For twenty-five years, the visitor reading your website was a person. They arrived with a question, formed an impression from the design, and either called you or didn't. What's changing now isn't that they read faster. It's that a growing share of them aren't reading at all, because software reads on their behalf and comes back with a shortlist. In this post we'll separate the two very different things people mean by "AI agents", look at what the evidence says you do and don't need to change, and work out where a smaller company's effort is worth spending this year.
The scale of the forecast is worth stating plainly. Gartner predicts that by 2028, 90% of business-to-business buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through what it calls AI agent exchanges [Gartner, 2025]. That's not a prediction about chat assistants answering questions, it's a prediction about purchase decisions being brokered before a person sees them. If you sell to other businesses, whether you're a commercial cleaning company, a print shop, or an IT provider, the buyer at the other end may increasingly be an agent working from a budget and a list of requirements. Three years isn't long for a change of that size.
Two very different things are being called AI agents
The first kind reads the open web and writes an answer, citing a few sources as it goes. The second kind transacts, which means it needs live prices, real stock levels, and a way to place an order. Advice written for one is being sold as advice for the other, and that's where most of the confusion in this conversation comes from. The requirements are not the same, and neither is the work.
On the answering side, Google's guidance for site owners is unusually direct. It states that there are no additional requirements to appear in AI Overviews or AI Mode, that you don't need to create new machine readable files or markup, and that there's no special schema.org structured data you need to add [Google Search Central]. A page has to be indexed and eligible to be shown with a snippet, which is the same entry fee as before. Schema markup sold as the route into an AI answer is work Google says you don't need. The markup isn't the gate.
The buying side genuinely does run on structured data
Here the picture reverses, and this is the part worth understanding properly. To have products available for purchase inside ChatGPT, a merchant supplies a product feed in CSV or JSON carrying identifiers, descriptions, pricing, inventory, media, and fulfilment options, submitted first as a sample for validation and then refreshed as daily snapshots [OpenAI]. In January 2026 Google published the Universal Commerce Protocol, an open standard built with Shopify, Etsy, Wayfair, Target, and Walmart, which lets an agent discover what a business offers, read real-time inventory, and open a checkout session [Google Developers Blog]. Both are integrations, not tags.
That distinction decides who this applies to. If you sell physical products through a mainstream platform, most of the plumbing is your platform's job rather than yours, and the practical question is whether you've switched the relevant features on. Shopify's recent agentic storefront work shows how quickly that layer is moving. If you sell services, though, and most small businesses do, there's no product feed to file and no protocol yet to join. A dental practice, a law firm, or an HVAC contractor has nothing to plug in. For them this door isn't open yet, and pretending otherwise wastes a quarter.
What holds up either way is being legible
Strip out the protocol talk and the remaining advice is the same for everybody, which is a good sign rather than a disappointing one. A machine has to be able to extract what you sell, who it's for, what it costs, and how it works, from text that exists as text rather than inside an image, a slider, or a script that runs after the page loads. Dates and prices have to be current, because stale information reads as unreliable to a model exactly as it does to a customer. Reviews and mentions elsewhere still carry weight. Nothing on your own site can corroborate itself.
There's a pleasing overlap here that doesn't get mentioned enough. The work that makes a site usable with a screen reader, clean markup, real labels, headings in the right order, is largely the same work that makes it readable by a language model. Accessibility stopped being only a courtesy and quietly became a distribution channel. We covered the answering half of this in more detail in what changes for your website when AI answers first, and the underlying instruction hasn't moved. Write plainly, in HTML, and keep it current.
The part of the pitch the research doesn't support
The urgency framing around all this deserves a harder look. Gartner's own survey work from May 2026 found that 69% of B2B buyers turn to sales representatives to validate AI-generated insights, and that 51% say they're more likely to encounter misleading information from generative AI than from a rep [Gartner, 2026]. Read alongside the 2028 forecast, that describes a split job rather than a replaced one. The agent assembles the shortlist, and a person still checks it before anyone signs. Being legible to the machine gets you considered. Being credible to the buyer still closes it. That part hasn't moved.
A reasonable first step, if you only do one thing
Open your own site with images and scripts disabled, or paste a page's raw text into an assistant and ask it what this business sells and what it charges. Whatever comes back is roughly what an agent has to work with, and for most sites it's less than the owner expects. Then ask three assistants what they'd recommend for the problem you solve, and write down who they name. That's twenty minutes of work, and it produces a baseline. A baseline is the thing most missing from these conversations. We run that kind of visibility check alongside conventional search work in our SEO, SERP and GEO packages, mostly because reporting them separately produces two documents that disagree.
The wider point is older than any of the protocols. A business that can state clearly what it does, what it costs, and who it's for has always been easier to recommend, whether the one doing the recommending is a search engine, an agent, or a neighbour. The machines have just made the cost of being vague a little more visible. If you'd like a hand working out which of these two doors your business is actually standing at, we're glad to look at it with you.
Sources
Gartner, Top Predictions for IT Organizations and Users in 2026 and Beyond, October 2025. Google Search Central, AI features and your website. OpenAI, Agentic Commerce key concepts. Google Developers Blog, Under the Hood: Universal Commerce Protocol, January 2026. Gartner, 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights, May 2026.
