Your site has more audiences now: Understanding MX and the machine-readable web

MX is one of those terms that’s starting to mean different things depending on who you ask. It can mean machine experience, multi-experience, multisensory design, or some other conference buzzword definition. In this article, I’m referring to the most commercially relevant definition: Machine Experience.

More specifically, I’m talking about whether ChatGPT, Claude, or the growing number of AI models out in the wild can make sense of your site and digital assets. There are now millions of these models, with very mixed quality, and many won’t be used directly by people. They’ll sit behind agents and tools, interpreting your content on someone else’s behalf. With that in mind, can they tell what you stand for, what you sell, what matters most on the page, and where to go next? Or are you expecting them to piece together an answer from inconsistent fragments across multiple sources?

For brands that rely on digital discovery and digital commerce, don’t think this is tomorrow’s challenge, I can tell you now, it’s already starting to impact your numbers.

A new chapter

For the last 30 years, the deal was simple enough: let search engines crawl your content, and they will send you the traffic you want. However, the rise of LLMs and the influence of answer-focused experiences is stressing that agreement. That pact is still in place, but the journey is getting messier as search engines, platforms and models start to demand a bigger slice of the pie. More of the discovery, comparison, and summarising now happens before a person lands on your site, and much of it is being done by systems rather than people.

Cloudflare’s recent reporting points in the same direction, stating AI and search crawlers are consuming more web content and increasingly returning less referral traffic. That changes what your website is doing. It’s no longer just there for customers and other stakeholders. It is also being read, interpreted and represented by systems you don’t control, and that will make their own call on what your content means.

In practice, this means designing for two very different readers. A person wants reassurance, clarity and ease. A machine wants structure, consistency and explicit meaning.

Having said that, it’s important not to overstate the challenge or complicate the solution. Google has already said there are no extra technical requirements for appearing in AI Overviews or AI Mode, and no special AI markup you need to rush onto your site. 

The fundamentals of crawlability, indexability, internal linking, useful and unique content, and structured data all remain critical.

Those same fundamentals were always about helping systems understand your site. The difference now is what happens next. It’s no longer just about being indexed or cited. Machines are starting to act on that understanding, searching, filtering, comparing and, in some cases, moving closer to transactions on a user’s behalf.

The move towards MX isn’t a reason to abandon sound SEO, accessibility, or content design. If anything, it’s a reason to take them even more seriously, as the cost of getting them wrong is now even higher.

This is why MX should not be treated as a replacement for UX. Instead, it should be seen as its partner in crime, like UX’s Torvill to MX’s Dean. Human-centred design still matters because people still compare, doubt, trust and decide. But AI is now shaping more of the journey earlier on, reading, filtering and narrowing options before a person ever reaches your website. And, in some cases, they won’t reach it at all. A user's entire experience of your brand may be mediated through a machine. That changes the role your content plays. It’s no longer only about what a customer sees on screen; it’s also about how clearly your content, data and digital estate can be consumed, interpreted and represented by AI.

If an agent has to scrape a page and infer meaning from visual layout alone, you have already made the experience harder than it needs to be. A machine should not have to guess whether a number is a price, a review score, a delivery date or a discount. The same applies to product specs buried in images, vague labels, or key details trapped within components that look fine to humans but are difficult for machines to parse. This is where machine readability becomes commercially relevant. Structured content, clean markup, consistent labelling, schema, feeds and APIs all make it easier for systems to understand what your site is actually saying and where you fit in the digital authority rankings. 

What machine-readable really means

So, what does machine-readable mean in practice? Mostly, it means getting the basics right. There’s no raft of new AI requirements; it’s still just about maintaining good web practice. Google’s guidance says the same thing – that SEO fundamentals still apply. MDN, Mozilla’s developer team, along with the W3C, the organisation that defines web standards, also makes the same point: semantic HTML and clear, readable data are how systems understand a page. 

Teams working in accessibility have spent years pushing for clearer structure, better semantics and more reliable labelling. The aim was always to make pages understandable to systems that do not see or navigate like a fully sighted user with a mouse and plenty of patience. Now AI is reading the same web, and suddenly that discipline has acquired a new commercial significance.

First impact

This evolution will hit commerce teams before most other parts of the market, because product details are full of information that must be read precisely. Price, availability, and delivery windows are all important. As are returns info, variants, dimensions, review counts, promo terms, shipping thresholds, and subscription options. Get any of them wrong, and you’re reducing your ability to leverage the opportunity and allowing others to gain the advantage.

As humans, we are broadly tolerant of mess. We can scan, compare, cross-check, contextualise, and make reasonable assumptions. Machines are far more literal. Give them a well-formed product page, and they can usually identify the essentials. Force them to infer the same information from layout, inconsistent naming, tabs, accordions or text trapped inside images, and the odds of a bad answer rise.

A good example is multi-step checkouts. When we work through a buying process online, we understand that delivery is likely to be added later or that thresholds apply. A machine often won’t. It will take the first price it finds and treat that as the total, because the rest is buried further down the flow or revealed conditionally.

Google’s product documentation is really clear on this: Structured product markup exists so systems can recognise things like price, availability and key details in a standard format. There is separate documentation for shipping and return policy markup for the same reason – the goal is clarity and consistency.

That has obvious implications for search and shopping, AI summaries, and agent-led workflows. Can a system tell the difference between a sale price and a standard price? Does it understand that one size is in stock and another is not? Does it pick up the correct delivery promise, or drag through an outdated one from somewhere buried in your 2018 terms and conditions that you thought you’d unpublished? Those might seem trivial until they start to impact whether someone chooses to click, compare or buy.

There is also a quieter problem. Machines do not always fail in dramatic ways. Sometimes they get things slightly wrong, and that can be worse, because it is harder to spot. You do not get an alert telling you an assistant has misread your delivery threshold. You just get the effect. In some cases, the issue isn’t so much the absence of information as it is the distortion of it.

When you lose control

That loss of control can have big implications for your brand. The moment a prospect encounters your brand inside Google’s AI overview or an LLM interface, they are no longer in a brand-controlled environment. They are not seeing your carefully structured page in the order you intended, with the emphasis you chose. Instead, audiences are seeing a version of your business reconstructed from whatever the model can find, interpret and prioritise.

That means key points can be compressed, reordered or ignored entirely. Supporting detail can be either elevated or removed, and critical context can be lost. What comes back might be broadly right, but still not the way you define it on your own channels.

This is where MX stops being a technical issue and starts becoming a brand and commercial one. If different systems can pull back different versions of your business, you lose control of how you are compared, recommended, and chosen.

Ready for your closeup?

At this point, lots of SEO teams would be tempted to play with the schema, feeds or APIs. That’s understandable, but also not necessarily the best starting point. The underlying problem is often that the content itself is not structured properly.

A lot of sites, especially lower-cost builds, are still designed visually first, with the content model bolted on afterwards. I’ve seen it dozens of times where a single “body copy” editor field in the CMS ends up carrying hacked-in HTML to display product facts, selling points, disclaimers, links, FAQs, and anything else that needs somewhere to live. The page is then arranged so it looks acceptable on the front end. That works well enough when a human is digesting it, but for a machine, it can resemble a hot mess that is impossible to decipher.

The answer is to separate different things properly. Product name, descriptions, specifications and features, price and offer terms, shipping, reviews, author, category, and FAQs should all be distinct in the back end. Once those are clear, everything else becomes easier to manage, reuse and present consistently across channels.

If you’re reading this thinking “we’re fine, we don’t sell jumpers, holidays or phone cases”, you’re wrong. This isn’t limited to the world of commerce. The need for structure and consistency applies equally to service businesses’ proposition pages, case studies, and thought leadership. 

In fact, these sites can actually be worse. Brands in these spaces often lean into long-form content, with key points buried in paragraphs rather than clearly defined in the back end. That might read well to a person, but it makes it harder for a machine to understand what matters and what the page is trying to say.

Another risk is consistency. If the same claim appears one way in the page copy, another way in metadata, and slightly differently again in a downloadable PDF, a human might not notice, but a machine is more likely to judge that negatively and look elsewhere for something it can validate. That’s where this becomes a question of reliability as much as visibility. If a system encounters your brand in multiple places, does it come away with the same understanding each time? In many cases, the honest answer is no.

Where APIs fit

APIs matter most when the information is dynamic or critical. Things like stock levels, delivery estimates, booking availability or pricing logic can all do more damage if you get them slightly wrong compared to them being absent completely. Direct, structured access removes the need for interpretation and lowers the risk of bad assumptions.

Championing consistency and structure isn’t controversial. Most systems prefer explicit information when they can get it; it’s just that humans are much better at coping with ambiguity, so we get lazy and let things slip, knowing the user can still work it out easily enough. Machines, on the other hand, reward clarity and punish anything that is potentially confusing.

Still, most businesses do not need to turn their site into a sprawling API strategy overnight. A much more sensible starting point is to work out which pieces of information genuinely matter, where those facts live, how many times they are duplicated, and where meaning is most likely to get lost. That tends to be a more productive conversation than immediately buying into whatever new AI framework is doing the rounds.

Heavyweight UX

None of what I have been talking about here removes the fact that people still make the final decision in most journeys. The numbers will evolve depending on how good AI gets at understanding its audience’s objectives, but to a lesser or greater degree, people will still look for brand, reassurance, clarity, speed and trust. They will still respond to good design and messaging, and they will still walk away from bad experiences. Even if AI is shaping more of the shortlist, humans ultimately decide whether they believe the brand, whether the offer feels right, and whether they want to part with their hard-earned cash.

So, the obvious challenge is to make the experience hold together from both directions. A person should arrive and find something clear, persuasive and easy to use. A machine should be able to identify the core meaning without improvising. 

IDHL and MTM clients already have a head start here. Not because somebody gave our designers and developers a look into the future, but because we have always focused on the fundamentals. Clean structure, sensible content modelling, consistent data, and decent publishing habits have always been critical; it’s just that the penalties for not following this approach are becoming much easier to see.

A sensible audit

When we’re reviewing a site through this lens, we start with a few simple questions.

• Can a machine tell what this site is about within seconds?
• Are theimportant factsexplicit, or are they implied through layout?
• Is the same information consistent across page copy, metadata, schema, feeds, supporting assets, and channels?
• Would the page still make sense if the styling disappeared?
• Are any key details trapped inside images, tabs,accordionsor components that are hard to parse?
• Could an assistant summarise the page accurately without having to fill in too many blanks?

That last question is relevant because every time a system has to fill in the gaps, it has room to get it wrong. Sometimes that is harmless. Sometimes it changes what gets shown, recommended or selected. We’ve all heard of AI hallucinating and outright making shit up, so if the information is commercially important, guessing is a poor operating model.

Why performance teams should care

If it’s not clear by now why all this matters, this article has missed its mark. But to recap, MX is already affecting performance marketing, organic search, CRO, content, and measurement across industries, audiences, and territories.

The old model was simple. I get it. A person searched, clicked, landed, and hopefully converted. The new model is less linear but still pretty straightforward. Discovery may happen within an AI-generated answer, comparison before the visit, or a recommendation may be shaped by product data from elsewhere. All these things could be happening right now, so in some cases, the click never comes at all.

‘Crawl-to-refer’ data points in the same direction. AI and search platforms are consuming large volumes of content, often with a striking imbalance between the amount they crawl and the traffic they return. The exact numbers move around by platform and time, but the broader pattern is clear enough to take seriously. 

For performance teams, the risk is equally straightforward and manageable. If you still think of your site purely as a destination for human sessions, you’re missing a growing slice of the pie and increasing risk. None of the required work is especially exotic, and none of it sits outside the normal concerns of a good digital team, so you shouldn’t be daunted by the task at hand. 

At the end of the day, MX is already here, whether people like the label or not. The more useful question is whether your site is making life easy for machines or leaving them to guess. If it is the latter, that could become a commercial problem surprisingly quickly.