The open web ran on a feedback loop: a user searched, clicked, read, and by doing so generated the signals — subscriptions, backlinks, dwell time, repeat visits — that helped the next user find the same quality source. AI Search is breaking that loop twice. First by suppressing the click. Then by eliminating the signal the click would have produced. The first loss is measurable. The second is structural, and no licensing deal or traffic-recovery strategy can fix it.

  • A randomized field experiment by Agarwal and Sen (SSRN, April 2026, revised June 2026) found that Google AI Overviews reduce organic clicks by 39.8% on queries where they appear — the first causal estimate, not observational correlation.
  • User satisfaction did not improve when AI Overviews were present: survey scores for information quality and ease of finding information were statistically indistinguishable between treatment and control groups.
  • Alphabet's own Q1 2026 earnings confirm the downstream effect: Google Network ad revenue — the proxy for publisher monetization via AdSense and Ad Manager — fell 4% year-over-year to $6.97 billion, while Google's owned search revenue grew 19%.
  • The deeper risk is not traffic loss but signal loss. When clicks vanish, so do the subscribers, backlinks, bookmarks, and reputation markers that helped search systems and future readers find quality sources — what researcher Jason Chan calls 'durable attention capital.'
  • Brands that re-architect around citation-based visibility and publishers that price on being the source AI engines cite will compound an advantage. The rest will discover the cost of waiting.

What durable attention capital is and why it matters now

Durable attention capital is the accumulated stock of quality signals that a piece of content or a publisher builds through repeated human interaction: subscribers, repeat readers, backlinks, bookmarks, search authority, and reputation. Jason Chan's June 2026 paper, reported by The Register on July 1, frames this concept as the overlooked casualty of AI Search. An AI answer, Chan argues, can use publisher content while keeping the user inside the AI interface. The user may be served in the short run. But the source loses the visit, the revenue from the visit, and the quality signal that the visit would have produced.

This is not the familiar 'publishers lose traffic' argument. It is a systems argument: when the signals that distinguish costly human information from cheap AI imitation stop forming, the entire discovery layer degrades. The web does not collapse because content disappears. It collapses because the ability to tell good content from bad disappears with the signals that measured it.

The click economy and the citation economy, side by side

In the open-web click economy, a search result was a referral. The publisher supplied the content, Google ranked it, the user clicked through, the publisher monetized the visit with display advertising, and the user's behavior — time on page, scroll depth, subscription, return visit — generated signals that reinforced the publisher's authority. Brands bought visibility by bidding on the click. Publishers funded content production by selling the impression the click delivered. The unit of account was the visit.

In the agentic-web citation economy, the search result is a synthesized answer. The publisher still supplies the content — AI Overviews draw from an average of 13.34 sources per response according to SE Ranking's 2025 data, nearly double the 6.82 of 2024 — but the user stays inside the AI interface. The publisher's content is cited but not visited. The brand's name may appear inside the answer, but no click is registered. The signal chain that fed the old economy — visit, impression, revenue, authority signal — breaks at the first link.

What three sources reported, and what we verified

PPC Land's July 1, 2026 coverage of the Agarwal and Sen SSRN paper (abstract ID 6513059) reports a 39.8% reduction in organic clicks when AI Overviews are present, based on the paper's June 17, 2026 revision. The original April 3 version of the same paper reported 38%. Multiple independent sources — Search Engine Journal, Citera, and the MacMD Viewer statistics roundup — cited the 38% figure from the first version. We use the revised 39.8% figure as the most current. The zero-click increase of 34.5% and the null result on satisfaction are consistent across both versions.

Fast Company's Pete Pachal, also writing on July 1, 2026, highlighted the New York Times–Oumi accuracy study: AI Overviews are accurate 91% of the time, which at Google's scale produces millions of inaccurate summaries per hour. Pachal's piece noted QuickSEO's April 2026 figure of 60.23% AI Overview prevalence. That number sits at the high end of a wide range — Conductor's Q1 2026 benchmark across 21.9 million queries puts prevalence at 25%, BrightEdge's commercial-vertical tracker at 48%. The variance reflects different keyword mixes and detection methods. What is not in dispute is the direction: AI Overviews are present on a material and growing share of queries.

The Register's July 1, 2026 report on Chan's research adds the signal-loss dimension that neither the Agarwal–Sen experiment nor the Fast Company piece addresses. Chan's thesis — that AI answers strip not just revenue but the measurement events that sustain content quality — is the piece the traffic-decline conversation has been missing.

Why 'better clicks' is not a defence of the status quo

Google's quality-click argument at its strongest

Google VP Liz Reid has argued that AI Overviews cut low-quality 'bounce clicks' and that the clicks users do make after seeing an AI summary reflect more targeted intent. The strongest version of this position is that AI Search acts as a filter: it handles the informational queries that never needed a site visit and sends only high-intent users downstream. If true, publishers would lose volume but gain conversion quality, and the trade-off could be net positive.

Why the data contradicts the quality-click thesis

The Agarwal–Sen experiment tested this claim directly. Three engagement measures — back-button navigation probability, sub-ten-second bounce rate, and time on page — showed no meaningful difference between sessions that included AI Overviews and those that did not. The additional clicks generated when AI Overviews were removed were, by every measured dimension, just as engaged as the clicks that existed already. Google's quality-click argument has no public data behind it. The first randomized experiment to test it found nothing.

And even if a future study did find a conversion-quality gain, it would not address Chan's signal-loss problem. A higher-quality click still requires a click. When the click itself vanishes — as it does in 83% of AI Overview sessions and 93% of AI Mode sessions, according to Seer Interactive and Semrush respectively — no amount of conversion improvement can regenerate the attention signals that funded and ranked the source.

What this means for brands and for publishers

For CMOs, media buyers, and agencies: the signal gap is a strategy gap

If 40% of the organic clicks your brand relied on for attribution are being absorbed by AI answers, your measurement stack is already lying to you. The brands being cited inside AI Overviews see a 35% lift in organic clicks and a 91% lift in paid clicks compared to uncited competitors on the same SERP, according to QuickSEO's 2026 analysis. Citation visibility is not a future problem. It is a current channel that most marketing teams have no reporting for. CMOs should treat AI visibility as a distinct discipline with its own budget, KPIs, and owner — not a line item inside SEO.

For publishers: the revenue model that fits citation, not traffic

Alphabet's Q1 2026 earnings laid the asymmetry bare: Google's owned search revenue up 19%, publisher Network revenue down 4%. Google monetizes the answer its AI wrote using publisher content. The publisher whose content grounded that answer gets neither the visit nor the ad impression. Cloudflare's July 1, 2026 shift from Pay Per Crawl to Pay Per Use is the first infrastructure-level attempt to reconnect publisher value to AI output. But crawl-pricing and licensing deals are stopgaps. The structural answer is a native ad layer inside AI-generated answers — one that compensates the publisher whose content is cited, not just the platform that rendered the answer.

Four signals that show the citation economy is pricing in

First, Cloudflare's September 15, 2026 deadline will force AI companies to separate search crawlers from training and agent crawlers, or face default blocks on ad-monetized pages — a clear market signal that the industry is moving from voluntary norms to enforced economic boundaries. Second, Cloudflare's Pay Per Use model, launched July 1, 2026 with partners Ceramic.ai and You.com, pays publishers when their content appears in AI answers rather than when it is crawled — the unit of value has already shifted from fetch to citation. Third, 54% of US marketers plan to implement Generative Engine Optimization strategies within the next three to six months, according to Exposure Ninja's 2026 survey. Fourth, OpenAI has posted job listings describing six different ad formats for ChatGPT, confirming that the ad layer inside AI answers is not theoretical — it is being engineered now.

Conclusion

Hold on to this: the open-web click economy is being replaced by an agentic-web citation economy, and the deeper casualty is not lost traffic but lost signals — the durable attention capital that made quality content findable and fundable in the first place. Smalk AI is built for exactly this gap — an AI Search ad network that places native ads inside AI-generated answers, giving brands citation-based visibility and giving publishers a revenue stream tied to the content AI engines actually cite. What to watch next: Cloudflare's September 15 crawler-separation deadline and OpenAI's emerging ad formats will set the first structural pricing norms for the citation economy — the moment both signals converge, the category prices in.