.
At some point, the patterns become obvious. The same blockages come back. The same steps are missing. The same deliverables keep failing.
At WeAreTheWords, we formalise. It is our signature, our method, our conviction.
Which is why I am proud to present our three training programmes in editorial engineering and AI: Offer Architecture, Voice Architecture, Editorial AI Assistant — a progressive curriculum built on the logic of a governed editorial infrastructure.
| Programme | What it stabilises | The central deliverable | Without it |
|---|---|---|---|
| Offer Architecture | The discursive framework of your offer | A Clear Offer Framework in 4 steps: formulated promise, named mechanism, narrative structure, integrated proof | The AI rewrites your offer ten different ways; none carries authority |
| Voice Architecture | The brand’s editorial identity | A Brand Voice Framework across 9 operational pillars | The AI produces in the generic voice of your sector, not yours |
| Editorial AI Assistant | Governance of AI content production | An assistant configured with your voice and offer, tool-agnostic | Editorial quality depends on one person, and disappears when they leave |
Before LLMs, this debt was costly. With LLMs, it is catastrophic. AI amplifies what it receives. Feed it a vague offer and it produces industrialised vagueness.
Why AI training without editorial strategy is not enough
Most marketing teams do not lack content. They lack consistency across that content. The website says one thing. The sales deck says another. The marketing manager’s LinkedIn post says a third. No one is lying. Everyone is interpreting. And the offer dilutes with every interpretation.
This has a name in our method: editorial debt. It accumulates silently, mission after mission, campaign after campaign — until validation cycles stretch, briefings multiply, and the brand can no longer introduce itself in twenty seconds without hesitation.
Before LLMs, this debt was costly. With LLMs, it is catastrophic. AI amplifies what it receives. Feed it a vague offer and it produces industrialised vagueness. Feed it an unstable voice and it produces dilution at scale. The model does not correct the foundations. It exposes them, without repairing them.
That is the starting point of the WeAreTheWords triptych. Not “how to use AI”, but “on what editorial foundations are you running your LLMs?”
Offer Architecture: the foundation of every engagement
On every Clarity engagement, we start by listening. To the documents, but also to the people. How they present the offer in a meeting. How they answer “what exactly do you do?” The words they use spontaneously. The hesitations. The formulations that shift depending on who is in the room.
What we hear reveals everything. An unstable offer narrative does not show in the slides. You feel it in the room.
The question is never “is your offer good?” It is: can your offer hold in one firm sentence, without over-explanation, without conditions, without variation depending on who is carrying it? Can it survive a pass through an LLM without distortion?
Offer Architecture is the formalisation of this diagnostic into four steps drawn from our client work: identify the formulation breaks, stabilise the core sentence, name the central mechanism, deploy without dilution. Participants do not leave with a marginally improved promise. They leave with a narrative framework stress-tested in the field.
The Clear Offer Framework GPT is integrated into the programme as a testing tool. Submitting an offer promise to an LLM and observing what it does with it is one of the most revealing exercises we have developed.
What participants build: a formulated and tested promise, a clearly named mechanism, an intelligible narrative structure, proof integrated into the message, discursive standards that can be handed to the team. An asset. Not a project deliverable.
Explore Offer Architecture and upcoming sessions
Voice Architecture: what you build once the offer holds
A clear offer is not enough. It can remain inaudible if it is expressed in an interchangeable voice.
What we observe on Growth engagements: teams that have been producing with AI for months have abundant, uniform content. The volume is there. The distinctiveness has gone. Texts look alike because LLMs optimise towards the statistical norm of the category — yours included.
Brand voice is not a stylistic antidote. It is an infrastructure of constraints. It narrows the model’s probability space by setting precise rules: this lexicon rather than that one, these archetypes rather than those, these assumed narrative tensions rather than their erasure in favour of smooth consensus.
The Brand Voice Framework across 9 pillars was born from this necessity. Each pillar corresponds to a resistance point identified across our client work: brand statement, values, personality, positive and negative character traits, archetypes, emotions, brand and meaning narratives, language register. Nine pillars because that is the minimum for a voice to be governable. Fewer, and it remains too abstract to be transmissible. More, and it becomes impossible to maintain.
Over one day, participants activate the 9 pillars on their own brand, test their voice against an LLM used as an editorial mirror, build their narrative cores, and integrate the framework into their internal GPT. AI is not the subject of the programme. It is the test bench.
One participant — an editorial director at a Belgian media group — named the issue with a precision we could not have improved: “Narrative tensions are a card to play. We never give enough time to strategy.” That is exactly what this programme resolves.
Explore Voice Architecture and upcoming sessions
Editorial AI Assistant: what you deploy once the foundations are in place
When a brand has a clear offer and a structured voice, configuring an editorial AI assistant becomes an act of governance. Nothing to do with a prompt engineering exercise.
Without these foundations, the assistant is a noise amplifier. With them, it becomes a transmissible asset: a system that produces in the brand’s voice, respects the validated vocabulary, excludes prohibited formulations — regardless of which team member uses it, regardless of which LLM it runs on.
This is the third lesson from our client work: editorial quality cannot depend on a single person. The day the content manager who “knows how to prompt” leaves the team, the system goes with them. Editorial governance means inscribing the rules into the system, not into individuals.
The programme is structured in four steps drawn from our internal workflows. First, an AI editorial maturity diagnostic: which use cases, which scope, which voice is available, which knowledge base needs building. Then, translating the brand into operational rules: system instructions, semantic blacklist, archetypes and tone registers. Next, configuration and deployment: versioned knowledge base, settings, internal data protection, first test in real conditions. Finally, a testing and maintenance protocol: a four-point review framework, internal editorial standard, long-term evolution plan.
The programme is LLM-agnostic. ChatGPT, Claude, Gemini: the protocol adapts to all models. Because dependence on a single tool is a governance risk. What we teach is the architecture logic, not the mastery of one interface.
What participants take away at the end of the day: an assistant configured with their voice and offer, an operational semantic tag, a framework shareable with the team, a documented review protocol. Not a technique. A replicable, durable editorial asset.
Explore the Editorial AI Assistant programme and upcoming sessions
Why this sequence, and not another
The triptych follows a structural logic: the offer is the foundation, the voice is the framework, the assistant is the distribution system. An editorial assistant configured with a non-existent voice reproduces incoherence at scale. A voice built on a vague offer produces an identity dissociated from the commercial narrative. You do not put the roof on before the walls.
That said, the curriculum is not a sequential constraint. Some participants arrive with an already stabilised offer and a formalised brand voice. For them, the Editorial AI Assistant programme is the natural entry point: they have the foundations, they come for the governance system that perpetuates them in the LLM era.
Others arrive with the sense that something is stuck upstream. That their AI assistant keeps producing generic content despite their parameterisation efforts. That the problem is not the tool. The full triptych is built for them: Offer Architecture first, Voice Architecture second, Editorial AI Assistant to deploy what has been built.
In both cases, the objective is the same: a Content Operating System. A governed, documented, transmissible editorial system, resilient to the dilution that LLMs accelerate.
Each programme stands alone. Each programme builds on the previous one. Together, they form what WeAreTheWords calls a Content Operating System: a governed, documented, transmissible editorial system, resilient to the dilution that LLMs accelerate.
Who this AI and editorial strategy curriculum is for
These programmes are for those who govern content, not only those who produce it.
- The content manager using AI daily who notices that tonal consistency erodes as production accelerates.
- The consultant who needs to deliver an editorial system to a client and is looking for a documented, reproducible, defensible method.
- The marketing director whose team produces at volume but whose brand remains inaudible in its sector.
- The communications manager whose offer message shifts depending on who is carrying it.
In every case, the problem is not the tool. The problem is the absence of an editorial framework upstream of the tool. That is precisely what the triptych builds.
The programmes are available separately. They are designed to be taken in sequence. Participants who complete the full curriculum leave with three interdependent assets: a stabilised offer, a governable voice, an assistant that perpetuates both.
Your content deserves more than an editorial calendar. It deserves a system.
FAQ
In what order should the three programmes be followed?
The offer first, then the voice, then the assistant. An assistant configured with an unstable voice reproduces incoherence at scale. Participants who arrive with a stable offer and a formalised brand voice can start directly with the Editorial AI Assistant programme.
Which AI tools does the Editorial AI Assistant programme work with?
It is LLM-agnostic: ChatGPT, Claude and Gemini. The method covers the architecture of the assistant, not one interface, because depending on a single tool is a governance risk.
What do participants leave with?
A deliverable built on their own brand: a tested offer framework, a Brand Voice Framework across nine pillars, or an assistant configured with their voice and offer, with a documented review protocol.