Producing at scale requires brand governance at the same scale

Trust is already the main barrier

The use of generative AI is widely used by major advertisers, but it is accompanied by an explicit concern about how it is used: 63% of the major brands surveyed report using it in marketing, and 80% say they are concerned about how their agencies use it. The risks cited are legal, ethical and reputational.

WFA · Annual Report 2024 · 2024 · WFA member advertisers — View source

The mechanism is scale-specific: an isolated error is corrected, an error reproduced on thousands of content and points of contact becomes a systemic risk.

Controversies often reveal an error in brand judgment

Mediated cases are not first of all technical failures. These are poorly calibrated brand decisions, which the automated production has made visible faster and more widely. Two sufficiently documented cases illustrate this.

Google, "Dear Sydney" (2024). The announcement of the Olympic Games was withdrawn after a strong rejection by the public, which found that it proposed to delegate an emotionally significant gesture to AI. The difficulty was not with the quality of execution but with the choice of territory.

The Verge · Google sweaters its Gemini Olympics ad after backlash · 2024 — View source

McDonald's Netherlands (2025). A Christmas advertisement entirely generated by AI was withdrawn after a massive rejection, even though its production met the expected standards. The problem was the difference between the brand's expected register at that time of year and the product register.

The Drum · McDonald's sweaters AI ad · 2025 — View source

In both cases, the tool did not produce a factual error. It faithfully executed a trademark decision that should not have been made.

AI does not automatically cause a loss of quality

The counter-example is necessary in order not to confuse the tool with the use. Holidays Are Coming Coca-Cola generated a creative backlash, but got 5.9 stars at the System1 test, its maximum score, thanks to the preservation of the brand's historical codes.

Kantar · Rethinking AI-generated advertising · 2025 — View source

System1 · Nation's Favourite Ads 2024 · 2024 · advertising test, consumer panel — View source

Unilever followed the same logic by endorsing its Beauty AI Studio to Brand DNAi, its global brand governance framework.

Unilever · Annual Report 2025 (Brand DNAi) · 2025 — View source

It is therefore not the use of AI that determines the outcome, but the ability to maintain brand codes and make them operational in production.

The brand must become an operational layer

This is the displacement that BCG describes as brand intelligence layer : purpose, rules, identity, context and decision criteria encoded in the systems themselves rather than transmitted to humans by a document.

BCG · Making the agent marketing transformation a reality · 2026 — View source

1. 01
   Brand historical bookPrinciples written to be read, interpreted and applied by people.
2. 02
   Brand intelligence layerIdentity, rules, references, constraints, decision criteria and validation points, expressed in a form usable by a system.
3. 03
   Implementation systemThese rules apply in production, control and publication workflows.

The brand must become machine-readable and operationally manageable, without becoming fully automated. Encode a rule does not exempt it from deciding when it applies.

The second risk is not error, but loss of distinctiveness

BCG warns against what it calls the scaling sameness : systems built on the same models and patterns produce similar outputs through thousands of interactions.

BCG · Making the agent marketing transformation a reality · 2026 — View source

The risk is more insidious than the bad buzz, because it does not trigger any alert: even without factual error, a generic production repeated on a large scale gradually reduces the distinctiveness of a brand. Agents without organisational identity diffuse the same correct — and interchangeable — response.