For twenty-five years the open web ran on a visible loop: publishers produced content, search engines sent traffic, advertisers paid for that traffic, and publishers reinvested. AI-generated answers are replacing that loop with a new one — content is ingested, synthesized, and delivered without a click — and a governance debate has opened about who controls the gate. The debate is necessary. But it is incomplete. Transparency about how answers are assembled matters less than the economic rail that determines whether the sources behind those answers get paid.

  • AI chatbots still account for under 1% of publisher page-view referrals, according to Chartbeat (March 2026), but the direction is structural: Gartner forecast a 25% drop in traditional search volume by 2026, and zero-click searches hit 69% when AI Overviews appeared.
  • Fifteen domains capture roughly 68% of all AI citations across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, per the 5W Citation Source Index (May 2026) — a concentration steeper than Google's PageRank ever produced.
  • The open-web search economy paid publishers through traffic. The agentic-web citation economy uses their content but has no native mechanism to pay them for it.
  • Governance frameworks — transparency, epistemic diversity, multistakeholder oversight — are the right aspirations. But without a payment rail inside the answer, they remain aspirations.

What AI information gatekeeping is

AI information gatekeeping is the process by which a small number of language models decide which sources are surfaced, weighted, and cited when assembling an answer — and which are set aside without explanation. In the open-web era, the search engine was a gateway: it ranked pages, but the user chose which to open. In the agentic-web era, the model is a mediator: it reads the sources, weighs them, and delivers a single synthesized response. The step that disappears is the one where the user consulted the sources directly.

The traffic economy and the citation economy, side by side

In the open-web search economy, the unit of value was the click. A publisher earned revenue because a user landed on a page, saw an ad, and the advertiser paid for that impression. The entire chain — content creation, indexing, ranking, traffic, monetization — was visible and measurable. In the agentic-web citation economy, the unit of value is the citation. A publisher's content still grounds the answer, but no click follows, no ad impression fires, and no revenue flows back. Cloudflare data from mid-2026 quantifies the asymmetry: Anthropic's crawlers read 11,122 publisher pages for every single referral they return; OpenAI's ratio sits at 857 to 1; Google's traditional search exchange is roughly 5 to 1.

The mechanic has changed, but the underlying need has not. Brands still need to be visible where decisions are made. Publishers still need to be compensated for being the source. What is missing is the economic layer that connects the two inside the new surface.

What CircleID reported, and what the data confirms

Simone Catania's July 2026 essay on CircleID frames AI-mediated discovery as an internet governance challenge. The core argument: a handful of models now sit between users and the web, a handful of domains sit behind those models, and the combination concentrates information access in ways the open web was designed to prevent. The piece cites Andrew Peterson's concept of knowledge collapse — the risk that language models, by generating toward the center of their training distribution, push minority and peripheral knowledge out of view — and a 2025 empirical study testing 27 language models across 155 topics that found almost all produced less epistemically diverse answers than a basic web search.

The numbers check out against independent sources. The 5W Citation Source Index (May 2026), synthesizing over 680 million citations, confirms that the top 15 domains absorb 68% of all AI citation share and that Reddit alone accounts for roughly 40%. Chartbeat data (March 2026) confirms AI chatbots deliver under 1% of publisher referral traffic. The Reuters Institute Digital News Report (June 2026) found that only 4% of users who encounter news through AI chatbots click through to the original source. The concentration is real and measurable.

Why governance without economics is an incomplete answer

The strongest case for transparency and multistakeholder oversight

The CircleID essay makes its strongest point when it argues that information access should remain anchored in open, accountable systems. The domain name system — governed through bottom-up, multistakeholder process for decades — is offered as a model. Defenders of this framing, including the Internet Society's Global Internet Report and scholars like Konstantinos Komaitis, argue that AI reverses the internet's original logic by recentralizing agency into a few firms that control compute, data, and distribution. The solution, in this view, is structural: transparency about which sources are used, epistemic diversity as a public value, and user representation in standard-setting. This is a credible position. It is also, on its own, insufficient.

Why transparency without a payment rail does not protect publishers

The limitation of the governance-first framing is that it treats the problem as one of oversight when it is equally one of economics. Knowing which sources an AI engine cited does not pay the publisher whose content grounded the answer. Epistemic diversity as a principle does not fund the newsroom that produced the reporting. The open-web search economy worked — imperfectly — because traffic created a payment rail: publishers got paid when users arrived. The agentic web has severed that rail without replacing it. AI licensing deals have emerged — OpenAI has roughly 20 verified publisher partnerships as of mid-2026, anchored by a reported $250 million News Corp deal — but these are bilateral, opaque, and available only to the largest outlets. The long tail of authoritative publishers that AI engines cite daily receives nothing.

Meanwhile, AI advertising is arriving, but it sits beside the answer rather than inside it. eMarketer estimates that over 80% of US AI advertising spend in 2026 runs adjacent to AI-generated responses — the old search auction renting a new wall. OpenAI launched its self-serve Ads Manager with CPC and CPM bidding in May 2026 and reportedly hit $100 million in annualized revenue within weeks. Google is integrating ads into AI Mode. But none of these formats include the publisher in the value chain. The ad money flows to the platform, not to the source whose content powered the answer.

What this means for brands and for publishers

For CMOs, media buyers, and agencies: visibility now means citation, not ranking

The overlap between ranking first on Google and being cited in an AI answer collapsed from roughly 70% in early 2024 to under 20% by April 2026, according to analysis cited in the 5W report. Brands built for ten blue links are not automatically visible in AI-generated answers. CMOs should ringfence a 2026 test budget for AI citation visibility as a discipline separate from SEO and SEA. Media buyers should audit how much of the decision funnel already passes through AI-mediated surfaces. Agencies should build AI visibility as a practice line — not a sub-task of SEO — before clients ask why a competitor is being cited and they are not.

For publishers: price the future on citation revenue, not traffic recovery

Search referrals fell roughly 60% for the smallest publishers between 2024 and 2025, against approximately 22% for the largest, according to Chartbeat data reported by Axios. That traffic is not coming back. Multistakeholder governance will not restore it. What publishers can do is demand a model that pays for citation — native advertising served inside AI-generated answers, with the publisher in the revenue chain. Bilateral licensing deals with OpenAI or Google are available to a few dozen outlets; the structural answer is a scalable ad layer that compensates every cited source, not just the ones large enough to negotiate directly.

Three signals that will tell you when the payment rail arrives

First, buying controls: brands still cannot purchase or exclude placement inside AI-generated answers with any precision — when granular targeting ships for in-answer inventory, the format is real. Second, publisher revenue share: Perplexity's revenue-sharing program is the earliest attempt, but payouts remain a fraction of licensing deals; when a platform ties publisher compensation to per-citation usage rather than lump-sum fees, the pricing model has shifted. Third, standardization: no standards body, industry coalition, or measurement vendor has yet defined a pricing benchmark for a sponsored citation — the moment one does, the category prices in and early movers compound their advantage.

Conclusion

Hold on to this: the governance debate about who controls AI answers is real, but governance without economics is aspiration without enforcement. The open-web link economy is giving way to an agentic-web citation economy, and both brands and publishers need a payment rail built for it. Smalk AI is building that rail — a generative engine advertising network that places native ads for AI agents inside AI-generated answers while opening a new revenue stream for the publishers whose content fuels those answers. What to watch next: which platform, standards body, or coalition sets the first pricing benchmark for a sponsored citation — that is the moment transparency graduates from principle to infrastructure.