Machine Media is here. The playbook is two years behind.

The data stopped being subtle months ago. AI bot traffic on Cloudflare's network grew 187% in 2025 while human traffic grew 3.1%. Agentic AI traffic alone grew 7,851% year over year. Global publisher Google traffic fell 33% in 2025. Small publishers lost 60% of their referral traffic in two years. Some lost 90%.

Mike King has been calling this Machine Media. Rand Fishkin has been writing eulogies for the open web. Wil Reynolds spent his SEO Week keynote telling a room of marketers they have to decide what kind of marketer they want to be. The thesis, across all of them, is roughly the same: the architecture the web was built on — humans visiting websites — is breaking down, and the channels we've optimised for two decades are being absorbed into a single agent-mediated layer.

I think they're broadly right. I also think the industry response to all of this has been remarkably weak, and that weakness is the actual story.

Because here's what's strange. The diagnosis is correct. The numbers are real. The infrastructure shifts are documented. And yet the practical advice trailing the diagnosis is, almost without exception, two years behind where it needs to be. "Build inimitable products." "Become an authority." "Invest in brand." All true. All also the advice you would have given in 2019. None of it engages with the specific operational changes a UK business actually needs to make in the next 90 days to remain visible inside an agent-mediated web.

This piece is an attempt to close that gap. Not the philosophical gap — the operational one. If Machine Media is the medium, what does the work actually look like? Who does it? What do you stop doing? What do you start measuring? And what does it mean for a small or mid-sized business that doesn't have an enterprise martech stack and can't pivot its entire org chart in a quarter?

That's the question. The grand-statement pieces aren't answering it. So let's answer it.

The diagnosis everyone agrees on

Strip the rhetoric away and the consensus position looks like this.

For 25 years, the web operated on a click economy. A user issued a query, a search engine returned ranked links, the user clicked one, landed on a website, and the website's owner had an opportunity to convert that visit into something economically meaningful. SEO existed to maximise the probability of being the link clicked. Content existed to maximise the probability of converting the visit. Analytics existed to attribute the visit to a channel. The entire industry — agencies, tools, training, conferences — was built around that loop.

That loop is now broken in three specific places.

First, the search interface increasingly answers the query directly. AI Overviews, AI Mode, ChatGPT, Perplexity, Claude, Gemini. The user gets the answer without visiting the page. Google's own data, as reported in the I/O coverage, shows AI Mode has surpassed one billion monthly users with queries more than doubling every quarter.

Second, the agent increasingly completes the task on the user's behalf. Universal Cart, agentic checkout, the Universal Commerce Protocol, Comet's logged-in shopping flow. The transaction happens without the user visiting the merchant's site. The Amazon v Perplexity case at the Ninth Circuit on June 11 will decide whether the agent counts as an authorised visitor at all, which is a question nobody was asking 18 months ago and is now the most consequential question in digital law.

Third, the data that used to flow back to the merchant — what the customer searched for, what they considered, why they chose what they chose — increasingly stops at the agent. Armando Roggio's framing in the I/O coverage was exact: merchants still own the transaction, but not the purchase intent or product discovery. The visit was the data. No visit, no data.

That's the loop. And we built it.

So far, so consensus. The disagreement starts when you ask what to do about it.

Where the industry response is failing

The dominant response, in the pieces this week and across the conference circuit, has been to argue for some combination of four things: build a stronger brand, create more authoritative content, make your product genuinely differentiated, and invest in earned media so AI systems cite you. All of which is correct. None of which is operational.

The industry is articulating the problem at the altitude of the keynote and the practitioner is left to translate.

Telling a UK plumbing business with eight engineers and a marketing manager to "build an inimitable product" is not advice. It's a sentiment. The product is plumbing. The differentiation is reliability, pricing, geographic coverage, and the engineer who turns up on time. That's already inimitable in a way an LLM can't replicate. The question isn't whether the business has something worth recommending — most do. The question is what the business should be doing operationally so that the agent recommending plumbers in Bristol at 11pm on a Tuesday recommends them.

Telling a B2B SaaS company with a small content team to "become the authority in your niche" assumes they weren't already trying. They were. The advice that mattered five years ago was: publish useful long-form content, earn backlinks, build a community, win conference talks. That advice still applies. But it doesn't address the specific change, which is that the surface where authority gets converted into business outcomes is no longer their website.

The grand-statement pieces are written for an audience that runs strategy. The practical question — what does the work look like on Tuesday morning — is being skipped.

The industry is articulating the problem at the altitude of the keynote and the practitioner is left to translate.

So let's do the translation.

Pillar 1: Treat your website as a data source, not a destination

This is the conceptual shift that has to happen first, and it's the one most businesses haven't made. The website used to be the storefront. In the Machine Media model, the website is one of several places your information lives, and increasingly it's not the place the customer encounters it.

Operationally, this means a few specific things.

Your content has to be parseable by machines, not just readable by humans. That sounds obvious until you look at what most business websites actually publish — paragraphs of prose with the key facts buried in the third sentence, prices listed only on a quote-request form, opening hours embedded in an image, service areas described in marketing language rather than postcodes. An LLM crawling that page extracts less than half of what a human reader would.

Your structured data has to actually exist and be accurate. Product schema, FAQ schema, LocalBusiness schema, Service schema. The schema markup conversation in the SEO industry is genuinely two years behind. Most agencies still treat structured data as a nice-to-have. In a Machine Media world, it's the difference between being legible to an agent and being invisible to one.

Your APIs and feeds matter as much as your pages. If you sell products, a clean product feed with accurate inventory, pricing, and attributes is now a primary marketing asset. If you provide services, a structured service catalogue with geographic coverage, response times, and capabilities is the equivalent. Mike King's framing — that your "website" becomes a data source, which could be your site, an API, an MCP server, a structured feed — is exactly right. The question for most businesses is which of those they're set up to provide.

This isn't a strategy. It's a hygiene programme. It's also the thing most businesses can start on Monday without a board meeting.

Pillar 2: Earned media is the new ranking signal

The retrieval-and-citation research from late 2025 onwards has been remarkably consistent on one point: AI systems cite brands that have been mentioned in places those systems trust. Reddit. Wikipedia. Industry publications. Trade press. News. Academic content. The branded-search data Tom Capper presented at SEO Week — that branded search now predicts rankings better than domain authority — points in the same direction from a different angle.

What this means operationally is that the PR and digital PR functions, which have been treated as adjacent to SEO for years, are now central to it. Getting your business mentioned, in context, on a site an LLM trusts is worth more than ten blog posts on your own domain. That was already true. It's now overwhelmingly true.

The specific work this implies:

Build relationships with the trade press in your sector. Most UK industries have two or three publications that LLMs treat as authoritative. Get quoted in them. Contribute pieces to them. Be a source they call when they need a comment.

Take Reddit seriously, but properly. Not as a place to spam your link — that doesn't work and the moderators will rightly destroy you for it — but as a place where your business should be discussed in context by customers, by employees being genuinely useful, by you participating as a recognisable expert. The subreddits relevant to most UK service businesses are small, active, and read by LLMs.

Earn citations on Wikipedia where they're defensible. Most businesses can't justify a Wikipedia entry. Many can be cited in entries about their industry, their location, or topics they're genuinely expert in.

Treat your YouTube and podcast appearances as primary content, not promotional artefacts. The transcripts get indexed. The mentions get picked up. The citation graph the LLMs build runs through audio and video as much as text.

None of this is novel advice. What's changed is the weighting. Five years ago, this was a complement to your on-site SEO programme. Now it's the programme.

Pillar 3: Measurement has to be rebuilt from the ground up

This is where I'd push back on the grand-statement pieces hardest. Saying "attribution is broken" or "the old measurement model is dead" is a true sentence that does not help anyone allocate budget on Tuesday. Businesses still have to decide what to spend on what, and decisions get made with whatever numbers are available.

The practical answer is a stacked measurement model with four layers, none of which is sufficient on its own.

Layer one is server-side analytics on your own properties. Real visit data, properly stripped of bot traffic — which is now genuinely difficult because legitimate AI agents are visiting on behalf of users and you probably want to count some of them. Log file analysis matters again in a way it hasn't for a decade.

Layer two is branded-search and direct-traffic tracking. If your branded search volume is going up while your overall organic traffic is going down, you're winning the Machine Media game even though your dashboards look like you're losing. If both are going down, you have a real problem. This is the most underrated metric in the industry right now and almost nobody is reporting on it in board packs.

Layer three is citation tracking. Are you being mentioned in AI Overviews, in ChatGPT responses, in Perplexity, in Claude, for the queries that matter to your business? The tools to do this exist, they're imperfect, and most of them are overpriced. Pick one and use it consistently. Trend over time matters more than absolute accuracy.

Layer four is downstream conversion data. Are leads still coming in? Are sales still closing? Is the cost-per-acquisition stable? In a world where the journey is increasingly invisible, the only honest measure of whether your marketing is working is whether the business outcomes you care about are still happening. If they are, the dashboard problem is a dashboard problem. If they aren't, you have a real one.

None of these layers is complete. Together, they give you enough signal to make decisions. The industry has spent the last 18 months pretending one of them — usually citation tracking — is the answer. It isn't. The stack is the answer.

Pillar 4: Your conversion surface is no longer your website

This is the operational implication most businesses haven't internalised. If the agent is increasingly the surface where the customer encounters your business, then the moment of conversion is whatever the agent recommends or transacts. Your job is to be the thing that gets recommended.

Practically:

For ecommerce, your product feed quality, your pricing competitiveness, your fulfilment reliability, and your reviews are now your conversion optimisation programme. The landing page split-test culture of the last decade is being made partially redundant by the fact that fewer customers are seeing the landing page. The split test that matters is whether your product gets added to the Universal Cart instead of a competitor's.

For service businesses, your structured information — pricing, availability, response times, service areas, certifications, reviews — is what the agent has to work with when it's recommending. A plumber with clean structured data showing genuine same-day availability in a specific postcode is going to be recommended over one with a beautifully designed website and no machine-readable hours.

For B2B, the comparison surface is where the work happens. The G2 reviews, the Capterra listings, the Reddit threads where your category is discussed, the industry analyst reports. These are now your primary funnel. Your website's role shifts toward closing — providing the depth and credibility a buyer who's already been recommended you needs to commit.

This isn't a complete abandonment of the website. Far from it. The website still matters for closing, for credibility, for the customer who does click through, for the search queries that still drive direct visits. But its weight in the overall mix has dropped. The businesses winning are the ones that have rebalanced their effort accordingly.

Pillar 5: The legal and contractual layer is now strategic

The Amazon v Perplexity case at the Ninth Circuit isn't just a legal story. It's a strategic one, because the outcome will dictate the terms on which AI agents can interact with your business for the next decade.

If the court rules for Amazon and the CFAA theory holds, businesses will have significant control over which agents can access logged-in areas of their sites. They'll be able to require agents to identify themselves, to negotiate access terms, to charge for it, to block it entirely. This is the world publishers have been hoping for and the world OpenAI and Perplexity have been operating against.

If the court rules for Perplexity, the user remains the authorised party at all times, agents acting on user instruction are functionally indistinguishable from the user, and businesses lose much of their ability to gate agent access without also gating their actual customers.

Either outcome has operational implications you should already be thinking about. Your terms of service, your robots.txt, your authentication flows, your rate limiting, your API access — all of these are now strategic decisions, not technical ones. Most businesses have outsourced these decisions to whoever set up their site three years ago. That's no longer adequate.

You don't need to predict the legal outcome. You need to know what your position will be under either ruling and what changes you'd make on day one of each. That's a 90-minute conversation with whoever runs your tech and whoever runs your legal. Have it.

The strongest counterargument

The strongest version of the opposing view goes something like this: the industry has been declaring the death of search and the rise of the agent for two years now, and the click economy is still mostly intact for most businesses. AI Overviews appear in a meaningful percentage of searches, but the majority of commercial queries still produce blue links and people still click them. Universal Cart and agentic checkout are real but barely used. The Cloudflare bot-traffic numbers are dramatic but they include legitimate agents acting on user instruction, which is a different economic event from a customer not visiting at all. The "Machine Media" framing makes a great keynote but the data showing customer journeys actually replaced by agent journeys at scale is much thinner than the rhetoric suggests.

The cost of being early is roughly six months of effort that compounds. The cost of being late is the business.

There's real force in that argument. The pace of change is slower than the keynotes imply. The transition will be uneven across industries — agentic commerce will eat travel and consumer electronics before it eats local services or B2B SaaS. Many UK businesses can probably operate on roughly the existing model for another two to three years before they're forced to change. The grand-statement pieces overstate the immediacy.

I'd concede that point but not the larger one. The direction of travel is settled even if the speed isn't. The businesses that start the operational work now — the schema cleanup, the structured data, the earned-media investment, the measurement stack rebuild — will be in position when the wave actually arrives. The ones that wait for the change to be visible in their own dashboards will be doing this work under duress, with revenue already declining, with budget already cut. The cost of being early is roughly six months of effort that compounds. The cost of being late is the business.

The honest version of the position is: act as if the transition is happening, but don't burn the existing playbook before it's actually dead. The existing playbook still works. It's just no longer sufficient.

What this actually means for your next 90 days

If you run a small or mid-sized UK business and you've read this far, the operational priorities cluster into three tiers.

Tier one, do in the next 30 days: audit your structured data and fix it. Audit your branded-search trend and start reporting it monthly. Set up basic AI citation tracking — pick one tool, even if it's imperfect. Read your terms of service and decide your position on agent access.

Tier two, do in the next 60 days: identify the three trade publications, two subreddits, and one industry analyst whose mentions of your business would have the highest LLM-citation value. Build a deliberate outreach programme to all six. Review your product feed or service catalogue and make it machine-legible. Have the legal-and-tech conversation about agent access.

Tier three, do in the next 90 days: rebuild your reporting to show the four-layer measurement stack. Have the harder conversation with leadership about what success looks like when the metrics you've reported for a decade are no longer the right ones. Stop budgeting for tactics that depended on assumptions that no longer hold.

None of this is glamorous. None of it is a keynote. It's the unsexy operational work that actually moves a business through a transition.

The grand-statement pieces this week were correct about the diagnosis and incomplete about the response. Machine Media is real. The open web's economic model is genuinely under pressure. The click economy is genuinely thinning. All of that is true.

What's also true is that businesses don't get to operate in the diagnosis. They have to operate on Tuesday morning, with the team they have, the budget they have, and the customers they have. The work is operational, it's specific, and it's mostly things the industry already knows how to do — just weighted differently than it has been for a decade.

The businesses that win the next five years aren't the ones with the cleanest theory of what's happening. They're the ones doing the boring work of making themselves legible to the systems that now mediate their customers. That work is available to anyone. Most won't do it. The ones that do will look, in 2028, like they were lucky. They weren't.