Business to Agent: Marketing That Declares Itself to Machines

The recipient of marketing is changing

My whole line of reasoning starts from a simple intuition I've held from day one: marketing has always been an act aimed at someone. For a century that someone was a person - with attention to capture, emotions to move, objections to dissolve. We called that relationship Business to Consumer, then Business to Business when the other side was a company. In both cases, though, the last metre was always covered by a human being who read, looked, decided.

That premise is cracking. A growing share of decisions no longer passes directly through a human eye, but through a system acting in its place: an assistant that searches for a supplier, compares options, filters, summarises and - increasingly - performs the next action. The recipient of the message is no longer only the person. It's also the machine the person has delegated. I call it Business to Agent.

What Business to Agent is, and what it isn't

Business to Agent doesn't mean talking to a chatbot, or putting a virtual assistant on your website. It means recognising that, between you and your customer, a third party is inserting itself: a software agent that reads the world on behalf of a human and reaches them with an answer, a shortlist, sometimes a decision already made.

This agent doesn't behave like a visitor. It doesn't scroll your homepage, isn't guided by visual hierarchy, isn't held by a strong image. It queries, extracts, compares. It looks for declared facts, not impressions. And when it doesn't find clear facts, it doesn't stop: it guesses. It's in that margin of interpretation that opportunities are lost, or won, today.

The term isn't mine: «Business to Agent», or B2A, is emerging internationally - from IBM to Kantar to the Y Combinator ecosystem - to describe exactly this shift. What I try to add is a precise reading: not just «optimise for agents», but declare your identity so that a machine treats it as fact. It's the difference between being found and being recognised.

Marketing has always been an act of translation

Seen this way, Business to Agent isn't a break, it's a continuation. Marketing has always been a work of translation: taking what a business is and making it intelligible to whoever is on the other side, in that recipient's language and codes. The channels changed - radio, print, television, the web - but the gesture was the same: translating identity into something the receiver could understand.

Philip Kotler told this story as a succession of eras: from product-centric marketing (1.0) to consumer-centric (2.0), to values-driven marketing (3.0), to the digital turn (4.0) and finally to Marketing 5.0: Technology for Humanity (2021), which puts artificial intelligence at the centre of the relationship. In all these eras, though, technology remained a tool in the service of a human recipient. Business to Agent introduces the break none of those eras foresaw: the machine is no longer only the tool, it's also the recipient.

Today the receiver changes, and so the code changes. Translating your identity for a machine doesn't mean writing more persuasive copy: it means making it readable as data. For an agent, the clearest form in which you can say who you are isn't a well-written sentence, it's a structured, machine-readable declaration that leaves no room for ambiguity. It's the same principle I've worked on all along: an entity is citable and reliable to an AI system not by how much content you produce, but by how readable you are.

The Patagonia case: from repeated values to declared values

I like to start with Patagonia, because it's the clearest example of what it means to become inseparable from your values in the eyes of a machine. If you ask a language model today who makes sustainable outdoor clothing, the name appears almost every time. Not by chance: across billions of pages read, the words «sustainable», «ethical» and «Patagonia» have become statistically inseparable. The company built that association through decades of consistency: in 2018 it rewrote its corporate purpose into a sharp sentence, «We're in business to save our home planet», and in September 2022 founder Yvon Chouinard transferred the entire ownership to a trust and a non-profit - the Patagonia Purpose Trust and the Holdfast Collective - declaring that «Earth is now our only shareholder».

And here comes the detail that changes everything, and that is the heart of my reasoning from day one. Patagonia didn't leave that sentence only in texts written for human beings: it put it in the data. In the JSON-LD of its European site, the Organization markup with which the company describes itself to machines, the owner field - the one that would normally hold a holding company or a person - doesn't list a company. It reads the string "The Earth".

{
  "@type": "Organization",
  "name": "Patagonia",
  "owner": "The Earth"
}

It's not a registry detail, it's a declaration. An agent reading that markup doesn't have to interpret the manifesto, the interviews or the press releases to understand who controls the company and why it exists: it finds it stated as fact, in a language it processes without ambiguity. It's the mission translated into the language of machines.

This is exactly Business to Agent. Not hoping a model will infer your values after reading a thousand pages, but declaring them in the structure, where an agent reads them as verifiable statements. Patagonia had decades of consistency to make those values inseparable from its name; the structured declaration puts that same possibility - stating who you are and what you stand for so a machine treats it as fact - within reach of anyone willing to structure their data. It's the difference between a bet on volume and time, which few win, and an act of sovereignty over your own narrative.

From persuasion to declaration

This is where marketing changes nature. Towards a person, the main instrument is persuasion: I order the arguments, choose the images, build an emotional path that leads to a decision. Towards an agent, persuasion counts for little. An agent doesn't feel emotion, doesn't respond to tone, isn't held back. What it looks for is a coherent, verifiable declaration: who you are, what you offer, on what terms, with what evidence.

This doesn't mean persuasion dies. The person remains, and the final yes is still theirs. It means that, before persuasion, a new layer opens: readability. If the agent doesn't select you, you never reach the stage where you might convince. The structured declaration becomes the condition for access to persuasion, not an alternative to it.

What an agent does that an answer engine doesn't

In recent months many have started talking about how to appear in AI responses: getting cited when someone asks ChatGPT, Gemini or Perplexity for advice. It's a first level, and it's real: that level now has a name and a literature, Generative Engine Optimization, formalised in a paper presented at KDD 2024 that shows how structured optimisation of content can increase its visibility in generative responses by up to 40%. But the agent goes beyond citation. An answer engine names you; an agent selects you and then acts: it compares terms, checks availability, prepares or initiates a transaction on the user's behalf.

This raises the stakes on data. To be cited, being clearly described is enough. To be chosen by an agent that has to act, you need operational data: a unique identity, a typed offer, terms, availability, proof of reliability. What for an answer engine was a bonus becomes, for an agent, the precondition. Without that data, the agent can't trust enough to proceed, and moves on.

And this isn't a theoretical scenario. In little more than a year an infrastructure has emerged for agents to act and pay: Anthropic's Model Context Protocol (November 2024) to connect them to data and tools, Google's Agent Payments Protocol and initiatives like Visa Intelligent Commerce and Mastercard Agent Pay (2025), through to OpenAI's Instant Checkout inside ChatGPT, built with Stripe (September 2025), which lets a purchase be completed without leaving the conversation. With over 700 million people using ChatGPT every week, the agent that acts isn't a distant promise: it's already at the door.

The entity becomes the brand's interface

The consequence is clear: your entity - the structured, linked set of who you are and what you offer - becomes the interface through which agents meet you. Not the website as a page to look at, but the entity as a surface to query. It's the machine-readable version of the brand: the infrastructure that lets an agent move from finding a solution to acting on it.

For anyone doing marketing it's a reversal of perspective. For years we designed what the user sees. Now we also have to design what the agent reads, and the two planes don't coincide. A website can be beautiful for a person and unreadable for a machine. A well-built entity works on both: it communicates to the human and declares to the agent.

Trust stops being perception and becomes structure

There's one final shift, and it's the deepest. Towards people, trust is built with signals: a polished brand, reviews, social proof, perceived authority. Towards an agent, trust is built with verifiability. A data point linked to a stable source, a VAT number that matches a registry, an identity anchored to recognised entities is worth more than any adjective.

It's a world in which trust isn't suggested, it's demonstrated in the structure. The sameAs that links your profile to an official registry, the anchoring to verified entities, the consistency of data across all sources: these are the new signals of reliability, because they're the only ones an agent can check on its own. Reputation, for a machine, is a chain of verifiable statements.

Why this isn't a distant future

One might think all this concerns a remote tomorrow. It doesn't. The first level - AI that searches and recommends - is already part of millions of people's habits. The second - the agent that acts - is emerging now, while the data infrastructure of most businesses is still poor. It's exactly the window in which it pays to move: whoever declares their identity now becomes one of the few clear reference points when agents start choosing at scale.

The numbers bear this out. Gartner forecasts that by 2028, 90% of B2B buying will be intermediated by AI agents, with over $15 trillion of spend flowing through agent exchanges, and that agents will outnumber human sellers tenfold - these are projections, not certainties, but they point in a direction that's hard to ignore. And on the consumer side the phenomenon is already measurable: according to Adobe Analytics, during the 2025 holiday season traffic to e-commerce sites from generative-AI sources grew by nearly 700% year on year, with conversion rates about a third higher than other channels.

And anchors to verifiable sources don't wear out. Unlike a campaign, which stops working when you stop paying for it, a structured declaration accumulates and strengthens over time. Moving early means being already readable when the others start to be: a position that's hard to displace.

What's really at stake: sovereignty over your narrative

There's a reason this topic, for me, isn't a technique but a matter of principle. When an agent doesn't find clear data, it doesn't give up: it fills the gaps. It decides who you are, what you do, what you stand for, based on what it can infer. In that moment you no longer govern your narrative: the machine governs it, with the information it found - including what's wrong, outdated, or belongs to a namesake.

Declaring your identity in structured data is how you take that control back. It's not optimisation, it's sovereignty: deciding what systems know about you, and how, instead of letting them decide for you. Business to Agent, ultimately, is this: the operational version of the idea that governing your digital narrative, in a world where AI systems define it, is a right to exercise, not to suffer.

And it's not only a marketing question. The vocabulary of sovereignty is already in public policy: the European data strategy openly speaks of «data sovereignty», AgID's Italian Strategy for Artificial Intelligence 2024-2026 addresses technological and digital sovereignty, and the European AI Act - Regulation (EU) 2024/1689, in force since August 2024 - sets the rules of the game. Narrative sovereignty is the individual and business version of that same principle: remaining the author of your own story even when decisions pass through machines.

Where to start

The first step is always the same, and disarming in its simplicity: ask the main AI systems what they know today about you or your business, and observe what's correct, what's missing, and what's invented. That snapshot tells you exactly how much of your narrative you're already delegating to the machine.

From there the work is to declare what you want treated as fact: identity, offer, values, evidence, linked to the right entities and to verifiable sources. Not to please an algorithm, but to remain the author of your own story even when the reader is an agent.

The shift towards intelligent agents represents the new frontier of GEO (Generative Engine Optimization), redefining how brands declare their identity and values directly to machines.

Key references

• Aggarwal P. et al., «GEO: Generative Engine Optimization», Proceedings of KDD 2024 (ACM SIGKDD) - arxiv.org/abs/2311.09735
• Lewis P. et al., «Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks», NeurIPS 2020 - arxiv.org/abs/2005.11401
• Anthropic, «Introducing the Model Context Protocol», 2024 - anthropic.com
• OpenAI, «Buy it in ChatGPT» (Instant Checkout / Agentic Commerce Protocol with Stripe), 2025 - openai.com
• Gartner, forecasts on B2B buying intermediated by AI agents by 2028 - gartner.com
• Patagonia, «Earth is now our only shareholder», 2022 - patagoniaworks.com
• European Union, Regulation (EU) 2024/1689 (AI Act) - eur-lex.europa.eu
• AgID, Italian Strategy for Artificial Intelligence 2024-2026 - agid.gov.it