Google’s latest financial results caught my attention. Search advertising revenue is growing, but the websites that supply much of the information people discover online face a very different reality.
For years, we have built the internet around clicks. Publishers create content; search engines index it; people click through; and those visits support advertising, subscriptions, and sales. Artificial intelligence is changing that relationship. An assistant can consume information across many websites and deliver an answer without sending any of those sites a visitor.
So I started wondering: why are we still financing information through human visits when machines increasingly consume it on our behalf? Why not build a system that lets AI pay for licensed content when it reads or uses it?
Google is making more money. So why is the web losing traffic?
Alphabet reported Google Search advertising revenue of approximately $63.35 billion for the third quarter of 2026, an increase of around 12% year over year. Paid clicks grew, too. Google itself is not collapsing, and it would be misleading to suggest that pay-per-click advertising is already dead.
But Google Network advertising impressions fell 14%, while declining AdSense revenue partly offset growth elsewhere in the network. Meanwhile, the Reuters Institute’s 2026 Digital News Report describes Chartbeat data showing an approximately 33% global decline in organic Google search referrals to more than 2,500 publishers between November 2024 and November 2025. In the United States, the decline was around 38%.
These different metrics should not be confused. Not all publisher traffic losses can be attributed to AI, and Google’s own advertising revenue is not a measure of independent publishing health. But together, the figures point toward a structural problem: Google can keep growing while publishers receive less traffic from the information ecosystem they help create.
Search engines are becoming answer engines
Today I can ask an AI assistant to compare sources, analyze technical documentation, summarize the news, and research a complicated topic. It often includes source links. Sometimes I open them, but frequently I already have what I need.
I enjoy that convenience. I do not want to go back to opening twenty tabs to answer every question. Yet from the publisher’s perspective, the economic exchange has changed. An AI retrieves the information, the user receives its value, and the original website may receive no traffic, ad impression, or subscription opportunity.
A 2025 Pew Research Center study found that users clicked traditional Google search results in 8% of visits where an AI summary appeared, compared with 15% without an AI summary. Just 1% clicked a link in the AI summary itself. A citation is valuable for transparency, but it does not necessarily pay to produce the source.
If AI consumes the content, shouldn’t the creator be able to charge for the licensed consumption?
My proposal: an open pay-per-use AI web
I believe we need a standardized economic layer for content access and licensing, independent of any one search engine, AI provider, hosting company, or payment platform.
Pay Per Crawl would let publishers charge verified AI crawlers when they retrieve certain resources. Pay Per Use would compensate creators for specific agreed downstream uses, such as including licensed material in research reports, summaries, or AI-generated answers. Premium retrieval would let an assistant buy access to an individual article, dataset, or report without requiring its user to subscribe to an entire website.
These models should be clearly distinguished. A system might crawl an article and never use it, or retrieve it once and use the licensed material in hundreds of answers. Training and fine-tuning a model would require separate terms from one-off retrieval. Copyright law, including distinctions between facts, excerpts, reproduction, and training, must be respected.
There must also be a free option. If I want my writing available to everyone and every AI agent, I should be able to declare that. Open-source documentation, public-interest resources, education, and personal writing should not be forced into universal paywalls.
The objective is publisher choice and an interoperable licensing system, not charging for every fact or link on the internet.
Give personal AI assistants their own research budgets
Imagine giving your AI assistant a monthly research budget of €10. You allow it to purchase resources costing less than €0.05 automatically and require confirmation for anything higher.
When you ask for research, the assistant first reviews free resources. It then discovers a specialist publication charging €0.08 for one report and a dataset available for €0.15. These numbers are illustrative. Based on your preferences and budget, the assistant requests approval or pays through a supported protocol, reads the licensed material, and completes the research.
Instead of buying subscriptions from dozens of publishers, users could pay for the exact information they need. Specialists and independent publications could earn money from interactions that would otherwise never send them a visitor.
Microtransactions would need prepaid balances, aggregation, or efficient payment infrastructure, since ordinary credit-card fees can exceed a very small content price. User privacy would be essential: purchasing a source must not reveal someone’s private AI conversation to the publisher.
And why not rethink AI advertising?
If AI increasingly mediates how people discover products and services, advertising will follow. AI platforms already experiment with sponsored placements. A broader exchange could allow advertisers to fund clearly labeled exposure while publishers earn from the licensed information used by AI providers.
But this separation is critical: advertisers should be allowed to buy visibility, not the assistant’s conclusions. Paid placements must remain clearly distinguishable from independent answers. Users should know when money influences a recommendation’s presentation.
Some companies are already working on this
What surprised me was discovering how many parts of the proposal are already under development.
Cloudflare launched Pay Per Crawl in 2025 and expanded its approach with Pay Per Use in 2026. Publishers can define terms and AI buyers can offer compensation for particular uses. The difficult question of verifying downstream AI usage remains important.
Really Simple Licensing (RSL) offers machine-readable content terms, including free access, attribution, pay-per-crawl, subscriptions, and pay-per-inference. A shared standard helps publishers avoid maintaining separate rules for each AI provider.
TollBit is developing tools for licensed AI access, publisher analytics, and content monetization.
x402 is developing internet-native payments through HTTP 402 Payment Required. It could allow authorized AI agents to encounter a paid resource, obtain payment authorization, and continue accessing content.
Google has also reportedly experimented with publisher payments based on contributions to AI-generated answers. The company behind the old search traffic exchange is itself exploring ways to compensate sources when an answer does not require a click.
Those are encouraging signs. But they are pieces of the larger system, not yet a universal, independent economic layer for the open web.
The hardest problem is proving what AI actually used
Measuring a direct HTTP request is comparatively simple. Measuring downstream use is not. An AI could cache an article, generate an answer from multiple sources, or rely on knowledge already inside its model. A source citation is not necessarily a reliable billing event.
A trustworthy system would need agreed licensing definitions, usage receipts, auditable accounting, and privacy-preserving reporting. Otherwise publishers would depend entirely on AI companies telling them how many times their content was used and what it was worth.
Disabling text selection or right-clicking would not solve content protection. Authorization must happen before protected material is delivered. And once readable content is delivered, contracts, audits, and enforcement still matter.
We need open protocols, not another gatekeeper
I do not want publishers to move their hosting, DNS, content, or intellectual property to one company simply to participate. Websites should be able to declare permissions, use their preferred infrastructure, select payment providers, and retain control over access.
Local AI assistants, open-source developers, smaller AI companies, and nonprofits should be able to participate alongside the biggest model providers. This connects with my broader vision of decentralized AI: intelligence should be able to work across personal computers, specialist networks, and cloud infrastructure without making us dependent on a handful of platforms.
We should not replace a search monopoly with a licensing monopoly.
From the attention economy to the information economy
For decades the internet monetized attention through visitors, impressions, clicks, and conversions. AI is increasingly separating the value of information from the website visit that traditionally financed it.
My proposal is an open AI content exchange: publishers choose whether content is free, licensed, private, or premium; agents discover and respect those terms; users set content budgets; advertisers pay for transparent placements; and creators receive meaningful compensation when licensed use occurs.
The web should not have to choose between giving everything away to AI and blocking AI entirely. There should be a third option.
If machines become the primary readers of the web, humans still need a reason to write it.
