Advertising is beginning to face an unprecedented circumstance: part of its audience is no longer human. Before a person considers, recommends or buys a brand, an artificial intelligence model may have tracked, interpreted and ranked the information available about it to decide if it deserves to be part of a response.
This introduces a second level to advertising effectiveness. Brands still need to excite, generate memories and build meaning for people, but this universe of messages, content and conversations is also being processed by machines that interpret communication with very different criteria.

Two investigations presented around Cannes Lions 2026 help to measure this difference. The “Creative ReCode” study, by Kantar, analyzed 786 ads in 48 markets and compared their effectiveness among people – based on six dimensions: engagement, branding, message, persuasion, meaning and differentiation – with their readability for machines. The correlation between both evaluations was only 0.10.

Research by Jellyfish and INSEAD on 480 works submitted to Cannes Lions reached a similar conclusion: humans and language models showed profoundly different evaluations of the same creative pieces.

The question that opens up for marketing is important. If AI becomes increasingly important as an intermediary between brands and consumers, Should we also start advertising for the machines?

Machines want explanations

Research points to two practically opposite ways of interpreting creativity.

Language models operate with what David Dubois, Associate Professor of Marketing at INSEAD, defines as a “narrow beam”. They favor product details, specific features, clear statements, functional explanations, recognizable structures and descriptive language that allows one option to be compared with another.

Human beings work differently. They are able to interpret what is not said, understand metaphors, recognize cultural references, complete narratives or react to humor and emotional intensity. Much of what makes an advertising piece memorable lies precisely in that implicit territory.

The divergence is especially pronounced in categories such as luxury, where much of the value of communication is constructed through codes, symbols and meanings that do not need to be made explicit.

Tom Roach, Vice President of Brand Strategy at Jellyfish, summarizes the new scenario by pointing out that “The future of creative effectiveness depends not only on what excites people, but also on what survives ranking, retrieval and recommendation by machines.”

But the conclusion of the studies does not seem to lead towards advertising designed primarily to satisfy the algorithms. Quite the opposite.

The brand appears as the common territory

If humans and machines process ads differently, there is a point where both systems can meet: a clear, consistent and differentiated brand perception.

Kantar proposes that this coherence must be built around explicit product benefits. Communication can continue to be emotional, creative or culturally relevant, but the relationship between what the brand promises, what it demonstrates and what it wants to be recognized for must be unequivocal. “It’s about bringing brand ideas to life in a consistent and coherent way throughout the ecosystem”explained Dom Boyd, Director of Customer Strategy at Kantar. The aspiration would be to build “a brand operating at scale with flexibility and contextual relevance.”

Boyd uses the image of Velcro to explain it: a single brand idea with numerous small attachments capable of adapting to different channels, formats and contexts without losing what holds the whole together.

Kantar points out three ingredients to achieve this. The first is clarity of perception: brands with more consistent perceptions register 111% higher growth, and being large does not necessarily imply having a clear positioning. The brand idea, the research points out, should be able to be condensed to fit on a refrigerator magnet.

The second is to be culturally connected, participating in the conversations and forces that shape the category. Here creators become especially important due to their ability to translate product benefits into relevant codes.

The third is the connection of the ecosystem: what a company says must be related to what others say about it. And this last element is especially relevant when the person interpreting that conversation is a machine, also taking into account that 63% of visibility in LLMs comes from long-term brand building.

This idea finds support in other research, collected by Warc, which has gained prominence during 2026. The British agency Charlie Oscar analyzed 30 brands belonging to six categories for eight months to study what factors explain their presence in the responses generated by language models. The result places the historical construction of the brand far above the most immediate factors.

“Two-thirds of visibility in LLMs is based on things brands did years ago”says Dan Wilson, co-founder and Chief Data Officer of Charlie Oscar.

The data introduces a particularly relevant reading given the growing interest in AEO – optimization for response engines -: the presence in artificial intelligence does not necessarily begin by optimizing content for ChatGPT, Gemini or any other system. A substantial part of that presence is the accumulated consequence of years of building notoriety, differentiation and conversation around the brand.

This also poses a difficulty for emerging brands. Large, established companies start with the advantage of heritage built over years, while brands challenger They need to generate enough external signals to enter the set of options that the models consider relevant.

What the brand says and what others say

The models do not only serve official communication. They seek consensus: reviews, recommendations, editorial references, user conversations and other signals distributed over the Internet contribute to building the representation that an AI ends up forming of a company.

Hence, Natasha Wallace, Chief Solutions Officer of Jellyfish, suggests that the objective should not be reduced to achieving visibility in searches, but rather to working on brand perception: “It’s not just about what you say, but how you activate what other people say about your brand to reinforce preference within the model”he explains.

In this context, distinctiveness retains all its importance. If two competitors appear before a model with practically interchangeable attributes, the AI ​​has little reason to favor one over the other. Distinctive assets, a recognizable proposition and a brand platform sustained over time thus acquire a second function: in addition to facilitating human recognition, they help machines understand what makes a company different.

The research also introduces nuances around one of the brands’ current big bets: social-first strategies.
Kantar warns that a strategy designed primarily for social platforms does not necessarily equate to an effective strategy for artificial intelligence assistants. Organic content and its ability to circulate around the web are particularly important.

According to the analysis, creator content published on YouTube appears much more frequently in LLM citations and results than content from TikTok, Meta or Instagram. However, only 11% of the creator content analyzed includes a proper link structure.
There is also a technical problem: much short-form vertical video can perform perfectly in front of a human audience while remaining virtually invisible to web crawlers used by artificial intelligence engines.

Kantar also points out that platform engagement and brand building only coincide in about a third of cases. The ability to generate interaction within a social network, therefore, does not guarantee that this activity ends up strengthening the representation of the brand outside of it.

How to advertise to two audiences

The temptation might be to fill the communication with explicit claims, keywords and product details to facilitate its algorithmic interpretation. Research advises against precisely this drift.

Optimizing primarily for machines can produce perfectly readable and barely memorable advertising. Kantar considers algorithmic readability “a nice extra”, but not the central objective of the briefing. His proposal for working before the two audiences is articulated around five principles:

• Make a clear promise
• Explain a relevant benefit
• Build a simple story aligned with that benefit
• Combine demonstration and confidence
• Keep the idea focused

Some campaigns show how this coexistence can be resolved. Kantar cites the advertisement “The Residue” by Finish Jet-Dry, where the visual narrative and the rational explanation of the product move in the same direction. The machine finds a recognizable benefit and a clear explanation of why it occurs, while the person receives an advertising story built around that same argument.

The question, therefore, seems less related to creating a new advertising for algorithms than to recovering a classic brand-building discipline: having something relevant to say, making it recognizable and sustaining it with enough coherence so that both people and machines can unequivocally associate it with the brand.

And that conclusion becomes especially relevant in the heat of the race to appear in the responses of AI assistants. Optimization can modify some of the present visibility; But, according to Charlie Oscar’s data, most of it is being played in a territory that is much more difficult to manufacture overnight: the brand that a company has spent years building.