The open web’s advertising model was built on the movement of people through content.
That movement created the raw material of performance media. Advertisers could read intent from behavior, and they could reach people because the same behavior placed them inside monetizable environments. Search, content, commerce and comparison all worked because the user’s path was visible enough to measure and populated enough to sell.
AI weakens that arrangement at the source. As retrieval replaces browsing, more intent gets processed before it becomes open-web activity. The user still has questions and preferences, but fewer of those signals spill into the channels that advertisers have historically used to understand demand. The market is left with less evidence of what people want. One study found that users clicked a traditional result on 8% of Google searches with an AI summary, versus 15% when no summary appeared.
The exposure problem matters more. Advertising cannot perform against intent alone. It needs a reachable person. If AI systems absorb more discovery, filtering and decision support, then many of the surfaces that once carried human attention become less reliable as media environments. The web may still contain the same content, but less of the decision-making path passes through places where an ad can meet the user.
That is the uncomfortable shift for performance marketers. AI does not have to end the open web to damage its economics. It only has to reduce the amount of human navigation.
The disappearing middle
A consumer journey can preserve the same demand while producing fewer opportunities to reach it. Research that once required moving between search results and multiple sites can increasingly be resolved inside one interface. The consumer may arrive at a merchant with a narrower set of choices and far less observable activity behind the decision.
For marketers, each removed step takes signals and exposures with it. The journey once revealed intent as it unfolded and created places where brands could influence consideration before the transaction. When an AI system handles more of that work upstream, the eventual purchase can look much the same even though the media opportunity around it has contracted.
Digital advertising optimized around the trail people left behind. Search, retail and affiliate channels did not always work cleanly, but they were connected by a shared assumption: consumer activity would remain broadly observable.
Agentic systems challenge that assumption. Their value comes from reducing the visible work between need and outcome. They synthesize information and route decisions through interfaces that may not expose the same behavioral residue to the rest of the market. From the consumer’s perspective, that can feel like progress. From the advertising market’s perspective, it removes part of the substrate.
Friction was inventory
This will force a difficult conversation about media quality. A large amount of digital inventory has depended less on durable attention than on digital friction. This inventory has included sources like open-web comparison pages, “10 best” roundups and SEO doorway content. These are all pages people pass through, not places they go. AI will make that distinction harder to avoid. Inventory built around low-value navigation will be more vulnerable than inventory tied to identity.
The distinction will become pointed. Environments can sustain attention even as AI strips away more of the connective tissue between destinations. Inventories that relied on an extra page to compare or summarize will face heavier pressure. Direct audience relationships should gain relative value as those incidental visits disappear. Merchants hold the holy grail of shopper intent, and sharing that data can benefit their monetization along with providing the consumer with stronger, meaningful suggestions.
Some budget will move into AI interfaces themselves. That market will develop quickly, but it will carry real constraints. Consumers use AI systems because they expect efficiency and trust. Advertising that makes those systems feel less useful will face resistance. The ad product inside an answer layer cannot simply recreate the formats of search, display or sponsored content without damaging the experience.
The largest walled gardens will also benefit. They have identity, scale and habitual use. They can continue to aggregate human attention even as parts of the open web lose it. But deeper dependence on closed platforms concentrates leverage in fewer hands and gives marketers less room to exercise judgment and oversight.
Durable attention sits closer to the transaction
Transactional utilities like loyalty ecosystems, order tracking and booking flows stand as an interesting alternative. They are attached to things people are already doing. The advertising opportunity is created by proximity to a real commercial or service moment.
Those moments are harder for an agent to erase because they come from an ongoing commercial relationship. As discovery contracts, these environments become more strategically important. They bring advertising closer to a person who is already engaged in a real activity, often with a recent transaction providing context. As frontier AI systems increasingly support MCP connections, platforms can make permissioned transaction and loyalty data available inside an agent, carrying that customer context into interfaces they do not control. The total surface may be smaller, but the remaining impressions can carry a clearer relationship to the consumer’s immediate needs.
AI will make the internet more efficient. Efficiency is usually good for consumers and dangerous for weak intermediaries. The practical response is to judge inventory by the human activity underneath it. Attention anchored in an active customer relationship should prove more durable than attention created by navigation alone. AI will leave advertisers with different places to appear across some parts of the web. The value of the surfaces that remain will depend increasingly on why the person is there in the first place.
