Visits Are Falling, WebMCP Draws In the Agents

Split image: on the left a customer relaxing on the sofa with a tablet, on the right a technician carrying a toolbox from the service van to the front door, connected by a cyan line — The request went through an assistant, the appointment is set.

Your future customers plan their trip with an AI assistant, a program like ChatGPT, Gemini or Copilot that answers questions in full sentences. That is also where they look for events at their destination. When something breaks, they ask the assistant to help troubleshoot first, then ask it who nearby can fix it. Often they never open the provider’s website at all.

The buying decision now happens inside the assistant, and the chat shows whoever has made their information and processes available to the agent.

Search visits are declining

In March 2025, the Pew Research Center tracked what more than 900 US adults clicked on in Google search results. When an AI summary appeared above the results, they clicked through to a regular result only 8 percent of the time. Without a summary, that figure was 15 percent. Once Google answers the question itself, close to half of the clicks a website would otherwise get simply disappear.

At the same time, the traffic assistants generate on their own is growing. HUMAN Security measured that traffic from AI agents and browsers with a built-in agent rose 7,851 percent year over year in 2025. More than 95 percent of it fell into retail, streaming and media, and travel and hospitality. Cloudflare reported on 3 June 2026 that for the first time, more requests on its network came from programs than from people, 57.5 percent. Cloudflare attributes the rise mainly to agents that read thousands of pages on behalf of an assistant to answer a single question.

In the bright office of a Tyrolean hotel business, the marketing manager and a colleague point at a monitor where the bar for AI agents towers over every other visitor source — A breakdown of visitor sources with agents at the top.

Whoever still reaches a website through an assistant buys far more often, according to Adobe. On US retail sites, these visits lead to a purchase 60 percent more often than visits from other sources, and their number grew 62 percent in July 2026 compared with the year before. These visitors have already made their choice in the chat.

The assistant recommends whoever it can pull information from

Whoever makes their prices, services and proof available so an AI agent can retrieve them shows up in the chat. If the agent cannot find a business’s price, service area or opening hours, it names the competitor whose information it can find instead.

The customer asks their assistant, the assistant retrieves the businesses' information; if it finds none, the competitor appears; if it finds prices, services, proof and processes, your business appears in the chat and the appointment gets booked — The path of a customer request through the assistant.

Whoever also opens up processes, from contact through booking to a purchase in the shop, gets the job directly through the assistant. The customer tells the chat they’ll take Thursday’s slot, and the assistant books it with the business.

On the left the information: prices, services and service area, proof and certificates, frequently asked questions; on the right the processes: get in touch, book an appointment, buy in the shop; below: the customer's assistant answers with your figures and books the appointment with you — What a business can make available to the agent.

Proof builds the agent’s trust

Certificates and processes that a business makes available to the agent raise the agent’s trust in that business. That includes a master craftsman’s certificate, manufacturer training, the terms of a warranty, and the steps of a repair from the first call to the invoice. An assistant booking an appointment for its customer would rather give that slot to a business whose proof it can check for itself.

A master plumber stands in his bright workshop beside a heat pump, holding a tablet with verified certificates; a master craftsman's certificate and manufacturer certificates hang on the wall — A piece of proof an agent can retrieve.

What an assistant learns on integrations.at

Our own website shows what that looks like. If a customer’s assistant asks about our prices, it gets the rate per AI hour for software engineering, the day rate for training, and the payment terms. If it asks for proof, it gets every Microsoft certification with its issue date and a link to check it directly with Microsoft. It finds the training courses with their level and day rate, the frequently asked questions for every service with their answers, and the way to reach us. It fills out the contact form for the customer, though the customer still has to submit it themselves.

On WebMCP Verify our page scores 100 out of 100. How that works on the technical side is covered in Agent-Ready Websites; the engineering behind it is part of MCP servers for your systems.

How to check your own site

Lighthouse is built directly into Chrome and scores your page’s load time, discoverability and accessibility. WebMCP Verify opens your address and shows what an agent finds there. Google’s Chrome extension Model Context Tool Inspector lists which functions your page offers the customer’s AI assistant.

After that, take the five questions a customer asks before buying from you, and write the answers the way you’d give them on the phone. Those answers are what we build on with you, until an assistant can book an appointment with you directly.