Insight / signal
Your website is becoming evidence for machines, not a destination for humans
For years the job of a website was obvious.
For years the job of a website was obvious.
Get found. Get clicked. Explain the offer. Capture the lead.
That model is not dead. It is starting to look incomplete.
The buyer’s first commercial conversation is moving upstream. It might not happen on your homepage. It might not happen on Google in the old sense either. It might happen inside ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, an internal procurement assistant, or some half-baked agent a vendor bolted into a CRM with too much confidence.
Messy, yes. But the direction is clear enough. More of the buying journey is happening before the click.
A Marketing School episode I captured this week put a number on it: roughly 68% of Google searches now end without a click. Other sources define the behaviour differently and report different figures, so treat the exact number as directional. They all point the same way. Buyers ask a question, get a summary, compare options, and often leave without visiting anyone’s source.
The lazy version of this story is “SEO is dead.”
It is not.
The useful version is stranger. SEO is being folded into a bigger job. Your website now has to persuade humans and feed machines at the same time.
That changes what a good marketing asset looks like.
A normal business website is built like a sales brochure. Big claim at the top. Vague benefit. A few service pages. Case studies that are either too thin to prove anything or too polished to trust. A blog full of generic thought leadership. A contact form. Maybe a PDF nobody reads.
That was already weak for humans. For an AI system it is worse.
An answer engine is not impressed by your hero section. It is looking for clean facts, repeated signals, structured information, recent updates, reviews, citations, comparison points, dates, names, and enough public evidence to decide whether you deserve to be mentioned at all.
So the marketing question changes. It is no longer only “How do we rank?” It is also “When an AI system explains this market to a buyer, are we part of the answer?”
That second question is more uncomfortable, which is usually a sign it is worth asking.
AirOps’ updated guidance on ranking in AI Overviews is blunt about the mechanics. Clear answers matter. Structure matters. Freshness matters. Real experience and proof matter. Most cited pages still rank well organically, but ranking alone is not enough. You have to be easy to parse and safe to cite.
Superlines’ AI search data points the same way. Sessions often end without a website visit. Citations and mentions are becoming authority signals. Platform behaviour varies wildly. One model cites you. Another ignores you. A third recommends a competitor with worse work but cleaner public evidence.
That last part matters. AI visibility is not a single leaderboard. It is a messy map of what different systems think the web says about you. If your public evidence is thin, stale or scattered, you are handing the machines almost nothing to work with.
This is not only a search problem. It is becoming a commerce problem too.
Visa and OpenAI announced a partnership to bring Visa’s payment network into OpenAI products, with agent-initiated payments, tokenised credentials, spending limits and fraud monitoring. AP covered the same move as Visa plugging its payments into ChatGPT so agents can shop and pay for users.
Do not overstate it. Nobody sensible should pretend the economy becomes autonomous checkout next Tuesday. The trust gap is still huge. commercetools’ Spring 2026 agentic commerce update is clear that people are far more comfortable using AI to compare prices and products than to place the order.
That is the point. AI is becoming the adviser before it becomes the buyer.
It researches. It compares. It filters. It narrows the shortlist. By the time a human lands on your site, the decision may already be half made somewhere else.
And when agents do start acting more directly, they will need what humans need, just more literally: trusted information, current prices, clear policies, machine-readable data, permission boundaries, and proof that the transaction is legitimate.
That is why OpenAI acquiring Ona belongs in the same story, even though it sits in coding rather than marketing.
OpenAI did not buy Ona because Codex needed another shiny interface. It bought secure, persistent, customer-controlled cloud environments so agents can work for hours or days inside governed systems. Ona’s CEO said the quiet bit plainly: agents need more than intelligence, they need a trusted workspace.
Exactly. And agents also need trusted source material.
A trusted workspace is where the agent acts. A trusted evidence layer is what the agent reasons from. Most businesses have not built that layer. They have built pages.
There is a difference.
A page says, “We help ambitious businesses grow.” An evidence layer answers the real questions: who you help, what you solve, what the offer includes, what it does not include, what proof exists, who said it, when it was last updated, what the process looks like, what buyers should compare, and how a human or an agent takes the next step.
That sounds less glamorous than a new brand campaign. It is a lot more useful.
This is where most AI marketing advice goes wrong. It still treats AI as a content multiplier. More posts. More variants. More newsletters. More synthetic noise sprayed across the internet in the hope that volume creates relevance.
It will not. If anything, the opposite is coming.
As answer engines and agents get more important, vague content gets easier to ignore. Thin service pages get easier to skip. Samey articles get eaten by better-structured, better-cited, more current sources.
The winning move is not to produce more. It is to make the business easier to understand, verify and recommend.
In practice that work looks boring. Rewrite service pages around the questions buyers actually ask. Add honest comparison pages without being slimy. Publish specific proof instead of invented case-study theatre. Keep dates and claims current. Add schema where it genuinely helps. Build review and testimonial loops. Answer pricing, process and fit properly. Use founder-led content to create real experience signals. Make your best thinking easy to quote. Then check what the AI tools currently say about you and your competitors, and track citations, mentions and absences, not just rankings.
This is not a replacement for brand. It is brand made legible.
And this is why the post-agency model matters.
A traditional agency sells campaigns. A content shop sells output. An SEO supplier sells rankings. All useful sometimes. But the bigger need now is an operating layer that keeps the commercial evidence clean.
Research. Positioning. Website structure. Source packs. Proof libraries. Reviews. Schema. Content updates. Distribution. AI visibility checks. Sales follow-up. Measurement. A learning loop that feeds the next cycle.
That is not a one-off campaign. It is a system. It is also what we build at Cleo, because a campaign expires and a system compounds.
Here is the part business owners need to grasp.
Your website is no longer just a destination for humans after they click. It is becoming evidence for machines before humans decide what to click.
If that sounds bleak, I do not think it is. It is a useful forcing function. It rewards clarity. It rewards proof. It rewards being specific. It punishes vague marketing bollocks, which frankly had it coming.
The first move is simple. Before publishing another AI-written blog post, ask the tools what they already think.
Ask ChatGPT, Gemini, Perplexity and Google AI Mode the questions your buyers would ask. Who are the best companies for this problem? What should I look for when hiring one? How do you and your main competitor compare? What does your company actually do? What are the risks with this type of service?
Then read the answers like a buyer would. Are you there? Are you described correctly? Are you missing? Is a competitor framed better? Are old claims being repeated? Are you being cited, or is your content invisible?
That audit will probably be annoying. Good. That is the work.
The next marketing advantage will not come from shouting louder into the feed. It will come from making your business easier for humans and machines to trust.
Not more content. Better evidence.
Jason Sibley is the founder of Cleo, a post-agency marketing and AI company. JasonVsTheNoise is where he writes about what is actually happening with AI, marketing, and how businesses should be thinking about both.