Coca-Cola and Meta let AI make (almost) all the calls: here is what went wrong.
The cases of Coca-Cola, Meta, and Toys R Us demonstrate an absence of governance rather than AI's inability to perform in creative marketing. The key distinction is between an operational decision (which tools to use) and a strategic decision (where AI can operate autonomously and where human oversight is required). Brands that avoid backlash have embedded this distinction directly into their system architecture.
– Coca-Cola delegated emotional content to systems incapable of handling emotional nuances; Meta removed human oversight from creative assets; Toys R Us entrusted its origin story to systems with no access to brand identity
– Spotify, Netflix, and Amazon deploy AI where data is abundant and errors are reversible; they maintain human oversight where errors cause long-term reputational damage
– The fallout from LLM responses to millions of brand and product queries doesn’t make headlines, but it silently erodes positioning over time without anyone measuring it
The premise was straightforward: AI would make marketing content production faster, cheaper, and more scalable. Instead, cases disproving this assumption are mounting.
Coca-Cola invested in a holiday campaign produced entirely with AI. Technically, it hit all the right notes: snow, red trucks, a familiar atmosphere. The public rejected it. Meta allowed Advantage+ to autonomously swap brand ads with algorithmically generated creatives, resulting in grandmothers promoting menswear, distorted limbs, and flying cars in campaigns completely unrelated to the actual product. Toys R Us used Sora to tell its origin story. The result felt more like a technical demo than a narrative.
These three instances are often cited as proof that AI fails in creative marketing. This interpretation is flawed and misses the actual issue.
Coca-Cola used AI to generate emotional content without evaluating whether AI systems could comprehend emotional nuances. Meta eliminated human oversight from the creative decision-making chain. Toys R Us delegated identity-driven storytelling to a system that lacked access to the core brand identity.
While these appear to be three distinct mistakes on the surface, they share a single root cause: no one defined where AI could operate autonomously and where human control was required. This is an absence of governance.
What AI governance means for a brand
AI governance in marketing addresses a specific operational question: which brand decisions can be delegated to automated systems, and which require human oversight?
The answer lies on a spectrum, which brands leveraging AI without facing backlash have successfully navigated.
Spotify uses AI to build personalized playlists without editorial oversight for every single recommendation, yet retains human curators for culturally significant moments. Netflix uses AI to determine which thumbnail to display to which user, but does not delegate content production decisions to algorithms. Amazon personalizes its homepage algorithmically without ever framing it as a core product feature.
In each of these cases, governance is embedded into the system architecture: AI operates where data is sufficient and errors are reversible. Humans maintain oversight where errors result in reputational damage that is difficult to repair.
The brands that faced backlash did the exact opposite: they delegated high-risk reputational decisions to AI—scenarios where an error becomes public and nearly impossible to justify.
The risk that doesn’t make headlines
There is another dimension to this issue that rarely enters the conversation.
AI systems represent brands whenever consumers ask questions about them. The mechanism works like this: a user asks ChatGPT, Perplexity, or Gemini for information about a brand; the system constructs a response by pulling from public sources that the brand does not directly control; the brand is then described, compared to competitors, positioned within a price bracket, and associated with specific values. All of this happens before the consumer ever visits the brand’s website, reads a review, or sees a campaign.
Traditional brand management focuses on what the brand produces and distributes. However, governance in the era of answer engines requires monitoring how the brand is represented by third-party systems that Geist information on behalf of users.
The backlash seen in the Coca-Cola and Meta cases is highly visible: pulled campaigns, critical threads, and press coverage. Conversely, the damage caused by LLM responses to millions of queries regarding brands, products, and services doesn’t make headlines. It silently erodes positioning over time without anyone measuring it.
The distinction that matters
Deploying AI in marketing is an operational decision. Governing AI in marketing is a strategic decision.
The former addresses the question: which tools will help us produce content faster? The latter addresses far more complex questions: which AI systems are already talking about our brand? What are they saying? In what context are they placing us? What information about us is distorted, incomplete, or missing?
Brands that treat AI purely as a production efficiency lever are optimizing the visible surface of the problem while the hidden risks grow. Brand AI governance requires active management on both fronts: controlling what the brand produces using AI, and managing how the brand is represented by AI.
Those responsible for market positioning must address both. It is a matter for the board.
