Guide your business into the agentic unknown

Ben says: “Guide your C-suite into the great agentic unknown, because if you don't, then nobody else is going to.”

What is the great agentic unknown?

“I'm talking about the nature of AI agents using websites in the same way that human customers conventionally would. They are researching a topic on behalf of a user based on a prompt, and it can go from researching all the way to getting quotes or even adding things to a shopping cart.

In some ways, it feels like a stretch for SEO practitioners and is probably taking some people beyond their comfort zone. However, nobody else in a business is in a better place to guide leadership into optimizing for this opportunity than SEO.”

How do you explain this to stakeholders without sharing too much technical information?

“It's an emerging technology, so you need to treat it as a hygiene factor, and an opportunity to do more business. It's not helpful to position it with too much hype or fearmongering.

I've seen some LinkedIn chatter saying that agentic AI is going to replace organic search, and there are going to be more agents shopping online than human users. I'm sure that will be the case eventually, but it's nowhere near the case now.

It wouldn't make sense to position this as a horror story. It makes more sense to position it as a hygiene factor. This is a way in which people can theoretically do business with you, so you should enable it.”

What needs to change on a website in order to enable this?

“There are two sets of considerations that you can use to look at this.

The first is more comfortable for an SEO practitioner because it's about how an agent may go about researching a topic: how it might determine what queries are useful for a user's prompting, and then go and find that information. The skills required for this have a lot in common with classic search. It’s about doing keyword research and providing good, structured content to meet that search intent.

The second is an area that might be a bit new for some people, and it's the notion of testing your conversion funnel as an agent. To do this, you can prompt something emerging like ChatGPT’s agent mode to try and convert through your funnel.

You can prompt it in such a way that it may make some common mistakes, and you can audit your site against those common mistakes.

I’ll give you a technical example. I've seen ChatGPT's agent mode struggle with layout shift in content. It tries to add an item to cart, but a dynamic message pops up where the ‘Add to Cart’ button used to be, and it struggles to conduct the next relevant step. I've also seen it struggle to identify the relationship between two subdomains from one company, so it can’t move through the customer journey seamlessly.

You can prompt an agent to try and conduct a transaction journey, and while you are doing that, you also need to think about the common mistakes and bad prompts that a potential customer may use that could cause the agent to struggle.

If someone is filling in a lead form and they want to refine their form, does the layout change? A proficient and competent customer may go through an inquiry form in one go, but a less proficient customer may take a couple of tries at refining that information. If an agent's replicating that behaviour, can it do it in the way that a human can?

You need to challenge your conversion journey through the eyes of an agent that is being badly prompted.”

What is Google Opal, and how significant do you think it might be?

“Opal is a disruptor to what's become the market standard technology. At the moment, in late 2025, n8n is what has emerged as the market standard for people who want to create their own agent.

The most commonly used tool to create AI agents at the moment is probably ChatGPT in agentic mode, which has become the market standard for creating completely customised agents, but it's not affordable to everyone. It's affordable to a professional, passionate hobbyist who is willing to invest in understanding agentic AI.

However, Google's Opal is a product which has many of the same features (perhaps a little more basic), but it enables people to experiment with creating agents or agentic-like automations for themselves. At the time of speaking, we have a very early iteration of the product. It’s likely to develop, and Google are well-placed to become a market leader.

The main use cases for AI agents are going to the web, browsing the web, and completing actions. Google have produced Opal, they have Chrome, and they have Google Search. Logically, they are best placed to dominate this space.”

Do you have to structure web pages to optimize for agents instead of humans?

“We should consider our websites very differently. We’re still at such an early stage when it comes to optimizing for agentic AI, and proficient agentic AI already makes websites look a bit hostile.

A publisher publishes a website based on a set of researched customer journeys and creates ways in which users can move through the most common types of funnel. Agentic AI, however, is creating every journey completely bespoke for the individual user. For even the best UX practitioners, the product of their work suddenly looks less successful compared to the product of a custom journey produced by agentic AI.

It's easy to imagine a future in which there are as many bot users of websites carrying out actions on behalf of users as there are actual human users. As such, a website should be equally accessible for a bot as it is for a user.

As SEO practitioners, we've spent years telling people to design and produce content for humans: ‘Google's algorithms are really smart. Don't worry about the technical SEO. Technical SEO comes into its own on enterprise-level sites.’ My advice is to rein that in just a little bit. A website that is being prepared for agentic AI should be equally designed for bots and for users.

These agents struggle with layout shift, they forget where buttons are, and struggle to see the relationship between different subdomains. These are classic technical SEO issues.”

Is there software that can test how AI agents visit your website, and help you understand the questions users might be inputting to AI search platforms?

“On an individual level, something like ChatGPT's agent mode will give a slow frame rate screen recording of the actions it's trying to carry out. It moves quickly, but you can see what something like this is trying to do.

To do this at scale, there's no reason why something that's capturing a user session recording would distinguish between an agent user and a human user – in principle. A screen recording can be captured from a user agent regardless.

Now, for analytics, this is potentially problematic. I've seen examples of agentic AI trying to complete journeys on a website, running into trouble, and then asking a user to confirm whether what it's doing is correct. For example, if it’s getting a quote for insurance and it’s asked a question like, ‘Are you the person who this policy will be for?’, it will ask the user to clarify. However, instead of picking up the journey where it left off, a typical AI agent will restart that whole journey.

For any web events you're tracking, you may have fired an event as a ‘journey’ having started. Then, the agent asked the user to clarify something and restarted that journey, potentially firing off a tag again.

As we see increased use of AI agents from individuals, we will see inflated traffic, and it will become a headache that the industry will have to settle into a rubric for managing. Traffic patterns will artificially go up, and certain tracked events will artificially inflate as well.

Picture a human user trying to fetch quotes. They’re doing research and price comparison, they look at two or three competitors manually, and it'll take half an hour. If they’ve got an agent that can do it for them, why stop at two or three? Why not get five, six, seven, or ten?

If you set an agent off to perform a task like this, you've immediately inflated the amount of web traffic being driven by one person. You're still only going to convert once, so the value of each of those searches is lower than it would have been in a pre-agentic era.”

Ben, what’s the key takeaway from the tip you shared today?

“Audit your conversion journey for AI agents.

Can an AI agent complete your conversion journey as easily as a human user can?”

Ben Howe is Head of Digital Global Healthcare at AXA Health International. Find out more over at AXAGlobalHealthcare.com.