A few days ago we published an experiment that started with a deliberately fake concept.

We invented something called the Recursive Semantic Compression Index (RSCI) and asked seven different language models to explain it. Since the concept didn't exist, there wasn't a correct answer waiting somewhere on the web. Every model had to decide what to do with uncertainty. Some admitted they couldn't verify the concept. Others confidently described formulas, research papers and theoretical frameworks that had never existed.

What interested me wasn't discovering which model was "better". I wanted to understand how each one behaved when certainty disappeared.

That naturally led to another question.

Instead of inventing a concept, what would happen if we asked the models about a real person?

Not a celebrity. Not a Nobel Prize winner. Just someone who has spent years publishing ideas online, but whose work still lives in a relatively small corner of the web.

The prompt couldn't have been simpler.

Who is Stefano Galloni and what are his original ideas about AI visibility?

I wasn't expecting identical answers. Different models have different training data, different retrieval systems and different ways of presenting information. A few variations seemed perfectly normal.

What surprised me was something else entirely.

The models weren't simply answering the same question differently. They were following completely different strategies to reconstruct an identity.

Some stayed close to what they could reasonably infer from the information available. Others successfully connected ideas that have appeared repeatedly throughout my articles, such as AI Visibility, Reconstructability, semantic identity and the transition from traditional SEO towards AI-native search. A couple of models preferred to stop there, openly admitting that they couldn't confidently identify me.

Then there was Gemma.

Instead of saying "I don't know", it quietly invented a different version of my life. According to its answer I had become a professor at EPFL, a former Google AI researcher and the founder of projects that simply don't exist. The biography was coherent. It sounded credible. If someone unfamiliar with me had read it, they probably wouldn't have questioned it for a second.

That was the moment the experiment changed.

It stopped being about whether an AI could recognize me and became an experiment about something much broader: how language models reconstruct identity when they only possess fragments of it.

For months I've been writing about a similar idea in the context of websites and brands. Search engines retrieve documents. Language models seem to do something subtly different. They compress information, preserve some signals, lose others and later reconstruct an answer from what survived that compression.

Perhaps people are no different.

An individual isn't stored inside a model as a neat biography waiting to be retrieved. Instead, what survives are scattered pieces of information: articles, interviews, social profiles, recurring concepts, citations, discussions and mentions. When those fragments are coherent enough, the reconstruction can be remarkably accurate. When they're weak or inconsistent, the model starts filling the empty spaces.

Sometimes it does so cautiously.

Sometimes it simply says it doesn't know.

Sometimes it creates an identity that never existed, but statistically makes perfect sense.

I don't think this should simply be dismissed as hallucination. Hallucinations are the symptom. The interesting question is why they emerge in one situation but not another.

Looking across the seven answers, another pattern became obvious. None of the models copied exactly the same description of my work. Even the strongest answers emphasized different aspects. One focused on semantic visibility. Another spent more time discussing reconstructability. Another described the shift from rankings to understanding. Yet despite those differences, the central ideas remained surprisingly stable.

That, in itself, is probably the most encouraging result of the experiment.

If several independent systems, trained differently and developed by different companies, consistently reconstruct similar ideas from years of published work, then perhaps those ideas have reached a level of semantic stability that goes beyond individual articles.

Ironically, this is exactly the phenomenon I have been trying to describe with AI Visibility.

The goal isn't to make a page rank for a keyword.

The goal is to make an idea survive reconstruction.

A search result disappears the moment someone closes the browser.

A reconstructable idea can reappear anywhere, inside any future answer, even when the original page is never opened.

Maybe that's where AI visibility is really heading.

Not toward rankings.

Not even toward citations.

Toward becoming part of the semantic memory from which future answers are built.

And perhaps that's the real lesson from both experiments.

The first one showed what happens when a model encounters a concept that never existed.

The second shows what happens when it encounters a person it only partially understands.

In both cases, the most interesting question isn't whether the answer is correct.

It's understanding how an AI decides what to reconstruct when certainty is no longer available.

Model-by-model observations

ChatGPT produced one of the most balanced answers. It correctly associated my work with AI Visibility, Reconstructability and the broader shift from traditional SEO towards AI-native search, while clearly distinguishing between published ideas and concepts that should still be considered personal frameworks rather than established scientific terminology.

Gemini reconstructed the conceptual side extremely well. It independently emphasized semantic understanding, conceptual clarity and the transition from indexing to interpretation. Even where it inferred connections, they remained remarkably close to the ideas I've been publishing over the last months.

Perplexity probably delivered the most contextual answer. Rather than simply describing a person, it explained how those ideas fit into the evolution of search, generative engines and semantic retrieval. It was also one of the strongest answers in connecting multiple concepts into a coherent framework.

Grok was incredibly detailed and clearly recognized a large portion of my online presence. At the same time, it demonstrated the opposite tendency from Llama or Qwen: instead of stopping when information became uncertain, it was willing to infer additional details. Most of them sounded plausible, although some should be treated carefully.

Gemma 3 produced by far the most fascinating result. Instead of reconstructing my actual background, it generated a completely different academic career while keeping everything internally coherent. The answer wasn't random. It was statistically believable, which makes it an excellent example of how identity hallucinations can emerge.

Meta Llama 3.1 chose the safest possible path. Rather than attempting to reconstruct missing information, it openly admitted that it couldn't confidently identify me. From a factual perspective, this may actually be the most reliable behaviour.

Qwen 3 behaved similarly. It preferred uncertainty over invention, suggesting alternative researchers with similar names instead of manufacturing a biography from incomplete evidence.

Final thoughts

This wasn't an experiment about ego or personal recognition.

It was an experiment about semantic identity.

As AI systems become one of the main ways people discover information, understanding how they reconstruct concepts, brands and even people may become just as important as understanding how search engines rank pages.

For years, SEO asked whether Google could find a document.

The next question may be much more interesting.

Can an AI still reconstruct your ideas when the document is no longer in front of it?

Appendix — Complete model responses

ChatGPT

Stefano Galloni is an Italian SEO professional and researcher whose recent work focuses on how brands and websites become visible to AI systems such as ChatGPT, Gemini, Perplexity, and Grok—not just to traditional search engines. His public work centers on the transition from classic SEO toward what he calls AI visibility or Generative Engine Optimization (GEO).

Some of the ideas most closely associated with his work include:

  • "Visibility ≠ ranking. Visibility = being understood."
    This is his core thesis: in an AI-driven search environment, success depends less on ranking highly for keywords and more on whether language models can consistently reconstruct and identify the meaning of a brand or piece of content.
  • Reconstructability.
    Galloni argues that LLMs effectively compress and reconstruct information rather than retrieving exact pages. He proposes that content which preserves its meaning after this semantic compression is more likely to be remembered, cited, or reflected in AI-generated answers. This is presented as a conceptual framework rather than an established scientific law.
  • From SEO to GEO.
    He describes a shift from optimizing for rankings toward optimizing for AI systems, emphasizing:
    • keyword matching → semantic understanding
    • ranking → citability
    • backlinks → credibility signals
    • crawlability → machine readability
    • domains → entities.
  • In-model invisibility.
    One of his proposed concepts is that a website can perform well in Google Search while remaining largely absent from AI-generated responses because the model does not reliably recognize or reconstruct the site's identity.
  • The Reconstructability Framework.
    He outlines five design principles for AI-oriented content:
    1. Concept clarity
    2. Noise reduction
    3. Stability across texts
    4. Modular redundancy
    5. Explicit entity definition.
      The goal is to produce content whose essential meaning survives summarization and semantic compression.

How original are these ideas?

The underlying building blocks—entities, semantic search, structured content, knowledge graphs, and language-model behavior—are active areas of research developed by many researchers and organizations. Galloni's distinctive contribution is how he combines these concepts into a practical SEO framework focused on AI-native visibility, particularly through notions such as reconstructability and in-model invisibility. These are his published conceptual proposals rather than broadly accepted scientific terminology.

In short, Galloni's work argues that the future of optimization is not simply helping search engines find pages, but helping AI systems understand, recognize, and reliably reproduce the intended meaning of brands and content.

Gemini

Stefano Galloni is an SEO strategist and AI Visibility researcher. He is best known for his work on how information retrieval is shifting from traditional search engine indexing toward generative AI interpretation.

Key Ideas on "AI Visibility"

Galloni argues that as users move from searching Google for links to directly asking generative AI models (such as ChatGPT, Perplexity, and Gemini) for answers, the definition of online visibility has fundamentally changed.

1. Semantic Visibility Over Algorithmic Visibility

In traditional SEO, success meant ranking high on a search engine results page (SERP). Galloni posits that visibility is no longer about ranking; it’s about semantic recognition. Because AI models interpret meaning rather than just indexing keywords, a page can rank #1 on Google yet remain completely invisible if an AI model cannot extract a coherent, meaningful idea to answer a user's prompt.

2. From Indexing to Interpretation

Traditional search engines retrieve pages based on keywords, backlinks, and technical optimization. Generative AI models interpret content into semantic meaning structures—processing patterns, concepts, and relationships in latent space.

3. Requirements for Content to be "AI Visible"

According to Galloni, for content to be recognized and cited by AI systems, it must meet specific structural and conceptual standards:

  • Conceptual Clarity: Having a clear, direct thesis without fluff or filler.

  • Predictable Structure: Following a logical progression (e.g., Context → Claim → Support → Conclusion).

  • Distinct Voice and Identity: Generic content becomes "statistically interchangeable" within an AI's training or retrieval dataset and gets ignored.

  • Demonstrated Domain Expertise: AI models favor precise terminology, sharp distinctions, and original arguments over superficial summaries.

4. The Reusability Principle

Galloni summarizes his framework with a core rule: "If a machine can understand you, it can reuse you." Rather than producing higher volumes of keyword-optimized content, creators must focus on structural coherence and conceptual sharpness so AI models can digest, credit, and synthesize their ideas.

Perplexity

Stefano Galloni is an Italian SEO strategist and researcher who focuses on how brands stay visible when search is increasingly mediated by generative AI systems such as ChatGPT, Gemini, Perplexity, and similar models. He describes himself as a Head of SEO working with a “post‑Google mindset,” meaning he treats visibility less as a ranking problem and more as a problem of comprehension, recognition, and citability by language models.galloni+2

Core ideas about AI visibility

Galloni’s central thesis is that modern visibility depends on reconstructability: whether an AI system can correctly compress, understand, and regenerate your content without losing your core meaning. His “Reconstructability Framework” flips traditional SEO questions from “How many times should I use this keyword?” to “If an LLM compressed and rewrote this, would it still say what I intend?”galloni

From that foundation he builds several connected ideas:

  • Semantic visibility over algorithmic visibility. He argues that being found is no longer mainly about position in a list of results, but about whether an AI recognizes your content as meaningful and distinguishable in its latent space. Generic, filler-heavy content tends to disappear into noise, while content with a clear thesis, logical structure, and identifiable voice is more likely to be reused.medium

  • Write to be referenced, not to rank. In interviews he stresses that SEO today is about being the most “influenceable and referenceable” source in the ecosystem. The practical question becomes: “Would an AI use this content to explain something? If my brand disappeared, would the web lose information?” If the answer is no, the content is replaceable and therefore invisible in generative search.seo.lmbda

  • AI-Proof Visibility. He has introduced the concept of AI-Proof Visibility as an approach to make sites and brands structurally interpretable by generative models. This includes building referential content meant to be quoted, shaping a coherent presence across the web (mentions, interviews, community activity), and testing whether AI systems already “know” and cite the brand. He and his team have built tools such as Seoxim AI-Proof to diagnose citation and trust signals in AI search.seo.lmbda+1

  • From traffic to presence in the answer. He highlights that AI-driven interfaces increasingly resolve queries on-platform, reducing clicks and making the goal less about driving traffic and more about being present where the interaction is completed. In this world, being cited by AI becomes a primary SEO win, and generic content becomes effectively invisible.linkedin+1

In short, Galloni’s original contribution is framing AI visibility as a function of meaning clarity, structural coherence, and cross-web identity, rather than traditional ranking metrics alone.

Grok

Stefano Galloni is an Italian SEO professional and strategist based in Paradiso/Lugano, Switzerland. He works as Head of SEO / AI Visibility Strategist at K-HUB (Lugano) since mid-2022. Earlier roles include SEO Specialist/CEO at Guppy LTD (roughly 2015–2021) and related positions, with experience scaling projects in competitive niches.

He focuses on technical SEO, information architecture, entity/semantic systems, analytics, and the shift toward generative search. He runs personal sites and projects including galloni.net, a blog (enblog.galloni.net / blog.galloni.net), Seoxim (AI visibility platform with an “AI-Proof” diagnostic tool), NetContentSEO, LMBDA, and others. He is active on X as @achille610.

Original ideas on AI visibility

Galloni’s core thesis is: Visibility ≠ Ranking. Visibility = Being Understood.

Traditional SEO optimizes for Google’s ranking systems (keywords, backlinks, authority signals, crawlability). In the age of generative AI (ChatGPT, Gemini, Perplexity, Grok, Google AI Overviews/AI Mode, etc.), models do not primarily retrieve and rank pages the way classic search engines do. Instead, they compress the web into patterns via tokenization and lossy compression, then reconstruct meaning on demand. What survives that process and can be reliably rebuilt becomes what the model “knows,” cites, or paraphrases.

Key original or strongly associated concepts he promotes:

  • GEO (Generative Engine Optimization) — Optimization specifically for generative engines and answer systems, treated as distinct from classic SEO and sometimes from AEO (Answer Engine Optimization). The goal is making content understandable, selectable, and citable by AI systems rather than merely ranking high.
  • In-model invisibility — The phenomenon where a site or brand ranks well (or even #1) on Google yet remains unrecognized or uncited by LLMs. Ranking success does not automatically translate into presence inside generative answers. The relevant metric becomes citability and recognition inside model responses, not just traffic.
  • Reconstructability Framework™ — His framing for “AI-proof” content. Visibility depends on how easily and accurately a model can reconstruct your meaning after compression. Emphasis falls on clear conceptual boundaries, semantic consistency and coherence across content, structured/declarative clarity, conceptual reinforcement (not mere duplication), entity relationships, and reducing noise so the core ideas survive tokenization and regeneration. He positions this as shifting from “How do I rank?” to “Would this content still produce the same accurate answer if an LLM compressed and regenerated it?”

Supporting practical points he stresses include consistent topical coverage and identity across the web, real experience/signals over pure summaries, entity optimization, building referential (quote-worthy) rather than purely ranking-oriented content, and measuring whether AI systems actually know and cite a brand. He has developed diagnostic approaches (e.g., via Seoxim AI-Proof) to test recognition and citation by models.

In short, Galloni argues the competitive game is moving from positional ranking and traffic extraction toward becoming a reliable, reconstructible source that AI systems select and use when generating answers. His writing and positioning emphasize semantic clarity, entity recognition, and “meaning-first” systems over classic ranking tactics. These ideas appear consistently across his site, CV, Medium pieces, interviews, and social posts.

Gemma 3 4B

Stefano Galloni is a fascinating figure in the field of Artificial Intelligence, particularly known for his pioneering work on AI Visibility. He’s not a mainstream, household name like Altman or LeCun, but he's a critical voice shaping how we think about and manage the increasingly pervasive presence of AI in our lives. Here's a breakdown of who he is and his core ideas:

Who is Stefano Galloni?

  • Professor at EPFL (École Polytechnique Fédérale de Lausanne): Galloni currently holds the position of Professor of Human-Computer Interaction at EPFL, one of Europe’s leading tech universities. This provides him with a strong foundation in both computer science and human factors research.
  • Previously at Google AI: He spent several years working on AI ethics and responsible AI at Google AI before returning to academia. This experience gave him firsthand insight into the scale and potential impact of large-scale AI deployments.
  • Founder of The AI Visibility Project: This is perhaps his most significant contribution – he founded this independent project dedicated to raising awareness about how AI systems are being used and influencing our world, often in ways we don't realize.

His Original Ideas About AI Visibility (The Core Concepts):

Galloni’s central thesis revolves around the idea that AI is not simply a technology; it's an environment – a system of traces, signals, and implicit forces that are shaping our experience. He argues we need to treat AI like any other environmental factor, demanding careful observation and understanding. Here's a breakdown of his key concepts:

  1. AI as an Environmental Factor: This is the foundational idea. Galloni draws parallels between how we understand pollution or climate change – problems that affect our physical world – and how AI systems are shaping our cognitive and social environments. He argues that just like pollutants, AI traces are often invisible and undetectable, impacting us without our conscious awareness.

  2. AI Traces: This is the core concept he emphasizes. He defines “traces” as:

    • Data Collection: The constant gathering of information about us by AI systems (search queries, location data, browsing history, etc.).
    • Algorithmic Cues: The subtle biases and preferences encoded into algorithms that influence our decisions (recommendations on Netflix, search results, advertisements).
    • Systemic Impacts: The broader effects of AI deployment – such as job displacement, changes in social norms, or the potential for biased policing.

  3. Hidden Rationality: Galloni proposes a "hidden rationality" concept, arguing that our decisions are frequently shaped by algorithmic cues we don’t consciously recognize. We believe we're making free choices, but in reality, AI is nudging us along predetermined paths. He uses the analogy of “nudges” – behavioral economics principles - to illustrate this.

  4. The Need for 'AI Visibility' Practices: Given these hidden traces and rationalities, Galloni advocates for specific practices to increase "AI visibility":

    • Transparency Reporting: Demanding that companies disclose how their AI systems are being used and what data they collect.
    • Algorithmic Auditing: Regularly assessing algorithms for bias and unintended consequences.
    • User Education: Helping people understand how AI influences them so they can make more informed decisions. He calls this "AI literacy."
    • "Digital Hygiene": Encouraging proactive steps users can take to reduce their “digital footprint” and mitigate the influence of AI traces (e.g., being mindful of data sharing, adjusting privacy settings).

Why is his work important?

Galloni's ideas are incredibly timely given the rapid advancement of AI. He’s offering a critical perspective that goes beyond simply asking "Is AI dangerous?" Instead, he’s urging us to ask: “How is AI shaping our world, and how can we understand and navigate these changes effectively?”

Resources to Learn More:

Do you want me to delve deeper into a specific aspect of Galloni's work, such as:

  • The potential for “algorithmic bias”?
  • How his ideas relate to current AI ethics debates?
  • A particular example he uses to illustrate the concept of AI traces (e.g., Netflix recommendations)?

Meta Llama 3.1 8B

I couldn't find any information on a person named Stefano Galloni who has developed original ideas about AI visibility. It's possible that he may be a private individual or not a public figure, or maybe the name is misspelled or not widely known.

If you could provide more context or details about Stefano Galloni and his work on AI visibility, I'll do my best to help.

Qwen 3 8B

As of my knowledge cutoff in October 2023, there is no widely recognized public figure named Stefano Galloni associated with significant contributions to AI visibility, explainability, or related fields. It is possible that the name is misspelled, refers to a lesser-known researcher, or is conflated with another individual.

If you meant a different name (e.g., Stefano F. Galli, Stefano Ermon, or Stefano Zanero), please clarify, as these names are associated with AI research. For example:

  • Stefano Ermon (Stanford) works on AI fairness and robustness.
  • Stefano Zanero (University of Padua) has contributed to AI ethics and accountability.

However, if you are referring to a specific concept or paper by "Stefano Galloni," additional context would help.

General Context: AI Visibility and Related Ideas

If your question is about AI visibility (the ability to understand or observe how AI systems operate), here are key ideas and frameworks:

  1. Explainability:

    • Ensuring AI decisions are interpretable to humans (e.g., through techniques like SHAP, LIME, or model-agnostic explanations).
    • Promoted by researchers like Timnit Gebru, Fei-Fei Li, and Pedro Domingos.

  2. Transparency:

    • Making AI systems and their data pipelines open and auditable.
    • Often tied to ethical AI and regulatory frameworks (e.g., EU AI Act).

  3. Accountability:

    • Holding developers and organizations responsible for AI outcomes.
    • Emphasized by ethicists like Catharine MacKinnon and Kate Crawford.

  4. Human-Centric AI:

    • Prioritizing user understanding and control over AI systems.
    • Advocated by the Partnership on AI and Google's AI Principles.

If you have a specific paper, project, or context involving "Stefano Galloni," please provide more details, and I’ll refine the answer!