“Thinking you can specifically optimize the brain of an LLM means nurturing an illusion of control”, Britney told us
In the debate around artificial intelligence, the noise is often louder than the substance.
Promises pile up on top of one another, narratives overlap and create nothing but confusion, while the rules of the game change without asking anyone’s permission.
To understand everything new that is moving in search marketing, we need voices capable of reading change without oversimplifying it, of telling real evolution apart from its representation, beyond slogans and ads.
Britney Muller is one of those voices. An international AI professional with over ten years of experience across machine learning and generative AI, she led the launch of BLOOM, the largest open-source, multilingual language model, taking the AI conversation out of the labs and into the global public debate.
Her work is about translating complex systems into understandable tools for marketers, entrepreneurs and leaders who need to act, not just understand. That is exactly why, with the Orange Labs community, she has built a space for ongoing exchange, founded on a conviction that is simple and radical at the same time: AI only has value when it is adopted in a conscious, practical and sustainable way.
This episode of SEO Confidential starts from here. From the need to go beyond the hype, beyond the labels, to observe what is really changing. And what is not.
Here is what Britney told us about all these topics, enjoy the read!
“There are no private conversations with an LLM: we are all just unpaid workers for AI companies”
Hi Britney, thank you so much for accepting our invitation. First of all I would like to ask you: Google sold us AI Mode as a major step forward, but many signals point to unstable results, inconsistent citations and significant variability between mobile and desktop devices.
Are these effects an attempt to redistribute traffic, or a symptom of an immature model generating random benefits?
Thank you Roberto, and thanks again for thinking of me. Well, as we know, when Google embedded LLMs into AI Mode, it introduced a completely new technology into search. LLMs are probabilistic, not deterministic. (Given the same input, they can generate different answers, precisely because they are based on statistical computations.)
This variability is not a bug or a limitation of the system. On the contrary, it is the very principle their operation is built on, allowing the model to adapt to context, reformulate information and produce more flexible and nuanced results than the classic mechanisms of search.
The hype surrounding AI is fueled mainly by aggressive communication strategies, often very far removed from what the models can actually do. How much weight, in your view, does this narrative carry in how the market and the public perceive AI?
We are currently in a phase where companies market these tools as “almost AGI” (Artificial General Intelligence), aiming above all to strengthen market perception, to drive up stock prices.
Meanwhile, the most solid value is being built elsewhere: among the marketers and professionals using Python to analyze data, structure content strategies and automate operational workflows (the ‘boring’ stuff, in short). It is these practical applications, often barely visible and far from the spectacular narrative, that form the foundations of what is destined to last over time.
The collapse of organic traffic is hitting almost everyone, even when the quality of the work is high. How do you explain these changes to clients and stakeholders without them thinking the SEO strategy has failed?
It is no longer just about competing for shelf visibility, as it was with traditional SEO. Today the goal is to earn the recommendation of the “shop assistant”, represented by AI agents. The strategy has not failed: the operating context has changed.
The shift is from a model based on “recommendations leading to transactions” to a scenario where browsers and AI agents read content, make decisions and act directly on behalf of users. In this landscape, the outcome is no longer measured only in clicks, but in the ability to build brand authority and a recognizable presence in the training data, so that the model is able to identify and understand who you are.
Many people talk about “GEO” as if it were a new discipline, when in reality much of LLM behavior simply reflects what is already happening in search engines. Does it make sense to talk about optimizing for LLMs? Aren’t we feeding an illusion of control that does not exist?
Everyone is rediscovering SEO through the lens of citations in language models. In reality, there are essentially two ways to appear inside an LLM.
The first concerns presence in the training data: it is a dynamic tied to brand awareness, mentions and PR activity. In other words, what counts is how well known and how widely discussed a brand is over time.
The second is direct citation in the answers: this goes through the search API. If an LLM cites a source, it means a search engine like Google or Bing identified it as relevant first.
Thinking you can specifically optimize “the brain” of an LLM means nurturing an illusion of control. The path remains solid SEO and brand authority: that is the only real way to influence what the model knows and uses.
Britney, to wrap up, I am very curious to hear your opinion on the controversy between OpenAI and the New York Times. The NYT, which sued OpenAI for copyright infringement, asked the court to compel the company to hand over 20 million private conversations of ChatGPT users in search of evidence of misuse of its articles. OpenAI argues that this request is a direct attack on user privacy because it involves millions of people who have nothing to do with the lawsuit. May I ask what you think?
There are no private conversations with an LLM. This controversy highlights the “black box” problem. Users were never clearly informed that their interactions could be requested or used as evidence in a copyright infringement lawsuit. This points to a transparency deficit and therefore a serious failure in educating the people who use generative AI systems.
If a court can order OpenAI to hand over users’ “private” chats to third parties like the New York Times, then the very idea of the “safety” of these tools is a myth.
The whole affair shows that we are all just unpaid workers for these companies and that our privacy comes second to their legal battles. It is a good reminder: better never to use AI as your only source for high-stakes decisions, and never to feed it data you would not want to see quoted in a courtroom.
Here is the video where I tell you how this interview with Britney can help you understand how the visibility of your business is changing in the AI era:
We are brands, and there is much more beyond the click: visibility, authority and the memory of the models
The interview with Britney Muller gave us a genuinely interesting lens on how the relationship between artificial intelligence, search and visibility is evolving.
LLMs operate according to probabilistic logic and introduce a new dynamic in how information is selected, interpreted and returned. For this reason, thinking you can specifically optimize “the brain” of an LLM means nurturing an illusion of control, as Britney pointed out during the interview.
In this scenario, SEO takes on a different shape: less tied to intercepting the single query, more oriented toward building authority that is recognizable over time.
Visibility, therefore, shifts from the click to trust.
AI agents read content, evaluate sources and act as intermediaries between brands and users. What emerges is a selection based on reputation, message consistency and a widespread presence across the relevant information contexts.
To put it simply, artificial intelligence amplifies what already exists, rewarding the brands capable of consolidating a clear, credible voice.
The future of online presence, then, hinges on a sharp change of perspective: attention shifts from executing isolated tactics to systematically building authority.
Shortcuts lose their effectiveness this way (if they ever had any!), while strategic consistency, editorial continuity and the ability to position the brand as a recognizable reference over time become ever more central.
It is a path that requires vision, discipline and alignment between marketing, technology and content, more interdependent today than ever.
A sincere thank you to Britney Muller for the clarity, depth and concreteness of this conversation.
See you next week with SEO Confidential, to keep the conversation going on how search is changing.
Happy holidays to everyone, we’ll catch up after Christmas!
