Agentic SEO: what an AI agent needs from your website

What an agent is

An agent is an assistant that takes actions rather than only answering questions. It opens pages, compares options, fills a form, books a table, adds to a basket and, with permission, pays. ChatGPT, Gemini, Perplexity and Copilot have all added agent features.

Agentic SEO is the work of making a website one these agents can use: find, read, understand and act on. It is not about tricking anyone. It is about removing the small obstacles that stop a helpful robot from completing a task a customer asked it to do.

For example, a customer might tell their assistant: find a plumber in Dublin who can come tomorrow and book a call-out. The agent will search, open a few sites, look for clear service details, opening hours and a booking form, then try to complete the booking. If your site hides that information or blocks the agent, the job goes to someone else.

What breaks an agent

Agents are patient but literal. They do not scroll, tap or guess the way a person does. If something is not in the page text or clearly labelled, it may as well not exist. Here are the most common blockers.

• Text that only appears after scripts run. If your prices, services or hours load via JavaScript after the page opens, many agents will see an empty page.
• Content behind a login or a cookie wall. An agent cannot create an account to read your menu or your rates.
• Pop-ups that cover the page. A newsletter pop-up that blocks the content stops the agent from reading further.
• CAPTCHAs on ordinary pages. These are fine for sign-in and payment, but they stop agents from browsing your site at all.
• Buttons without clear labels. A button that says 'Go' or an icon with no text gives the agent nothing to act on.
• Prices that appear only in images. If your price list is a photo, the agent cannot compare or confirm it.

Blocked crawlers are the quiet killer. An agent that cannot fetch the page cannot act on it. Many sites accidentally block AI crawlers in robots.txt while trying to block something else. Vasukai's robots.txt checker tests 18 AI crawlers at once, so you can see exactly who is allowed in. You can try it at robots.txt checker.

For example, a dentist in Leeds might ask their assistant to find a practice taking new patients on a Saturday. If the practice's site blocks AI crawlers, the assistant never sees the Saturday hours, even though they are listed on the page.

What Google points to

Google's May 2026 guidance on generative AI features mentions, for sites that want agent-driven experiences, its agent-friendly web guidance on web.dev and the Universal Commerce Protocol, an open way for agents to find, compare and buy from shops. It also points shops to Merchant Center feeds and local businesses to Google Business Profile.

The common thread is machine-readable facts: structured data, feeds, profiles, rather than clever pages. Google is not asking you to build a new website. It is asking you to publish the facts you already have in a format a machine can read.

You can read more about how this fits with other AI visibility work in our guide to Google's generative AI guidance. The short version: structured data, feeds and profiles do most of the work.

What to do now

Start with the pages that matter most: home, services, prices, contact and booking. Make them readable without JavaScript, with the facts in text and structured data. That means services, prices, hours, locations and products written out plainly, not just drawn in a widget.

Keep booking and checkout plain. Labelled fields, no surprise steps, a confirmation page. Test a booking yourself in a private window with scripts off, or ask a friend to try it on a slow connection. If you cannot complete it easily, an agent will struggle too.

Let the crawlers in. Check your robots.txt and remove blocks on AI crawlers unless you have a good reason. Keep Google Business Profile and any product feed current, because agents often read those before they read your site.

For example, a bakery in Toronto might add its opening hours, cake prices and order form as plain text with structured data, then test the order form in a private window. That is a morning's work, and it makes the site usable for both people and agents.

What to leave for later

Implementing agent protocols by hand can wait. Platforms such as Shopify, WordPress plugins and booking systems will add support; choose ones that say they will. You do not need to write protocol code yourself.

Meanwhile, measure the part that already matters: whether assistants name your business when asked. Vasukai checks eight assistants on a schedule, from $29 a month, with a free check that needs no card. It shows whether you are named, who is named instead, and which websites the assistants cite.

That gives you a clear signal about your visibility today, while the agent-friendly plumbing gets built by your platform. Do the readable-facts work now, and let the protocols arrive with your tools.