56% of people let AI decide which brand message deserves attention.
Not the customers themselves. The algorithm, which to do so will summarise, select and rewrite.
In short, the brand no longer speaks: it hands the floor to the machine. With consequences we can only imagine. Let’s find out more.
The study
The study The Preference Economy: Earn True Loyalty or Get Filtered Out, published by GALE on a sample of 3,000 consumers in the United States and the United Kingdom, revealed an alarming shift in the way purchasing choices are made.
The figure is this: 56% of consumers say they are comfortable letting an AI system filter a brand’s communications entirely.
I’ll repeat it for anyone who missed its full significance: one person in two already lets AI do the choosing. And almost one person in three has already used ChatGPT, Gemini or Copilot to prioritise certain brands over others.
In short, the brand’s message no longer arrives pure but mediated, summarised. Perhaps even rewritten by an intermediary the brand did not choose and probably does not even control.
Brands are losing control
For over a century, governing a brand meant controlling what was said, how, and in what sequence that message would reach the audience. Copywriting, tone of voice, the very structure of a press release: all levers in the hands of those managing communications.
But if an AI system steps between the brand and its audience, the chain breaks: the language model reads the brand’s original material, extracts what it deems relevant according to its own summarisation criteria and gives the user a mediated version. The tone may change. Nuances may be lost. Information the brand considered central may be reduced, moved or omitted altogether. It is the definitive loss of narrative sovereignty.
Naturally, the impact varies by sector.
What to do NOW
Trying to control the algorithm would be impossible. No organisation, however large, can do it, since the reprocessing is in the hands of third parties.
No, the only answer is to change what is made available to the AI filter.
Let us start with consistency.
Consistency is the real weapon of the agentic era: because an AI model summarises material that is already consistent better, and more faithfully. If the same information is communicated in different ways across different channels, the system synthesising it has more room to choose which version to favour, and that choice is not in the brand’s hands.
Second, depth. Precise data, certifications and technical specifications always survive a language model’s summarisation process. A vague claim or a purely emotional tone, by contrast, is the first element a summarisation system tends to compress or eliminate, because it carries information that cannot be measured.
The voice of tomorrow
And then there is a value that cannot be filtered: community.
Direct relationship programmes, genuinely listening to people: these actions create and sustain a relationship that no AI filter can touch. This is essential to remember.
After all, the loss of narrative control described in this research is already the norm for many brands trying to communicate today. And for all of them, continuing to write as if every word reached the end reader intact is a blind illusion.
Today, only the brand that builds its communications knowing that much of them will pass through a filter it does not control, and that prepares for that mediation with consistency and specificity, will keep a recognisable voice even when that voice arrives mediated.
New Connections (FAQ)
How can we check whether an AI is already filtering or summarising our brand’s communications?
The most direct way is to query the same assistants the public uses, ChatGPT, Gemini, Copilot, asking them to describe your brand or compare it with competitors. The answer the system gives shows exactly which version, often simplified or partial, is already circulating among customers asking similar questions. Repeated periodically, this exercise is now a form of reputation monitoring that no brand should ignore any longer.
Is investing in Generative Engine Optimization enough to solve the problem of algorithmic mediation?
It is part of the answer, not the whole answer. GEO helps a brand to be found and cited correctly by language models, making its material more readable and structured for machines. But it does not solve the deeper problem of summarisation: even perfectly optimised content can be summarised, simplified or taken out of context when the model returns it to the user. GEO improves the chances of being cited. The consistency and specificity of the source material determine how faithfully that citation will reflect the original message.
Younger customers seem less concerned about this loss of narrative control. Is it a real problem or merely a perception of those who manage the brand?
This is a concrete finding, and GALE flags it explicitly: among the younger consumers surveyed, the most common response to the question about the privacy of their preferences was not hesitation, but the fact that they simply had not thought about it. This does not mean the problem does not exist. It means that younger audiences perceive it as normal, not as a loss. For a brand, this changes the urgency of its communications strategy, not its necessity: algorithmic mediation happens regardless, whether or not those subject to it perceive it.