AI Agents Are Visiting Your Website. Can They Actually Use It?

Picture a prospect who asks their phone to find three bookkeepers in Vancouver, request a quote from each, and report back by Friday. A few years ago, that was a wish. Now it’s a task that an AI agent can do. 

The first sign that your business was part of an agentic AI process might be a quote request that arrives through your contact form, filled in by software.

Meta launched its personal AI agent, Muse, in September, and it’s available in Canada and the US. OpenAI has a similar product called dots. These agents don’t stop at reading your website for an answer. They click the buttons, fill in the fields, and try to finish the job.

That’s where AI agent SEO comes in. It extends search optimization past discovery, so that automated assistants can find your business, understand what you offer, and complete actions on your website. 

This post covers what agents see when they land on your pages, where they get stuck, and the practical fixes that help agents and people alike.

What is an AI agent, and how is it different from ChatGPT search?

An AI agent takes a task from a person and tries to complete it on the web. This means it doesn’t just read your website to answer a question—it also tries to use your website to book an appointment, request a quote, or make a purchase.

For most of the web’s history, the goal was to get found. SEO earned you a spot in Google’s results. More recently, generative engine optimization (GEO) added a second goal: getting cited when ChatGPT, Perplexity, or Google’s AI Overviews answer a question.

Agents add a third step. SEO helps people find your website. AI agent SEO helps software understand it and get things done.

Muse and dots are the two names to know right now. Both can run their own browser on a cloud computer, take a request from a person, and carry it out across the web, from researching a product to filling in a form and checking out. Slobodan Manić’s breakdown of Muse and dots for Search Engine Journal is the most thorough look at how they work so far, and we lean on his findings throughout this post.

How do AI agents get to your website?

AI agents reach a business in one of two ways: through a direct connection the platform has built with that business, or by opening the business’s website in their own browser and working through it the way a person would.

The direct route is mostly for large retailers and platforms. Meta has built connectors with major brands and the Shopify catalogue, and Shopify shares products with Meta by default. If you run a Shopify store, you may already be in the connected group without having chosen it, so it’s worth checking your sales channel settings.

Everyone else gets the browser route. For the service businesses, professional firms, and not-for-profits that we work with, that’s the door that matters most today, even if direct connections reach smaller organizations as these platforms add partners. With the browser route, the agent lands on your site, reads what’s there, and works with whatever the page gives it.

The agent still has to find you before it can visit, either by name or by searching. Neither Meta nor OpenAI has said which search engine its agent uses, which is one more reason to make sure you’re visible beyond Google, in Bing and Perplexity too.

What does an AI agent actually see on your website?

AI agents read websites in different ways. Meta says Muse works through the accessibility tree, the simplified version of a page that screen readers use, so it relies on names, roles, and text rather than your colours, layout, or visual cues.

In its write-up on how Muse was built for safety, Meta explains that the agent’s browser gets a snapshot of this tree instead of the page’s code. Other agents may also work from screenshots or the page’s underlying code, and Google’s guidance for site owners lists all three approaches. Meta’s method is the one documented in the most detail, which makes it the most useful lens for checking your own site.

Chrome builds the tree from your page and, according to Chrome’s own documentation, keeps only the elements that matter to a screen reader. Each item gets a role (button, link, heading), a name, and a state.

The styling gap is the one that catches people. In the same SEJ analysis, Manić checked a product page on a large Shopify store where six of seven sizes were sold out. A shopper saw them greyed out and crossed out. The accessibility tree showed seven identical, available-looking buttons. 

The same thing can happen on a booking calendar where unavailable appointment times are only greyed out, leaving an agent to try each slot one at a time.

In a recent audit for a B2B software company, we found a membership link in the navigation that appeared twice, and one of the two had no label at all. A sighted visitor wouldn’t notice. A screen reader would announce it as “link” with nothing after it, leaving the user to guess where it goes. The same site used “Title” as the alt text on its customer logos, so the section meant to build trust gave screen-reader users no customer names at all. 

Neither fix is big: one properly labelled link instead of a duplicate, and alt text that names each customer. Both are also exactly what an agent like Muse would stumble on, since it reads the same accessibility tree that a screen reader does.

How to See Your Website the Way an Agent Does

You can check this yourself in a few minutes with Chrome.

1. Open the page you want to test, right-click anywhere, and choose Inspect.
2. In the Elements panel, click the Accessibility tab. If you don’t see it, look under the double-arrow menu beside the other tabs.
3. Turn on the “Show accessibility tree” toggle. The full-page accessibility tree replaces the usual page code in the Elements panel.
4. Walk through your contact, quote, or booking form and read the name of every field and button.

If an item shows up as “button” with no name, or a field has no label, a person using a screen reader and an AI agent may both struggle to tell what it does.

7 ways to make your website work for AI agents

Making a website work for AI agents comes down to two jobs: making sure software can understand your pages, and making sure it can tell whether it finished what it started.

1. Label Every Form Field and Button

Every field and button that an agent needs to use should have a name it can read. That covers contact forms, quote requests, booking tools, donation forms, and newsletter signups.

Placeholder text inside a field isn’t a label (see why in our article on improving website accessibility), and neither is an icon on its own. This is one of the most common problems on the web: the 2026 WebAIM Million report found missing form input labels on 51% of the top one million home pages.

2. Put Key Facts in Visible Text

If a customer would need it to make a decision, it should appear as plain text on the page. This means content like hours, service areas, starting rates, eligibility, turnaround times, and availability.

Structured data still helps Google understand your business, so keep it. Just don’t let it be the only place that those facts live, since an agent reading the accessibility tree won’t see it. The same goes for prices or details baked into a graphic.

3. Use Proper Headings and Descriptive Link Text

Your heading structure becomes the agent’s outline of the page, so headings need to be actual heading tags in a logical order, not text styled to look big. Link and button text matters just as much. “Learn more” and “Click here” name nothing, while “Request a quote” tells both people and agents exactly what happens next. 

Our guide to CTA best practices for UX and accessibility covers this in more detail.

4. Check What’s Hidden in Tabs, Accordions, and Pop-Ups

Whether an agent can read collapsed content depends on how your theme hides it. Some methods keep the text readable in the accessibility tree, while others remove it until someone clicks.

You don’t need to know which method your site uses. Ask your developer, and in the meantime, keep essential facts like pricing and service details out of collapsed sections.

5. Test Your Third-Party Widgets and CAPTCHAs

Embedded booking tools, payment forms, and scheduling widgets come from other companies, and their accessibility varies widely. Run them through the same accessibility tree check as the rest of your site.

CAPTCHAs deserve a second look too. They’re designed to stop automated visitors, which may be exactly what you want, or it may mean a qualified lead never reaches you. If your site uses Google’s tool, our post on recent reCAPTCHA changes explains what’s shifted.

6. Make Errors and Confirmations Clear

An agent needs to know whether an action worked. If a form submission fails, the error message should say what went wrong and which field needs attention, in text rather than a red outline around the field.

When a quote request or booking goes through, the page should say so plainly, with a confirmation message or a dedicated thank-you page. That’s already a good UX and accessibility practice, and it gives an automated visitor the signal it needs to report back to the person, fix the problem, or stop retrying.

7. Keep llms.txt and Text-Only Versions in Perspective

Files like llms.txt and stripped-down text copies of your pages have been proposed as ways to help AI tools read your content, but their benefits for browser-based agents haven’t been shown. They also leave out the forms, buttons, and booking flows that an agent came to use, so treat them as a small extra rather than your plan for AI agents. The page itself is what the agent works with.

Should you block AI agents from your website?

Robots.txt isn’t a reliable way to control personal AI agents like Muse, because they don’t identify themselves as separate crawlers. The decision moves to your firewall or bot protection, where legitimate agent visits have to be weighed against spam and abuse.

Robots.txt has always been an honour system that works by naming specific crawlers. The robots.txt standard is clear that its rules aren’t a form of access control. Neither Meta nor OpenAI has published a name for these agents’ browsers, and Meta says Muse’s browsing shows up as the user’s own activity, which leaves nothing specific to put in the file.

Large retailers can push back in other ways. Amazon blocked Muse through its terms of use, citing concerns that included the agent not identifying itself, as GeekWire reported.

For a bookkeeping firm or a physio clinic, an agent submitting your contact form is usually being directed by a person who wants your services and has asked software to help. Automated traffic also includes spam bots and form abuse, so this isn’t a case for dropping your defenses. The goal is to know what your security rules are doing. 

Ask your host or developer how your firewall and form protection treat automated browsers, and make the call on purpose.

How will AI agent visits show up in your analytics?

AI agent activity can be hard to tell apart from human traffic, because some agents trigger ordinary page views and conversions while others may never appear in your website analytics at all.

Depending on how an agent works, its activity can land in your reports in a few different ways:

• An agent browses your site and fills in a form: likely to look like a regular visit and conversion, since Meta and OpenAI haven’t published an identifier to filter on. Whether it’s counted at all depends on factors like consent settings and how your tracking loads.
• Someone clicks a link inside an AI assistant: Google Analytics 4 has sorted these into an AI Assistant channel since May 2026, as long as the app passes along where the visit came from.
• A purchase is completed inside an AI app: it may never load your pages, so it can show up only in your ecommerce platform’s order records.

For service businesses, the practical step is low-tech. Ask new leads how they found you, and add “I used an AI assistant” as an option on your intake form. Pair that with the rest of your website performance measures and you’ll start to see a pattern long before the analytics tools catch up.

Is an agent-ready website just an accessible website?

Accessibility is the foundation of an agent-ready website, but agent readiness also depends on whether an automated visitor can finish the task, which is why clear errors and confirmations matter as much as labels.

The foundation is where most of the work sits, and it’s overdue for most of the web. The same WebAIM report found detectable accessibility failures on 95.9% of the top one million home pages in 2026, up from the year before.

Fixing those issues helps the people using assistive technology on your site today. It also keeps you in contention as more customers send an agent to compare their options. If accessibility hasn’t been on your radar, our article on why accessibility matters for every website is a good place to start.

Getting your website ready for AI agents

Getting your website ready for AI agents means making sure the pages where people take action, like forms, bookings, and checkouts, can be read and completed by software that relies on labels and text more than visual design.

Back to the example of a prospect asking for three bookkeeping quotes. Their agent will find a handful of firms, read their websites, and try to submit a form on each one. The firms whose forms work get a lead. The others never learn that they were considered.

None of this requires a rebuild. Start with the accessibility tree check on your highest-value form, fix what has no name, confirm that success and error messages are clear, and move the facts that matter into plain text.

Next steps

If you’d like a second set of eyes, our SEO consulting team can review your site for search visibility, AI citations, and the accessibility issues that trip up forms. Prefer to start on your own? Run through the Bad Website Checklist, then test your contact form with the steps above.