For thirty years, a business website has had one real job: convince a person to do something. Trust the company, buy the product, book the call, fill in the form. Every design decision, every line of homepage copy, every carefully placed testimonial was built for a human being sitting in front of a screen, making a judgment call.

That job hasn’t disappeared. But it’s no longer the only one. A growing share of the traffic reaching, reading, and acting on business websites right now isn’t human at all. It’s an AI system reading your site on someone else’s behalf, deciding what to tell them, whether to recommend you, and in a fast-growing number of cases, whether to complete a transaction without a person ever looking at your homepage.

Microsoft made this explicit in June 2026, describing the web as now operating across three overlapping experiences at once: the traditional human web, an LLM layer where AI assistants answer questions using information pulled from across the internet, and an emerging agentic layer where AI systems act on a person’s behalf rather than just informing them. Microsoft’s own data, released alongside that framing, claims that AI-driven traffic to websites is now growing eight times faster than traffic from human visitors. Whether or not that exact multiple holds everywhere, the direction is not in question. The website’s audience has split in two, and most businesses have only ever designed for one half of it.

What Actually Changed in 2026

This isn’t a vague, ongoing trend dressed up as news. Several concrete things happened this year that are worth knowing specifically, because they show this is infrastructure being built now, not speculation about the future.

Microsoft introduced Web IQ, its own AI-native search infrastructure built specifically to connect AI assistants, including Copilot and ChatGPT, to current, authoritative information about individual businesses. Alongside it, Microsoft added citation reporting to its Clarity analytics product, letting businesses see, for the first time, when and how AI systems are actually discovering, citing, and recommending them, something that was previously invisible in standard analytics. Separately, Microsoft has acknowledged a specific, avoidable problem: many businesses are still blocking AI systems from reading their sites by default, often without realizing it, through generic bot-blocking settings never updated for this new category of visitor.

At the same time, the infrastructure for AI to act, not just inform, has moved from pilot to production faster than most coverage of it suggests. Visa, MasterCard, Google, and OpenAI have each shipped real payment and checkout infrastructure over the past year specifically designed for AI agents to transact on a person’s behalf, and major retailers and platforms have begun adopting it. That’s a genuinely new category of infrastructure that didn’t meaningfully exist two years ago.

What’s worth noting, though, because it cuts against the more breathless version of this story, is that the most extreme version of agentic commerce has already been pulled back. OpenAI initially built the ability for a purchase to complete entirely inside a ChatGPT conversation, and then walked that back earlier this year in favor of a model where the AI handles discovery and recommendation, but the actual transaction still happens on the merchant’s own website. The merchant stays the system of record. That detail matters more than it might seem, and we’ll come back to it, because it tells you something important about what’s actually changing and what isn’t.

What an AI Actually Sees When It Visits Your Site

A human visitor and an AI system reading the same web page are, functionally, looking at two different things.

A beautifully designed site with its key facts buried in an image, a PDF, or a JavaScript widget that only renders after a click is, to an AI system, close to unreadable, regardless of how it looks to a person. This isn’t a call to abandon design. It’s a recognition that a website now needs to work for a reader that doesn’t see design at all, only structure and text.

Trust Has Become a Cross-Referencing Problem, Not a Design Problem

Here’s the part most coverage of this shift misses. A persuasive homepage used to be able to do a lot of the work of earning someone’s trust on its own. An AI system doesn’t work that way. Before it cites a business, recommends it, or answers a question about it, it’s effectively checking your claims against everyone else’s: review sites, directories, news coverage, other businesses’ comparisons, social platforms. If your website says one thing and three other sources say something slightly different, an AI system generally can’t resolve that conflict the way a curious human might, by clicking around to figure it out. It’s more likely to just leave you out of the answer, or hedge in a way that makes you look less certain than a competitor.

This changes what “credibility” actually means online. It’s no longer primarily about how convincing your own website is. It’s about whether the facts about your business, your pricing, your hours, your services, your location, your claims, are consistent everywhere they appear, including on pages and platforms you don’t control. A business with a modest but accurate, consistent web presence can now out-perform a business with a beautifully designed site that quietly contradicts itself across different pages, or hasn’t been updated to match what its listings and reviews say elsewhere.

What This Means for Discovery and Getting Cited

Search used to mean ranking on a results page a person would scan and click through. Increasingly, it means being the source an AI system pulls from when it answers a question directly, often without sending that person to any website at all, yours or anyone else’s. That’s a harder thing to measure and a different thing to optimize for. Ranking first is no longer the same thing as being chosen.

For a business, this means the practical goal is shifting from persuading someone who lands on your page to being legible and trustworthy enough that an AI system is confident citing you before the person ever gets there. We’ve written previously about what this looks like in practice, in the specific context of using AI tools directly on a website, in How Businesses Can Use AI to Improve Their Websites. This article is about something upstream of that: the fact that the wider web your website exists in has already changed, independent of whether you add any AI features to your own site at all.

What This Means for E-commerce Specifically

The infrastructure described earlier, the payment protocols, the checkout standards, the agent identification systems major card networks have spent the last year building, is real and moving quickly. But the OpenAI reversal is the more useful data point for most businesses than any individual protocol name. Even as the technical ability to complete an entire purchase inside an AI conversation became possible, the model that’s actually emerging keeps the merchant’s own website as the place the transaction happens and the record of what was agreed lives.

That has a specific, practical implication for any business selling online. The discovery and recommendation stage of a purchase is increasingly happening through an AI system reading your product data, not a shopper browsing your category pages. If your product information, pricing, availability, and specifications aren’t structured in a way a system can reliably extract, whether that’s through a proper product feed, structured data markup, or simply clear, consistent text rather than information locked in images, you can be functionally invisible at the exact moment an AI is deciding what to recommend, even though your actual checkout process is perfectly fine once someone arrives.

What a Business Should Actually Do About This

Not every business needs to chase every development described above, and a lot of the more elaborate agentic commerce infrastructure is built for enterprise retailers with dedicated technical teams. For most businesses, a shorter list of genuinely high-leverage priorities covers most of the actual opportunity:

• Make sure the facts that matter, pricing, services, hours, location, credentials, exist as plain, readable text somewhere on your site, not only in an image or a PDF.
• Check whether your site is unintentionally blocking AI systems from reading it at all, a surprisingly common and easily fixed oversight.
• Treat the information about your business on other sites, directories, review platforms, marketplaces, as part of your effective online presence, since inconsistency there now costs you visibility, not just tidiness.
• And build any new website or web application with this readability in mind from the start, since retrofitting structure into a site that was built purely for visual persuasion is considerably harder than designing for both audiences from day one, which is the kind of foundational decision that belongs in how a site’s web application is actually built, not something added on afterward.

The Website’s Role Is Changing, Not Disappearing

None of this means websites are becoming obsolete, and it’s worth being direct about that, because the more dramatic version of this narrative overstates what’s actually happening. Plenty of purchases, relationships, and decisions still involve a person actually visiting a business’s site, reading it, and forming their own judgment, and that isn’t going away for anything that requires real trust or nuance.

What’s changing is that the website is no longer the only, or even the first, interaction a customer has with your business. Often, an AI system has already read it, formed an impression, and shaped what a person believes about you before they ever click through, if they click through at all. A website’s job is expanding to include being a reliable source for a system that will summarize, recommend, or act on it long before a human being ever arrives. Getting that right is quickly becoming as fundamental to being findable as having a website was in the first place.