ChatGPT, Copilot and Gemini start to compare and buy for customers. A product page designed to convince a human does not convince a machine.
We still talk about it as a distant prospect. It is already there, in part.
Since the end of 2025, ChatGPT has made it possible to purchase within the conversation, without returning to the site. Shopify has deployed its “agentic storefronts” to millions of merchants: for a store that sells in the United States, its products can now appear natively in ChatGPT, without installing anything. Copilot, Perplexity and Google’s AI mode in turn deploy assisted purchasing functions, where an agent can compare, select, and sometimes finalize the order. The protocols that structure all this already exist, supported by OpenAI and Stripe on one side, by Shopify and Google on the other.
It’s not science fiction. It is a new layer of intermediation between a brand and its customer. And it changes a rule that we haven’t had to question for twenty years.
All online commerce was designed to convince a human
Think about what boutiques rely on today. A well-worked product page. Visuals. Reinsurance. Urgency. A tunnel optimized for clicking. Meta campaigns to bring the right visitor to the right page. This entire structure has a single recipient: a human being, in a session, whom we seek to persuade.
An AI agent deletes the session.
He doesn’t look at your banner. He doesn’t feel the urgency. It translates a request into constraints: budget, characteristics, delivery time, possible brand preference. Then it queries catalogs, price feeds, availability. The evaluation is done by data, not by visuals. The work of persuasion that you have carried out for years is not, at this moment, aimed at anyone.
The consequence is brutal in its simplicity
If the agent doesn’t escalate your product, your brand is not in the transaction. No further down a list. Absent.
It is a change of nature, not of degree. On Google, a bad position still left a chance: the user could scroll, click elsewhere, come across you. An agent that constructs its selection from structured data does not scroll. It picks a few options that match, and ignores the rest. If your data is not actionable, you are not a lower ranked option. You are not an option.
What decides now
In this context, the levers move towards unspectacular things, and that is precisely why they are neglected.
The quality and cleanliness of product data become decisive: standardized sheets, standardized identifiers, stocks and prices displayed in real time. Being on the right channels and protocols, so agents can simply find you. And, once again, the consistency of the brand as an entity: an agent brings up what is recognized, recommended, consistent across the web, not what shouts the loudest.
What weighs much less, at this precise moment: the advertising budget, the copywriting of the sheet, the creative which converted humans so well. These levers do not disappear, but they intervene elsewhere, further upstream, on brand preference, not on agent selection.
Nuance, so as not to overreact
You have to keep a cool head. It’s a start. A small share of buyers have already finalized a purchase via an AI assistant, and the majority of sales still go through humans who choose for themselves. No one should overhaul everything tomorrow morning for a still emerging channel.
But the trajectories set the tempo. Morgan Stanley predicts that around 2030, nearly half of online shoppers will use AI agents, for a significant portion of their spending. McKinsey estimates the market in thousands of billions by the same deadline. When a shift of this size is predictable, the question is not whether to prepare for it, but when. And the good news is that the preparation is thankless rather than costly: these are clean data, not a technological gamble.
This is the logical continuation of what we already observe
This subject is not isolated. It extends the same trajectory. First, we stopped getting the click, the answers coming directly into the search. Then, we understood that visibility depended on being cited as an entity, not just well positioned. Agentic commerce is the next step: we may soon stop obtaining choice itself.
At each stage, the same type of brand emerges: one that is structured, coherent, readable, recommended. Not the one who spent the most to show off.
The question I would ask a leader today is not very flattering, but it is useful. If an AI agent had to choose between your brand and three competitors, based on available data alone, would it only find you?
