Your Next Website Visitor May Not Be Human.

But Does Your Website Need to Change?

AI agents are beginning to search, compare and carry out tasks on behalf of customers. But before you start redesigning your websites for artificial intelligence, there's a more important question to answer: will these interactions actually generate business?

Black-and-white cartoon of a man instructing an AI agent robot to book an AI training course, with the robot displaying a completed booking confirmation.

For years, businesses have invested in websites designed to attract visitors through search engines. SEO has become an important part of website development, influencing everything from site structure and navigation to page content.

But being optimised for search engines doesn't necessarily mean a website will rank well, attract the right visitors or generate enquiries.

Now, with the development of AI agents, we're beginning to see another shift in how people might find and interact with businesses online.

Unlike a traditional search engine crawler, which primarily discovers and indexes website content, an AI agent may be able to interact with a website on behalf of a customer. It might research products, compare prices, check availability, complete forms or even initiate a booking.

This raises an interesting question for business owners. If more online activity is carried out by AI agents rather than people, should we start designing websites for machines instead of humans?

I don’t think so. I've recently reviewed two business websites that were struggling with search visibility and lead generation. Although they were different businesses, both appeared to have been designed primarily around SEO rather than the needs of their potential customers. Yet neither website achieved the search rankings the businesses expected, despite being optimised for SEO.

The thing is, your website's job isn't just to attract visitors, but to convince them to take a punt on you. To show them that it's worth taking the next step to get to know your business. How visitors interact with your website may change, but its purpose hasn't.


What Is an AI Agent, and How Is It Different from a Search Engine?

Most businesses are already familiar with search engine crawlers. Googlebot, for example, visits websites to discover content that may be included in Google's search results.

AI agents can potentially go further.

Rather than simply retrieving information, some agents can perform a sequence of actions to complete a task requested by a user.

Imagine someone planning a business trip to Leeds. Instead of visiting several hotel websites, comparing prices and checking availability themselves, they might ask an AI assistant to find suitable accommodation within a particular budget and close to their meeting location.

Depending on the agent's capabilities, the websites involved and the permissions granted by the user, the agent might be able to research options, compare availability and help complete a reservation.

The important distinction is that the customer is still making the request, but the AI system may carry out some of the work.

Traditional search interaction AI agent interaction
A person searches for a service A person asks an AI assistant to find a service
Search results present relevant websites The agent may research and compare several websites
The person visits and explores a website The agent may retrieve information or interact with the website
The person completes a form or booking The agent may assist with completing the task
Website analytics record measurable activity Some agent activity may be identifiable, while other interactions may be difficult to attribute

Of course, not every AI assistant can perform all these actions. Some primarily provide information, while others can interact with websites or connected services. Their capabilities also depend on the permissions and technical access available.

Nevertheless, the direction is worth paying attention to, particularly for businesses that rely on online enquiries, appointments, reservations or e-commerce transactions.

Does Your Website Need to Be Optimised for AI Agents?

There's already plenty of discussion about making websites more accessible to AI systems. Some recommendations focus on structured data, machine-readable content, APIs and ways for AI agents to interact with online services.

These developments are worth following, but I would be cautious about treating every new technical recommendation as something all businesses must implement immediately.

For a small business, a complete website redesign or an expensive new integration may be difficult to justify without evidence of commercial benefit.

Before considering additional investment, I would start with three questions:

  1. Are AI agents actually interacting with your website?
  2. Can they successfully complete the tasks that matter to your customers?
  3. Are those interactions producing measurable business results?

These questions help distinguish an interesting technological development from a genuine business requirement.

  1. Are AI Agents Actually Visiting Your Website?

The first step is to establish whether there is any meaningful agent activity to investigate.

This isn't necessarily as straightforward as opening Google Analytics 4 (GA4) and looking for a new traffic source called 'AI agents'.

Some AI systems identify themselves through user-agent information or other technical signals. Others may operate through browsers in ways that resemble ordinary visitors.

There is also an important difference between an AI crawler accessing your content and an agent actively completing a task on behalf of a potential customer.

A crawler reading your product information does not necessarily represent someone who intends to buy.

Where should you look?

Depending on your website's hosting environment and technical setup, useful sources of information may include:

  • Server logs: These can reveal requests from identifiable automated systems, including some AI crawlers and agents.
  • Content delivery network and security reports: Services such as Cloudflare may provide information about automated traffic and bot activity.
  • GA4: Useful for examining recorded website interactions, although it cannot reliably identify all AI agent visits.
  • Form and transaction records: These can help determine whether identifiable automated interactions are associated with enquiries or purchases.

It's important not to assume that every automated request represents an AI agent acting for a customer.

Equally, the absence of identifiable AI agent traffic doesn't prove that no agents are interacting with your website.

For now, I would treat this as an investigation rather than a new marketing KPI that every business must report on.

  1. Can AI Agents Complete the Tasks That Matter?

If you establish that AI agents are interacting with your website, the next question is whether they can successfully carry out a useful task.

This is where many of the principles of good website design remain relevant.

Think about the information a prospective customer needs before making a decision.

For a financial services business, they might want to know which services are available, who they are suitable for, whether the adviser covers their location and how to arrange an initial consultation.

For an online retailer, the questions may concern product specifications, prices, delivery charges, availability and returns.

For a hotel, customers may need accurate room descriptions, booking dates, cancellation conditions and a clear reservation process.

If this information is difficult for a person to find, it may also create problems for an AI system attempting to complete the same task.

Start with the existing customer experience

Before introducing specialist technology, review whether your website already provides:

  • Clear descriptions of products and services, written in language customers recognise.
  • Accurate pricing, availability and important terms where applicable.
  • Logical navigation and links between relevant pages.
  • Accessible forms and booking processes.
  • Consistent business information and contact details.
  • Structured data where it accurately describes the page content and is appropriate for the website.

These improvements are not exclusively about AI agents. They can also benefit human visitors, search engines and other systems that access website content.

However, a website that is easy for people to use is not automatically accessible to every AI agent.

Some agents may struggle with JavaScript-heavy interfaces, authentication requirements, CAPTCHAs, complex booking systems or pages that require several interactions.

This is why testing matters.

Rather than assuming an agent can complete a task, try representative tasks using an agent capable of interacting with websites. Observe where it succeeds, where it fails and whether any failures reflect genuine customer problems.

For example, can the agent find a suitable product, confirm its price and reach the checkout? Can it locate the correct service and start an enquiry?

Testing should use appropriate permissions and avoid submitting real purchases, personal information or unwanted enquiries without authorisation.

A failed test doesn't automatically justify rebuilding the website. It simply identifies something worth investigating.

  1. Are AI Agent Interactions Generating Genuine Business Results?

This is the question I consider most important.

It's easy to become distracted by new metrics when technology changes. We saw this with website traffic, social media impressions and, more recently, AI search visibility.

But more activity doesn't necessarily mean better business performance.

Suppose your website receives 500 identifiable requests from AI systems in a month.

That number might sound impressive. However, if most requests involve crawling content and none can be connected to meaningful customer activity, what does it actually tell you about your marketing performance?

Very little.

Now imagine that a smaller number of identifiable agent interactions lead to completed reservations or qualified enquiries.

That would be a much more interesting development, provided those outcomes can be verified and reasonably attributed.


Measure outcomes, not just activity

The same commercial measures we use for other marketing channels remain relevant.

Metric What it helps establish
Genuine enquiries Whether interactions lead to legitimate customer interest
Qualified leads Whether enquiries match the business's target customers
Completed bookings Whether a customer successfully reserves a service
Purchases and revenue Whether transactions generate measurable sales
Cost per acquisition Whether the cost of generating a customer is commercially viable
Conversion rate Whether identifiable interactions result in meaningful actions

There is, however, a measurement challenge.

Existing analytics platforms may not capture every agent interaction. Some agents may access information without executing conventional browser tracking, while others may complete actions through interfaces that are not visible as ordinary website sessions.

Attributing a booking or sale to an AI agent may therefore require additional technical evidence, such as identifiable request logs, transaction identifiers or integration records.

Even then, it is important to distinguish between an agent that completed a transaction and one that simply helped a customer research their options.

The objective shouldn't be to report as many AI agent visits as possible. It should be to establish whether those interactions contribute to genuine commercial outcomes.

What Does the Latest Research Tell Us?

There is already evidence that AI agents are interacting with commercially important parts of websites.

In October 2026, Microsoft Clarity published research examining more than 5.2 million sessions involving AI agents across 230,511 website projects.

The research looked at activity recorded between 13 August and 9 September 2026, including form submissions, checkout interactions, appointment bookings and other website actions.

Bar chart showing actions recorded in AI agent website sessions, including checkout (25.21%), form submissions (19.85%) and appointment bookings (8.42%).
These figures do not prove that an AI agent independently completed a purchase or booking. Sessions may include human intervention.

What I find particularly interesting is that these interactions aren't limited to information gathering. They're also occurring in areas of websites associated with enquiries, bookings and purchases.

However, the findings need to be interpreted carefully.

Microsoft acknowledges that its data cannot always distinguish between actions completed autonomously by an agent and those involving a human taking over the session. A checkout event also doesn't necessarily mean a successful payment.

The research demonstrates that AI agents are reaching commercially important website functions. It doesn't yet establish how much additional revenue those interactions generate.

For businesses, that's an important distinction.

You can read the full research here: What AI Agents Do on Websites: Insights From 5.2 Million Sessions.

What Should Businesses Do Now?

I wouldn't recommend creating a separate AI agent optimisation strategy for every business just yet.

Instead, I would incorporate a few practical checks into your existing website and marketing reviews.

First, establish what you can measure. Check whether your current analytics, hosting or security tools can identify agent-related activity. Microsoft Clarity, for example, now offers an Agent Activity Dashboard for projects with detected agentic traffic.

Second, review your most commercially important website tasks. Focus on activities such as submitting an enquiry, finding a product, requesting a quote or making a booking. Check whether the information is clear and the process works as intended.

Third, test where appropriate. If AI agents are relevant to your customers, conduct a few controlled tests to see whether they can find information and progress through an important task. Document any issues before deciding whether technical changes are necessary.

Fourth, protect your customers and your business. Not every automated system should be given unrestricted access. Security, privacy, fraud prevention and customer authorisation still matter. The objective is to support legitimate activity without weakening existing safeguards.

Finally, review the commercial evidence before committing additional budget. If identifiable agent activity begins contributing to genuine enquiries or transactions, you have a stronger reason to invest in improving that experience.

For many businesses, the most useful work may still involve improving existing website content, fixing broken forms, simplifying navigation and making sure conversion tracking is accurate.

These are improvements worth making regardless of how AI agents develop.

The Visitor May Change, but the Business Objective Remains the Same

AI agents introduce an interesting development in how customers discover and interact with businesses online.

Some customers may continue to visit websites themselves. Others may ask an AI assistant to research options, compare suppliers or help complete a transaction.

Businesses may eventually need to consider both types of interaction when reviewing website performance.

But I don't think that means abandoning the principles of good digital marketing or rushing to adopt every new AI-related recommendation.

A website still needs to communicate clearly, provide accurate information and make it easy for customers to take the next step.

And whether that step is completed by a person or an AI agent acting on their behalf, the commercial question remains the same:

Did the interaction produce something valuable for the business? 

Until we can answer that question reliably, I would focus on what matters most: meeting the needs of our customers and measuring the commercial results, rather than optimising for a new technology simply because it exists.