TLDR: The internet’s foundational media logic is breaking. Here’s what’s happening:
Bots outnumber humans: 51% of web traffic in 2024 was automated, with sophisticated AI bots now mimicking human behavior at scale
The core assumption fails: Digital advertising was built on the idea that impressions equal people. That’s no longer reliably true
AI creates abundance everywhere: Every traditional signal (views, clicks, engagement, dwell time) can now be manufactured at zero marginal cost
Only one thing remains scarce: Verified human presence. The actual person making a decision with stakes, in real time, in front of others
Money is already moving: Marketing budgets are shifting toward live events, owned communities, and verified creators where proof of humanity is built into the structure, not the analytics report
The new threshold: The only question that matters is whether a human originated the action and whether you can prove it. Everything else will be priced to zero
The industry spent 30 years optimizing for attention. The next 30 will be spent proving whose attention is real.
As long as there has been media, the entire business has run on a single question: how do we get in front of people?
The question hasn’t changed. The answer is breaking.
The infrastructure built around it is still operational. The trading desks. The attribution models. The impression-based pricing. The agency compensation structures keyed to media spend. Every assumption inside media operations traces back to attention being scarce and reach being the asset. For decades, brands paid for impressions because impressions were proof that a person had looked at something they otherwise would not have seen.
That logic is breaking quietly, and almost no one in the industry has updated their behavior.
THE LONG PATTERN OF SCARCITY
Technology has always organized itself around scarcity.
Sometimes it creates the scarcity. Sometimes it exploits one that already existed. Either way, the scarce resource is where value accumulates, and the technologies that win are the ones that figure out which scarcity will matter next.
Steam and the factory created scarcity around capital and coordinated labor. Broadcast created scarcity around airwaves and prime time: three networks, a finite number of evening hours, and an entire advertising economy organized around buying access to those windows. Digital created scarcity around attention. Inventory became infinite, distribution became frictionless, and the only resource that could not be manufactured at scale was the willingness of a person to look at something. The platforms that captured and resold that willingness became the largest companies in the world.
In every case, abundance somewhere produced scarcity somewhere else. The pattern held.
AI breaks the pattern by accelerating it past its own limits.
It is producing abundance everywhere at once. Content is infinite. Engagement is generatable. Personas are manufacturable. The signals the industry has historically priced as scarce can now be produced at zero marginal cost: views, clicks, dwell time, and sustained behavioral patterns. There is no remaining bottleneck inside the digital system that AI cannot saturate.
Which means the next scarcity has to come from outside the system.
The thing AI cannot generate is the actual human on the other end of it. The body in the room. The person making a decision in real time, with stakes, in front of other people who will know if they are faking. That presence is the new bottleneck. The only input the technology can’t simulate, scale, or replace.
AI IS THE FIRST TECHNOLOGY THAT MAKES HUMAN PRESENCE MORE SCARCE
For two centuries, technology has made human labor more abundant. AI runs the other direction. The digital substrate it operates inside is being saturated with signals that no human originated, and in a media economy built on signals, the only signal that retains value is the one that proves a human was actually there.
Call it proof of humanity.
The phrase sounds technical because most of the work being done on it has been technical. Bot detection. Identity verification. Cloudflare blocking AI scrapers. The detection-and-evasion arms race that runs in the background of every major platform.
Those tools matter. They are also the perimeter of something much larger.
The real shift is operational. The silent assumption that made the old media business work, that an impression equaled a person, is no longer reliably true. Once that assumption fails, every system stacked on top of it starts to wobble in ways that are difficult to see from the inside.
The numbers have been quietly arriving for those who care to read them. Imperva’s 2025 Bad Bot Report, the firm’s twelfth annual study of automated traffic across its global network, found that for the first time in a decade, automated traffic surpassed human activity on the internet. Fifty-one percent of all web traffic in 2024 was non-human. Bad bots alone, the kind specifically engineered to mimic human behavior for fraudulent purposes, accounted for thirty-seven percent. The remaining human share has been declining for six consecutive years.
The trajectory matters more than the headline. The synthetic share registered in 2024 was crude by the standards of what is coming. The systems generating bot activity in 2026 were trained on human behavior at scale and have become very good at imitation. The signals the industry has historically used to filter for humans are now generatable: cursor movement, dwell time, and scroll velocity.
What sits underneath the modern internet is a layer of activity that is no longer being produced for any human to read or watch. It is being produced because optimization systems demand it, because other optimization systems consume it, and because the loops between them have become an economy of their own.
That economy can sustain itself for a long time before anyone with a P&L notices.
THE DASHBOARD LIES QUIETLY
Inside a marketing operation, this does not arrive as a crisis. It arrives as a slow disconnect.
Numbers that pencil cleanly on the dashboard but produce nothing in the field. Conversion lifts that show up in the model but not in the warehouse. Engagement rates climbing while the brand teams admit, in rooms with the door closed, that they cannot tell whether anything is actually moving. The instruments are calibrated. They’re monitoring the wrong signals.
Optimization engines do not care whether the data is human. They optimize against whatever signal they receive. If conversion behavior in the dataset is half synthetic, the model will confidently learn to drive more of it. The lookalike audiences will model audiences that do not exist. A creative test validated by bot traffic will greenlight a bad ad. The more sophisticated the operation, the deeper the contamination, because automation amplifies whatever it is fed.
The brands with the most reliable signal turn out to be the ones with the simplest measurement. That is an inversion the industry was not built for.
I’ve sat with senior performance marketers who describe the gap as a feeling first. The numbers say one thing, the gut says another, and the gut is right more often than they will admit out loud. Eventually, someone in the room asks the quiet question: whether the underlying data is even describing real people.
That question, asked once, changes how a marketing operation thinks about every input it receives.
PARTICIPATION AS PROOF
This is where participation stops being a soft word for engagement and starts being the only signal worth measuring.
Participation is proof of humanity expressed through action. A click can be produced by a script. A view can be paid for. A follower can be rented. The deeper forms of participation are harder to fake. Entering an experience, contributing to it, responding in real time to what other participants do, being shaped by what unfolds: that pattern degrades quickly when the underlying agent is not human, because the agent has no body, no history, no stake in the outcome, no social cost for getting it wrong.
Bot networks can simulate shallow participation. The simulation breaks down fast as the participation deepens. Real-time decisions in a shared environment, in front of other people who recognize coherence and incoherence, generate a signal that machines have not yet learned to forge well.
Anyone who has trained in jiujitsu has felt the cleaner version of this. You can roll with someone for thirty seconds and know whether they’ve had actual time on the mat. The pressure is wrong. The timing is wrong. The body lies to itself in ways that other bodies have already learned to read. A pretender in jiujitsu and a practitioner of jiujitsu register as different things almost immediately, and not because of any single signal you could name. It is the absence of the lived weight.
Real participation has weight. The simulator does not. The closer you stand to it, the easier it is to read the difference.
THE ORIGINATOR TEST
The objection I get is always some version of the same one. AI is going to participate too. Agents will comment, attend, coordinate, and even create. How is that different from a bot?
The answer is the originator.
A person orchestrating an agent is participating. The agent is a tool: the way a microphone is a tool, the way a phone is a tool, the way any extension is a tool. The intent originates with the human. The chain of action terminates in a body somewhere with social context and consequence. That is participation, even when most of the execution is automated.
A bot operating without any human originator is a different thing entirely. Nothing is being expressed. No one is participating. The activity is the byproduct of an optimization process that has no relationship to the people the activity is being measured against.
The threshold is not whether AI is involved but whether a human is.
Everything downstream of that distinction reorganizes once you accept it. Verification systems become provenance systems. Engagement metrics get audited for their chain of custody. The question every marketer learns to ask, on every signal, becomes some version of: did a person originate this, at what level of involvement, and through what structure?
WHERE THE MONEY IS ALREADY MOVING
The brands that figure this out a few years before the industry does will be the ones that move budget while it still costs nothing to move. They will fund the environments where the human signal is structural rather than promised. Direct community. Owned audience. Verified creators with real followings. Live events. Membership. The kinds of placements where proof is built into the architecture and not appended to the report.
The early evidence is already showing up in the spend data. Event Marketer’s EventTrack 2025 found that roughly three-quarters of Fortune 1000 marketers were increasing experiential investment heading into 2026. The trade press has been calling this an experiential rebound, which might be underplaying what is really happening. It is an early proof-of-human reallocation, even if the marketers doing it cannot yet articulate why their gut is pulling them there.
The most visible evidence is at festivals.
Coachella activations got materially better this year. Not incrementally better. Visibly better, in a way that anyone walking the grounds could feel. Adweek, reporting from on the ground at the festival, observed that the standout activations “felt less like marketing moments and more like experiences built around festival-goers.” The brands that showed up showed up with effort. Pinterest, a platform whose entire business is capturing attention through phones, built Coachella’s first-ever phone-free activation, asking guests to lock their devices away on entry in exchange for a more present, analog experience. It is the kind of choice that looks like marketing self-sabotage against a CPM model and looks rational against a model that prices verified human presence at a premium.
Which raises the question worth asking out loud. The Super Bowl has been the cathedral of impression-based advertising for forty years, the purest expression of the old logic: maximum simultaneous attention, premium pricing, the ads themselves becoming the cultural ritual. The model worked because the audience was real, the moment was singular, and proof of human was incidentally guaranteed by the format.
The question is whether the budget that has flowed there for two generations starts redirecting toward environments like Coachella. Live festivals. Cultural moments where participation is structural. Places where the brand is not interrupting the experience but is part of it, and where the proof of humanity is not something the analytics team has to argue for after the fact.
Live sports work this way. Newsletters with verified subscribers work this way. Creator partnerships with audiences that talk back work this way. Direct community works this way. None of these are new categories. What is new is the pricing logic.
PROOF OF HUMAN AS INVENTORY
Once you accept that proof of humanity is the new underlying scarcity, the consequence is that it becomes an inventory class.
The industry already organizes its buying around inventory classes. Network television was its own class. Programmatic display was its own class. Connected TV is its own class. Each one carries its own pricing logic, its own buyers, its own measurement standards, its own seat at the upfronts. They got organized as classes because they shared an underlying signal type and could be priced against each other.
Proof of humanity inventory will get organized the same way. The shared signal type is verified as a human originator at a usable level of involvement. The placements that clear that threshold will trade against each other; the placements that do not will trade against a different and rapidly cheapening pool. Live, opt-in, community-anchored, creator-led, membership-bounded environments belong in the first pool. Most programmatic display, most synthetic-amplifiable social, most reach-extended digital belongs in the second.
This is the part that the industry has not built the language for yet.
When it does, the consequences will be structural. Pricing changes, because proof-of-humanity inventory commands a premium that the old impression-based math cannot generate. Planning changes, because plans get built around human density rather than reach curves. Compensation changes, because billing against verified-human signal looks different from billing against impression volume. Measurement changes, because every metric becomes a provenance metric, weighted by the chain of custody on the underlying signal.
The naming has not happened yet. It is going to.
THE NEW LINE IN THE SYSTEM
This is not a moral argument about AI. The synthetic internet is not going away. AI will continue to extend what people can do, including the work of participating, and that is fine.
The threshold is the originator. Above it: proof. Below it: noise. Everything that happens inside marketing for the next decade will eventually have to clear that line or be priced as if it does not.
The industry spent thirty digital years optimizing for attention.
The next thirty will be spent proving whose attention is real.
Everything that cannot prove a human will be priced to zero. Everything that can will be the only thing worth buying.