The draft was fast. Approval was slow.
The five-person team was publishing in English, French and Arabic, and AI had made the first draft almost instantaneous. Approval still took too long because editors inherited generic arguments, unsupported claims and language that sounded translated. Measuring generation speed concealed the work being transferred to the next person in the process.
A reusable editorial system
Each request specified the audience, buying stage, commercial purpose, message, action and proof required. Approved product information and research were separated from creative direction, so headline exploration could not quietly become a new factual claim. Reference material supplied the knowledge the model should use rather than leaving gaps for it to fill.
Reusable prompt versions captured those requirements for recurring asset types. The team could change a prompt deliberately and retain its improvements instead of relying on an individual writer’s private technique.
Make the feedback improve the next draft
English, French and Arabic each had guidance on terminology, tone and local expression. Reviewers checked the source facts and whether the text sounded natural to the intended reader.
Repeated corrections became updates to the brief, reference material or prompts. An editorial owner kept those changes current, so the workflow learned from review instead of repeating the same mistakes.
Give the draft boundaries before asking it to write.
The brief defined the audience, source facts, tone and approval owner. English, French and Arabic shared the same purpose without being forced into the same phrasing.
| Input | Treatment | Editorial check |
|---|---|---|
| Product facts and approved research | Supply the relevant source; do not let the draft fill an evidence gap | Can each material claim be supported? |
| Commercial brief | Fix audience, decision, message and desired action | Does the asset serve the intended buyer stage? |
| Language guidance | Adapt terminology, syntax and register to the language | Does it read naturally beyond literal accuracy? |
| Creative expression | Allow alternative structures and angles within the brief | Is the argument useful and recognizably on brand? |
Speed mattered at approval, not at generation. The editor still owned the final claim and local expression.
Language quality and ownership
The prompt system needed consistency in facts and terminology without prescribing the same introduction, argument and conclusion for every asset. Otherwise the team would exchange one kind of rework for another: accurate content that all sounded alike.
Human editors retained responsibility for factual accuracy, local expression and publication. Restricted information belonged only in approved tools, and recurring corrections fed back into the reference material or prompt version rather than remaining an individual writer’s workaround.
Eight-week production results
Across eight weeks, brief-to-approved-draft time fell from six hours to three and a half, approximately a 42% reduction. First-review acceptance rose from 42% to 68%, while every published asset retained human approval.
The eight-week pilot measures production efficiency and editorial acceptance. Neither figure establishes that the resulting content generated more pipeline; that requires separate performance measurement. The useful operating change is a faster approved asset, with the publishing decision still owned by a person.
The useful saving was editor time, not generation time. A team can produce more approved content only when the brief, reference material and review process improve together.