WebMCP: How AI agents will use your website and what it means for your brand visibility

For thirty years, websites have been built for human eyes and human clicks. A new standard backed by Google and Microsoft is quietly preparing them for a different kind of visitor: an AI agent that does not browse your site, but uses it. The brands that can grasp what that changes will be best placed for the next era of digital presence.

Something new is happening to the web. AI assistants are no longer just reading it on our behalf, summarising pages and answering questions. They are beginning to act on it: booking the table, comparing the policies, filling in the form, completing the purchase. The industry calls this the agentic web, and it has a problem. Websites were designed for people. An agent arriving at a page built for human eyes has to guess what the buttons do, and guessing is fragile. It misreads layouts, breaks when a design changes, and gives up.

WebMCP is the emerging answer. In plain terms, it is a proposed web standard that lets a website hand an AI agent a clear menu of what it can do. Search this. Book that. Add this to the basket. Instead of guessing from the screen, the agent reads a published contract of actions, with defined inputs and outputs, and calls them directly. It is a joint effort between the Chrome and Edge teams, and already available in early preview in Chrome, with native support targeted for the second half of 2026…. So it’s happening now! 

The signals of momentum are worth noting. Household names including Expedia, Booking.com and Shopify are reported to be experimenting with it and support has begun appearing beyond the browsers themselves, with ChatGPT reported to be using the tools that enabled pages to declare. When the two biggest browser makers and the world’s most used AI assistant converge on the same idea, it stops being a curiosity and starts becoming an agreed direction.

From being seen to being used

To understand why this matters for brands, notice what actually changes. Until now, every machine that visited your website came to read. Search crawlers read it to rank it. AI models read it to summarise it. WebMCP takes the next step. The machines stop being readers and become users, able to operate your site’s features on a customer’s behalf.

Two design choices make this less alarming and more interesting than it first sounds. The standard is built around human workflows rather than unsupervised automation: the customer stays present, directs the task and approves what matters. The agent acts inside the customer’s own browser, on the live page. For brands, that is quietly good news. It keeps your own site, your owned territory, at the centre of the transaction instead of surrendering it to somebody else’s platform.

What it means for search: cited first, usable second

The sharpest strategic insight in the WebMCP conversation is about sequencing. Agents do not start at your website. They start inside an AI answer. A customer asks their assistant, the assistant chooses which brand to act on and only then does an agent arrive at a site. If your brand is not in that answer, your website never gets called. Visibility comes first. Usability comes second. Both are now required.

That means being found is split into two jobs. The first is visibility, which now means visibility to AI systems as much as search rankings, being the brand that models cite, recommend and choose when people ask. The second is genuinely new: usability by machines, being a site an agent can actually operate once it lands. One without the other is wasted effort. A brand that is visible but unusable will watch the agent move on to a competitor it can operate. A brand that is usable but invisible will simply never receive the visit.

The reassuring part is that becoming agent-ready does not mean rebuilding anything. The practical first steps are modest: annotating the forms a site already has so agents can understand them, registering a handful of the actions that matter most, such as search, booking, checkout or enquiry and publishing the small, standard files that tell AI systems what a site is and does. These are incremental moves and because they all rest on clarity and structure, they improve your conventional search presence at the same time.

The effects: a compressed funnel and a higher burden of proof

When an agent handles discovery and shortlisting, your customer may encounter your brand for the first time seconds before the decision. The parts of the funnel that marketing has always focused on, awareness and consideration, compress dramatically. What replaces consideration is validation: a human glancing at a choice that has already been assembled on their behalf and deciding whether to confirm it. As Wix’s head of AI search has observed, brands have never faced this level of burden of proof. You have moments, not minutes, to look credible.

Two forces are pulling in opposite directions here and both matter. The first is the risk the industry calls disintermediation: if the whole transaction happens on somebody else’s surface, a brand can be reduced to an interchangeable point behind an interface it does not control. The second force pushes the other way. Research reported by Harvard Business Review found that 64% of shoppers want at least one safeguard in place before they would let an agent buy on their behalf. People are happy to delegate the research. They guard the decision. Which means the brand moments that survive the compression become more valuable, not less. The brand a customer recognises at the moment of validation, the name they tell their agent to use, the reputation that makes delegation feel safe in the first place: these are the new competitive assets, and every one of them is built long before the agent arrives.

There is one more effect worth sitting with. Agents pick winners. A results page offers ten options and lets a user decide. A model acting on someone’s behalf tends to choose one. Categories will tilt further and faster towards the brand that is most clearly articulated, most credibly evidenced and most consistently recommended. Distinctiveness has always been valuable. In an agent-mediated market it becomes decisive.

The landscape changes but brand foundations do not

There is always a similar pattern, because it is the story of every digital shift we have navigated in thirty years. Print to web. Desktop to mobile. Owned media to social. Search results to AI summaries. Now summaries to actions. Each time, the channel rewrote the rules of expression and each time the winners were the businesses whose identity, clarity and point of difference were strong enough to move with it. The agentic web will be no different but it does sharpen three things for brand-builders.

First, the website’s job expands. The website is a brand’s most important strategic asset, the one owned space where identity is expressed. WebMCP raises those stakes again. The website becomes not only the brand’s face but its hands: a shopfront and, now, a counter that machines can transact at. Treating it as a legacy channel is precisely the wrong instinct.

Second, brand moves into the machine layer. The names you give your site’s actions, the descriptions attached to them, the structure of your data: this is brand expression now, written for an audience of agents. Clear language has become literally functional. A muddled proposition produces muddled data, which gets skipped. It is the oldest principle we have, strategy before style, wearing new clothes. You cannot annotate what you have never clearly defined.

Third, trust is the licence for delegation. An agent acting as the signed-in customer, inside their session, with their permissions, is the deepest trust the web has yet asked of anyone and it will be extended only to brands people already believe. The trusted brand gets the delegation and in turn, gets the transaction.

The question that matters most

WebMCP is still in it’s infant phase. It is a preview standard, parts of its security model are still maturing. However, that is not a reason to wait, because everything the agentic web will reward, clarity of proposition, visibility in AI answers, structured and legible digital foundations, and a reputation worth delegating to, pays for itself today, whatever timeline the standard follows. That is the definition of a no-regrets investment.

So the question we would put to any brand watching this unfold is a double one. When a machine is choosing on your customer’s behalf, will it choose you? When your customer glances up to confirm, will they recognise a brand worth trusting? Chosen by machines, believed by humans. Two tests, one foundation: clarity, distinctiveness, evidence and trust, including the places only the machines will ever read. The challenge is new but the answer is not. 

At Underscore we help brands walk the tightrope between the human world and the digital one, advising for what is next without losing sight of what works now. If you are thinking about how your brand shows up in AI-driven search, and whether your current digital presence and set-up is ready, we would be glad to talk.

Explore our thinking and our work at underscore.co.uk, or get in touch at [email protected]