Attention Fades. Money Moves.

Attention Was a Subsidy

For three decades, much of the consumer internet operated under a familiar arrangement: users paid little or nothing directly, while advertising funded a large share of the service. Major search and social platforms turned user attention into advertising revenue, often using behavioural signals to target and measure ads.

Advertisers did not pay for digital content out of generosity; they paid for access to human visual focus. The pageview, the banner ad, and the sponsored link worked as business models because a person was looking at a screen, seeing promotions, and leaving behavioural footprints.

When AI assistants step between users and websites, that cross-subsidy can weaken. An automated tool reading product details or summarising news does not view banner ads, click on sponsored links, or generate mouse-hover trails. As software handles more routine tasks, direct human page visits may become less frequent, and the ad engine funding free infrastructure could lose efficiency.

It helps to distinguish between four levels of AI assistance:

• Answering: The assistant replies from its training data without visiting websites.
• Retrieving: It fetches live information from webpages or APIs.
• Comparing: It gathers options from multiple sources and ranks them.
• Acting: It completes transactions on behalf of the user.

The economic impact grows as assistants move from answering questions to taking actions. Not every AI interaction removes a website from the process: assistants will still depend on authoritative sources, structured APIs, product feeds, booking systems, and payment infrastructure. But the trend towards delegation is what matters.

Paying For Fresh Information

If advertising generates less revenue from each user interaction, publishers and service providers may look for ways to charge the systems that still depend on their information. One possible response is to establish digital tollbooths: charging AI developers when automated tools request their data.

But static information is difficult to keep charging for. If an AI reads a page of historic dates or basic product dimensions, that information can be stored and reused without continually checking back.

• Information that stays static: Basic product measurements, historical records, or standard documentation.
• Information that goes stale quickly: Live seat availability, dynamic prices, traffic conditions, or current stock levels.

The faster information goes stale, the more valuable reliable access to the current source becomes. A cached flight price is not enough; the assistant must know whether the fare is still available, under what conditions, and whether the booking can actually be completed.

In practice, this may look less like charging for every webpage request and more like licensing structured data, selling API access, sharing revenue, or taking a commission when an assistant generates a transaction.

Websites may monetise transaction access more than content access.

Paying For Finished Results

As AI tools move from retrieving information to executing tasks, pricing may shift from access and raw computing usage towards finished results. A user may not want to pay an agent for ten minutes of processing; they may prefer to pay only if it successfully books the trip, finds a qualified lead, or resolves the support issue.

If a system needs repeated attempts or gets stuck in a loop, the buyer pays for processing without necessarily receiving usable work. A result-based model shifts attention to whether the task clears a defined completion test and stays within a spending limit.

Examples could include:

• A qualified sales lead that meets predefined criteria.
• A booked trip that satisfies specified constraints.
• A customer-support issue resolved to the user’s satisfaction.

Outcome-based pricing works best when success can be measured clearly. It becomes harder when the user is buying judgement, creativity, exploration, or advice.

Sponsoring Consideration, Not Attention

Commercial influence is unlikely to disappear if visual page visits decline, but its mechanics are shifting. When users rely on an AI assistant as an advisor, inserting hidden advertising directly into its recommendations risks eroding user trust.

Instead of paying to catch a human eye, advertising may shift from buying attention to buying consideration.

In this setup, the recommendation process separates into two distinct steps:

• The primary recommendation: The AI searches available sources to identify relevant options based on user preferences such as price, specifications, or quality.
• The commercial offer: Separately, the system checks whether a relevant brand offers a verified option—such as a discounted upgrade or matching promotion—and presents it clearly as sponsored.

For example, an assistant could first identify hotels that meet a user’s budget, location, and quality requirements, then separately show a sponsored room upgrade or promotion from one of those suitable hotels.

“Sponsored consideration” means a company may pay to have a qualifying offer considered when the assistant compares options. The commercial relationship would still need to be clear, and users would need to understand whether sponsorship affected inclusion, ranking, price, promotion, or the final recommendation.

Sponsorship should influence which relevant offers are shown, not secretly override what is best for the user. Commercial sponsorship could remain viable, but move from capturing visual attention to presenting relevant options at the decision moment. Trust depends on clear disclosure and a genuine match between the offer and the user’s needs.

Who Controls The Last Mile

As AI capabilities expand, leverage may shift towards entities that manage the final step between the user and the real world: the device, user identity, payment permissions, and the primary customer relationship.

Here, “last mile” does not mean physical delivery. It means the step between deciding what to do and actually doing it: authenticating the user, confirming permission, making payment, placing the order, and handling the result.

• Identity: Verifying who the user is.
• Permissions: Determining which tools can access data or execute actions.
• Payment: Handling secure financial approvals.
• Relationship: Operating as the primary interface the user interacts with daily.

An assistant may find the best flight, but the company controlling the user’s identity, wallet, device, and approval flow may control the booking—and potentially the economics around it. That company may also determine which assistants can act, what data they can access, which payment method they use, and how easily users can switch to competing services.

Control also brings accountability. Once an assistant moves from advice to action, users need to know who is responsible when something goes wrong, who handles refunds or disputes, who protects payment details, and whether an action can be explained, cancelled, or reversed. Responsibility may be divided among the assistant provider, the data source, the merchant, the payment provider, and the platform controlling the transaction. The entities that can credibly manage that risk and redress may capture part of the value too.

Control of the last mile is not automatically a winning position. Hardware costs, consumer preferences, and open web standards may continue to limit how much power device makers or operating-system providers can exercise. But controlling the user’s primary interface remains a strategically powerful position.

What Becomes Valuable

A decline in attention-funded digital services would not mean internet business models disappear. It would mean they adapt. Some routine browsing and simple comparisons may move into assistants, while authoritative sources, live services, trustworthy recommendations, and transaction-ready platforms become more valuable.

The emerging models differ in what they monetise, who pays, and who bears the risk.

As AI tools decouple user goals from visual page visits, value capture may move away from simply hosting visual screens.

Economic value does not disappear. It moves towards whatever remains scarce, trusted, and difficult to bypass.