Episode 18 · Article on the episode

Coca-Cola versus Telekom: why perfectly generated images feel cold

Two Christmas adverts, two methods, two reactions. The comparison works well beyond the holidays as a test of where generated content carries and where it tips over.

By Mark Zimmermann · 22 Dec 2025 · 4 min read · Auf Deutsch lesen

This special episode has an AI tell four Christmas stories from Father Christmas's perspective. Between the fireside kitsch sits a comparison that works as a case study.

The comparison

Coca-Cola's AI-generated Christmas advertising caused criticism online: soulless polar bears, jerky penguins, cold images without warmth. Against that stands Deutsche Telekom's advert, produced entirely without AI, with a palpable result.

The thesis derived from it goes: perfectly generated images often feel cold precisely because they lack the imperfect, the human.

The thesis is pointed and merits a refinement, because otherwise it is misunderstood as scepticism about technology.

Where generated images tip over

Movement is the weak point, not the single frame. A generated still image passes every check today. Moving images give themselves away in the intermediate steps: a penguin moving jerkily violates an expectation that nobody consciously formulates and everybody has.

Familiarity sharpens that. Polar bears and penguins from a decades-old campaign are known to the audience precisely. The more familiar a motif, the smaller the deviation that stands out. With an arbitrary motif the same deviation does not stand out.

The occasion has a say. With a product image nobody expects warmth. With a Christmas campaign, warmth is the product. Where feeling is the message, the method becomes part of the message.

The usable rule is therefore not “no AI in advertising”, but: the more a message rests on closeness and the more familiar the motif, the more expensive every visible deviation becomes.

The case is also instructive because the costs here did not arise in production but afterwards. An advert that provokes objection costs more than the saving in production brings in.

The technical part in between

Between the stories, both hosts discuss whether the much-quoted pilot phase of AI is really over, as the consultancy phrase has it, or whether companies are only now grasping what is coming towards them with agentic systems.

Mentioned in that context is work on latent collaboration, in which models apparently conduct internal negotiations of roles that cannot be directly observed. That is a further building block in the black-box character of today's language models and an argument against the widespread assumption that visible reasoning shows what is actually happening.

Also placed in context is the competitive situation between OpenAI and Google Gemini, including the reported code red at OpenAI, while Google stands there less dependent on Nvidia chips thanks to its own hardware.

The friendly side

The third story shows AI as a quiet helper with Christmas stress, from travel planning to the search for presents. Unspectacular and viable for exactly that reason: tasks where nobody expects warmth, but relief.

The fourth and most touching story is about the British organisation Action for Children and its magic elf mirror called 11.ai. An example of how technology can make donations more personal and more effective instead of replacing them.

The difference to the Coca-Cola case is revealing. Here the technology replaces no human gesture, it carries one. The feeling comes from the organisation and the donors, the technology makes it deliverable.

Conclusion

The sentence the episode ends on works well beyond the holidays: technology can support Christmas, but not replace it. Where the heart is missing, even the best generation stays superficial.

For practical application this can be put more sharply. Check two things before any generated content.

First: is closeness the message, or is it information? With information the method is a matter of indifference. With closeness it becomes part of the statement.

Second: does your audience know the motif precisely? The more familiar, the smaller the deviation that stands out, and the more expensive the failure.

And calculate the costs in full. The saving in production is known, the cost of a debate about your brand is not.