Coca-Cola is restructuring its agency roster while reshaping how marketing work is briefed and produced, and the two changes are connected. As generative tools take over first draft production, the brief becomes the highest leverage document in the chain, because it is the artifact that both humans and models read. Brands that treat the brief as an internal memo will find their content drifting, and their citation footprint shrinking with it.
The Roster Reshuffle
A roster reshuffle usually signals a shift in what the brand believes it needs to buy. Consolidating agencies reduces coordination cost and gives the brand a single point of accountability for creative output. Splitting the roster by capability, on the other hand, buys specialization at the cost of integration. Coca-Cola appears to be doing a version of the second, keeping specialist partners while centralizing strategy.
The AI dimension changes the calculus. Coca-Cola is reshaping marketing as its agency roster changesMarketing DiveWhen first draft production moves in house or into a tool, the agency value shifts toward strategy, taste and governance. That is a different contract, priced on judgment rather than on output volume, and it is the reason roster changes and AI adoption tend to arrive together.
Briefs That Machines Read
Structure Over Prose
A brief that a model can use looks different from a brief written for a creative team. It states the audience, the single message, the proof points and the constraints in discrete fields rather than in flowing paragraphs. The structure is not bureaucratic decoration: it is what allows a model to hold the constraints constant while generating variations, and what allows a reviewer to check compliance without rereading everything.
The practical benefit shows up in review time. When the brief carries explicit proof points with sources, the reviewer checks the citation rather than the claim. That reduces the back and forth that consumes most agency hours, and it produces content that is easier to defend when a claim is challenged publicly. Structure is what makes a claim auditable.
Agency Economics Under AI
Agency economics have rested on billable hours and on production volume. Both are under pressure. If a tool produces twenty variants in an hour, charging by the variant is untenable, and charging by the hour punishes efficiency. The models that survive are retainer plus performance, or a fixed fee for governance and a variable fee for distribution.
The transition is uncomfortable because it changes who owns the risk. Under a volume model, the agency absorbs rework. Under a governance model, the brand absorbs the cost of a bad output and the agency is paid for the process that should have prevented it. Marketing DiveContracts that name the review gates explicitly, rather than describing them generally, are the ones that survive the first serious dispute.
Brand Voice and Citation
Gates by Risk Level
Brand voice becomes a data problem once content is generated at volume. A voice guide written in adjectives is not executable; a voice guide written as paired examples with accept and reject labels is. The paired example approach also gives a model something to imitate rather than something to interpret, which reduces variance across outputs and across teams.
Citation is the downstream consequence. When a model answers a question about a category, it cites the sources that are structured, specific and attributable. A brand whose content is generated without proof points produces text that reads well and cites badly. Marketing DiveThe measurable outcome is a declining share of voice in AI answers even while paid reach holds steady, which is a confusing signal until the mechanism is understood.
What Other Brands Should Copy
Three Moves Worth Stealing
Three moves are worth copying. First, rewrite the brief as a structured artifact with explicit proof points and source links. Second, define review gates by risk level, so low risk content moves fast and high risk content gets human sign off. Third, measure citation share alongside reach, because the two now diverge and only one of them compounds.
The order matters. Brands that start with tooling end up automating a process they never fixed. Brands that start with the brief and the gates get a compounding advantage, because every subsequent tool plugs into a process that already produces auditable output. That sequence, not the tool choice, is what separates the brands that gain from the ones that merely spend.
