When AI agents replace human attention, advertising doesn’t die — it goes invisible.
Somewhere in a data center right now, a bot is loading a webpage, triggering an ad impression, and costing a brand money. Nobody will ever see that ad. It wasn’t designed for this visitor. The visitor doesn’t have eyes.
This has been happening for years, and the industry treats it as a bug — billions spent on fraud detection, traffic filters, verification layers. But what if the bug is becoming the product? As AI agents start browsing, comparing, and buying on our behalf, the question isn’t whether machines will see ads. It’s whether ads will be designed for them.
There are two stories about advertising and machines, and they sound alike but are nothing alike. The first is about waste — bots faking clicks, draining budgets, getting caught. That story is old and adversarial. The second is about AI agents that make real decisions for real people, and the market forming around influencing those decisions. That story is just beginning. The confusion between the two is where most of the skepticism lives. When people hear “ads for machines,” they picture the fraud. What they should picture is the agent with a credit card.
The old story is easy to tell because it’s already enormous.
Automated traffic accounts for roughly half of all internet activity, with bad bots alone making up about a third. The latest wave of AI crawlers — GPTBot, ClaudeBot, PerplexityBot — is pushing that share higher. Some publishers now get more visits from AI systems than from human readers.
The advertising industry has been fighting this for years. Anti-fraud programs saved U.S. advertisers an estimated $10.8 billion in 2023 — not the cost of fraud, but the savings after mitigation. The actual scale of ads served to non-humans is larger than most mid-size ad markets. It’s an arms race with a clear villain, and both sides keep escalating.
But the machines in that story are parasites — they extract value by pretending to be people. The machines in the new story don’t pretend to be anything. They’re agents, acting on behalf of real users, with real intent, and increasingly, real wallets.
Google is already placing ads inside AI Overviews — the summaries that appear above search results. Perplexity, the AI search engine, launched sponsored questions where the AI generates the answer but a brand pays to surface the topic. And on the payments side, the infrastructure for agents to actually buy things is being built right now. The major card networks are designing checkout flows and payment credentials specifically for machine-initiated transactions — not humans tapping a phone, but agents completing purchases autonomously.
The pipes are already being laid. By the companies that own the pipes.
The moment agents can transact, influencing their decisions becomes advertising in the most literal sense. Not the metaphorical, think-piece sense. The someone-is-going-to-build-a-rate-card-for-it sense.
Here’s the conceptual shift that matters most. Traditional digital advertising optimizes a funnel built for human attention:
Impression → Click → Conversion.
Every stage assumes a person seeing something, feeling something, deciding something. But when an AI agent is doing the browsing, there’s no impression to make, no click to entice, no emotional trigger to pull. The funnel rewrites itself:
Inclusion → Recommendation → Execution.
The bottleneck is no longer whether someone notices you. It’s whether you make the agent’s shortlist — a shift from attention to agency that I trace in Attention Isn’t Enough.
That changes everything about what “marketing” looks like in practice.
Consider a travel agent AI planning a weekend trip for its user. It knows the budget, the dietary restrictions, the preference for boutique hotels. It queries availability, compares reviews, and assembles an itinerary. At no point does it scroll past a banner ad and feel intrigued. But if one hotel chain has structured its data so the agent can parse it effortlessly — and has paid for priority placement in the agent’s recommendation layer — that chain has an edge. Not because the agent was persuaded. Because the agent was fed.
Or imagine a personal finance agent evaluating credit cards on your behalf. It doesn’t care about aspirational lifestyle branding. It cares about interest rates, reward structures, and fee schedules. But if one issuer’s terms are formatted in a way the agent can digest cleanly, and that issuer has paid the platform for “promoted comparison” status, it’s more likely to surface in the final three options the human actually sees. The ad didn’t target the human. It targeted the machine that curates for the human.
Some marketers are already calling this GEO — Generative Engine Optimization. The idea is the same as traditional search optimization, but instead of structuring content so Google ranks it highly, you structure it so AI systems cite and recommend it. It sounds absurd until you remember that “optimize for Google’s algorithm” sounded equally absurd in 1998. The SEO industry took over a decade to mature — years of trial and error, reverse-engineering ranking signals, building tools and best practices from scratch. GEO won’t take nearly that long. The playbook already exists. The strategists, the agencies, the tooling vendors — they all know the shape of this game because they’ve played it before. They just need to swap the target. Within a few years, there will be agencies that do nothing but optimize for agent visibility, and they’ll charge a lot of money for it.
The natural objection is: I’ll just tell my agent to ignore ads.
And maybe you will. If you run your own agent and control its instructions, you can strip out sponsored content entirely. But that assumes you control the agent. The more likely scenario — the one every major platform is building toward — is that the default AI assistant lives inside someone else’s ecosystem. Apple’s, Google’s, Amazon’s. And those companies will not build agents that refuse to monetize.
This is the real strategic battle: not who builds the best agent, but who builds the default agent. The default agent controls the ad surface. And history suggests that defaults win. Most people never changed their browser homepage. Most people won’t reconfigure their AI assistant’s recommendation settings either.
That said, the market only survives if users trust it. An agent that secretly recommends products because a brand paid for placement will lose users the moment that becomes obvious — and it will become obvious. So disclosure becomes table stakes. The surviving model will probably look like “three of these five options are organic, two are promoted” — closer to Amazon’s sponsored listings than to some covert manipulation of agent reasoning. Transparency won’t kill this market. It’ll just make the market legible.
Which means the agents that win will be the ones that are useful enough to keep using despite the ads — exactly like search engines, exactly like social feeds, exactly like every attention surface before them. The pattern isn’t new. The audience is.
But there’s a question underneath all of this that nobody seems to be asking, and it’s the one that should make you uneasy.
If agent traffic becomes the dominant form of web activity, and the ad money follows the agents, what happens to the internet that was built for people?
We’ve seen a version of this before. Search advertising reorganized the entire content economy around Google’s algorithm. We got SEO farms, keyword-stuffed articles, listicles engineered for clicks rather than insight. The web became worse for humans because optimizing for the algorithm and optimizing for the reader diverged — and the money was in the algorithm.
Now run that forward. If the valuable traffic is coming from AI agents, and the ad revenue follows that traffic, content will optimize for machine consumption. Structured data over narrative. Parseable formats over beautiful design. Product specs over storytelling. The websites that thrive will be the ones agents find most useful — which may have nothing to do with what humans find most meaningful.
The human web doesn’t die. It just stops being where the money is.
That sentence is worth sitting with. Not because the human internet vanishes — people will still write, design, and create for other people. But the economic engine that funds the web, that pays for servers and salaries and content production, will increasingly orient toward the audience that transacts at scale. And that audience won’t be reading your homepage. It’ll be parsing your API.
“Agent-first” won’t just add a new surface. It removes the assumption that there’s a person on the other end at all.
We spent two decades building an internet optimized for capturing human attention. We got clickbait, infinite scroll, and dopamine loops — an economy built on one problem: making people look at things.
That problem is about to become irrelevant. The new audience doesn’t get bored, doesn’t get emotional, and doesn’t doom-scroll at 2 AM. It processes, evaluates, and executes. And the market that forms around influencing those executions will be quieter than the attention economy, less visible, less dramatic — but possibly larger.
The ad that nobody sees might be the most important ad ever served. Not because it changes a mind. Because there’s no mind to change. There’s only a decision, and someone paid to shape it.