AI search changes the arithmetic of digital publishing faster than publishers can adapt
Qatar-based Al Jazeera saw growth in its monthly visits by 66% month on month and 53% year on year to 85.3 million in June 2025, at the same moment Reuters fell 20.7% to 80.6 million visits. The reversal was sharp enough to scramble the hierarchy of English-language news traffic in a single month. It arrived just as many publishers attributed traffic declines to Google’s AI Overviews and AI Mode, tools that answer readers before they ever reach a publisher’s homepage.
That collision — between AI systems that consume information and publishers that still depend on being visited — is beginning to expose which parts of the digital media economy were built on assumptions that no longer hold. News organisations spent two decades chasing audience scale through search and advertising. AI search changes the arithmetic because it extracts value from reporting while reducing the need to click through to the source itself. The companies losing traffic are not obscure outlets on the edge of the market. Reuters, whose business model rests partly on syndication and authority, still saw one of the biggest month-on-month declines. The pressure is moving upmarket.
The response from publishers has become increasingly contradictory. On one side, they warn that AI systems are hollowing out referral traffic. On the other, they are building the data infrastructure that makes those systems more effective. Al Jazeera tells readers that it uses information about past and present engagement with its site and services, including how often you visit and what type of content you usually like to view, to present more relevant content, make business and programming decisions, and deliver personalized advertisements. Readers are encouraged to consent because “we recommend that you allow us to use all this information to get the most out of our digital content.”
Publishers built a surveillance economy that AI companies now need to expand
The dependence runs deeper than recommendation engines. Precise geolocation data, persistent cookies and behavioural tracking became the machinery that financed digital publishing after advertising replaced subscriptions as the dominant revenue source. Al Jazeera’s own history shows how fragile that balance always was. The network started in 1996 by an emiri decree with a loan of 500 million Qatari riyals and later aimed to become self-sufficient through advertising by 2001. It failed. The Emir agreed to several consecutive loans instead. Even today, the organisation describes editorial independence as resting on funding through loans or grants rather than direct government subsidies. Advertising never disappeared as a structural necessity; it accounted for 40% of the station’s revenue in 2000.
AI systems now threaten the same advertising logic that kept digital newsrooms alive. The problem is not simply fewer clicks. It is that the surveillance architecture publishers built to maximise advertising efficiency is becoming politically harder to defend just as AI companies need larger pools of behavioural data. Cookie restrictions were supposed to weaken that system. Yet the evidence suggests much of the tracking economy survived intact. According to research, 54.773% of all cookies survived an imposed lifetime restriction of 30 days. Thousands of cookies continued generating value despite the restriction. The infrastructure proved resilient enough that regulators are now considering loosening parts of the framework instead of tightening it further.
That is the significance of the European Union’s proposed reforms. The bloc would delay stricter risk-management rules for “high-risk” AI until 2027 and allow tech firms to use anonymised personal data to train AI models. The same package would cut back on website pop-ups asking permission to use cookies. Henna Virkkunen, the EU tech chief, argued the changes would boost European competitiveness by simplifying rules about AI, cybersecurity and data protection. Her diagnosis was blunt: European companies are “held back by layers of rigid rules.”
European regulators are loosening the same safeguards publishers once defended
The trade-off is becoming visible in real time. Publishers spent years telling readers that consent frameworks protected privacy while simultaneously warning advertisers that restrictions would reduce targeting efficiency. Now AI companies are arguing that the same restrictions threaten Europe’s competitiveness. The result is a convergence of interests between platforms hungry for training data and publishers desperate to recover audience and advertising losses. Both sides increasingly depend on weakening the barriers that were introduced to limit behavioural surveillance in the first place.
That creates an uncomfortable position for the publishers currently posting growth. Al Jazeera’s audience surge arrived inside an ecosystem where the economics of attention are shifting away from the institutions producing reporting and toward the systems organising it. The company still asks users to accept tracking because disabling it means “some content and ads you see may not be as relevant to you.” But relevance is no longer the scarce commodity. Distribution is. AI systems increasingly control it, and the publishers surrendering behavioural data to remain visible are also helping train the infrastructure that reduces the need to visit them directly.
The contradiction is already embedded in European law. The EU AI Act applies outside the EU if AI outputs are used inside the bloc. A company in the EU can send data to an AI provider outside the EU, receive processed outputs back, and still fall within the framework. That means the regulatory perimeter follows the data even when the economics do not. Publishers carry the cost of producing reporting, consent management and audience acquisition. AI systems increasingly capture the interaction that once financed all three.
The market signal is not Al Jazeera’s growth or Reuters’ decline in isolation. It is that both now depend on the same weakening assumption: that gathering more behavioural data will preserve the economic value of being the place readers actually arrive. AI search is already pricing that assumption differently than publishers are, and the institutions still organised around advertising-driven traffic are discovering that the asset they spent two decades optimising — the visit itself — is no longer where the value settles.
