AI Search vs. Open Web: How AI Overviews Threaten Publisher Traffic

For much of the web’s commercial life, the bargain was straightforward: search engines helped people find information, and the sites that created that information received visits in return. Boston Review’s new five-writer symposium argues that AI-led search may be putting that arrangement under new pressure.

Published under the title What Was the Internet?, the collection does not offer a single forecast or a conventional product review of artificial intelligence. Instead, it brings together Siva Vaidhyanathan, Avery Dame-Griff, Joanna Walsh, Chad Wellmon and Cory Doctorow to examine what could be lost as answers, discovery and online communities become more dependent on large platforms and generative AI systems.

Its central question is deliberately broad: if people increasingly receive a synthesized answer before they encounter the sites, writers and institutions behind it, what happens to the economic and civic systems that produced those sources in the first place?

The issue is not only AI it is how people reach the web

The symposium’s sharpest concern is about the movement from links to generated responses. In his essay, Vaidhyanathan argues that conventional web search, however flawed, still presented users with visible sources that they could compare. AI-generated summaries may instead make a response appear self-contained, reducing both the incentive and the opportunity to visit the underlying publisher.

That is not simply a philosophical concern. A Pew Research Center analysis of browsing data from 900 U.S. adults found that users clicked a traditional search result in 8% of visits where an AI summary appeared, compared with 15% of visits without one. The study tracked 68,879 Google searches from March 2025. Users clicked a link cited within an AI summary in 1% of those visits.

Those figures do not prove that every AI summary causes a publisher to lose a visit. The study captured behaviour during a specific period, and search queries with AI summaries may differ from those without them. But the results help explain why publishers are watching AI search closely: traffic from search is often tied to advertising, subscriptions, donations and the chance to build a direct relationship with a reader.

Five writers, five different ways of describing a changing internet

Boston Review’s contributors approach the question from distinct directions.

• Siva Vaidhyanathan focuses on the power of search infrastructure. His essay argues that the key change is from a system that directs people toward sources to one that generates an answer on their behalf. He frames this as a problem of power over how public knowledge is organised and encountered.
• Avery Dame-Griff looks back at the smaller, often volunteer-run online communities that helped people form friendships, explore identity and build local connections. His concern is not that the early internet was ideal, but that scale and engagement-driven platform design have made meaningful connection harder to sustain.
• Joanna Walsh considers what platform changes mean for writers and independent literary publications. Her contribution links the weakening of social referrals and search traffic to the difficulty of finding audiences for work outside commercial publishing.
• Chad Wellmon traces the intellectual history of search itself. He contrasts the older experience of using results pages to explore, compare and follow references with a model designed to anticipate and satisfy a question immediately.
• Cory Doctorow places the web’s decline in a policy context. His argument is that weak competition enforcement, privacy law and repair rights helped create the concentrated platforms that now shape so much online life.

The result is less a eulogy for a vanished internet than an argument about infrastructure. The writers disagree in emphasis, but they share a concern that the systems controlling attention increasingly sit between people and the sources they rely on.

Google presents a different case for AI search

Google’s own position is more optimistic. In its guidance for site owners, the company says that AI Overviews and AI Mode show supporting links, can help people explore information and may surface a broader range of useful pages than a classic search. Google also says clicks from pages with AI Overviews tend to be higher quality, meaning visitors are more likely to spend time on the site.

Google’s documentation adds that AI features use the same core search requirements as regular results. There is no separate “AI index,” special markup or additional technical qualification that guarantees a page will appear in an AI Overview or AI Mode response.

That leaves publishers with an unresolved measurement problem. Google records AI-feature traffic within the overall Web performance reporting in Search Console, while the company’s public guidance focuses on aggregate search performance. For a newsroom, nonprofit or small publisher trying to understand whether AI answers are helping or displacing its work, aggregate data may not fully explain the change in audience behaviour.

The disagreement is important. Boston Review’s writers emphasise the risk that summary-first search makes original sources less visible. Google argues that AI search can create more complex questions and more paths to relevant sites. Both can be true in different cases, which is why the question cannot be settled by a single traffic statistic.

Why the link still matters

A list of links asks readers to make judgments. They can see that one result is a government agency, another is a university, another is a local publication and another is a personal blog. That process is imperfect ranking systems have always reflected commercial incentives, spam and bias but it makes the source visible.

A generated answer changes the reading experience. It can be fast and useful, especially for narrow questions or complicated comparisons. But it can also compress disagreement, context and authorship into a fluent response. The concern raised throughout the Boston Review collection is not that readers must reject AI tools. It is that convenience should not obscure who produced the knowledge, who is accountable for it, and whether those producers can continue their work.

This is especially relevant for material that cannot be reduced to a quick answer: local reporting, specialist research, criticism, archives, creative work and community knowledge. These are areas where the value often lies in the source’s evidence, perspective and context not only in a short conclusion.

TRADITIONAL SEARCH ENGINE INDEXING VS. GENERATIVE AI-LED SEARCH
DIMENSION / METRIC	TRADITIONAL LINK SEARCH (1998–2023)	AI-GENERATED SUMMARY SEARCH (2024+)
Discovery Mechanism	Ranked list of diverse domains, institutions, and authors	Single synthesized, conversational answer block
Outbound Click-Through Rate	~15% of total search visits	~8% overall (only ~1% on cited source links)
Publisher Value Exchange	Delivers traffic for monetization (ads, subscriptions)	Extracts training data & real-time facts with minimal referral
Source Evaluation	User actively evaluates multiple competing sources	Algorithmic model blends sources into a unified summary
Monetization Architecture	Paid sponsored links alongside organic web results	In-engine ad modules embedded inside answer engines
Search Console Visibility	Clear separation of standard web impressions & clicks	Consolidated aggregate reporting without discrete AI breakdowns

What the symposium does and does not claim

What Was the Internet? is a collection of critical essays, not a neutral research report or a prediction that the web will disappear. Its contributors use forceful language and make normative arguments about corporate power, culture and policy. Those arguments should be read as the writers’ interpretations of a changing digital environment.

Still, the underlying question is practical. If the web’s original publishers receive less visibility and less revenue, who will fund the next explanation, investigation, review, essay or public resource that an AI system may later summarise?

Boston Review does not offer one technical fix. Its contributors instead point toward a wider set of choices: stronger public institutions, healthier competition rules, direct relationships between publishers and audiences, accountable platform design, and online spaces built for communities rather than endless scale.

The web is not gone. But the symposium argues that how people discover it and whether they still arrive at the work behind the answer is becoming one of the most consequential media questions of the AI era.

AI SEARCH OVERVIEWS & WEB TRAFFIC: MYTH VS. REALITY
MYTH	REALITY
"AI summaries always send higher referral traffic to cited sources."	False. Pew Research data shows only 1% of search visits result in a click on links embedded inside AI summaries.
"Web publishers can use special markup tags to guarantee inclusion in AI Overviews."	False. Google documentation confirms that standard web indexing rules apply with no dedicated markup for AI inclusion.
"Traditional search engines never made editorial choices."	False. PageRank and ranking algorithms always reflected commercial weights, though they preserved visible lists of distinct domains.

Sources and verification

• The five-essay symposium and the arguments attributed to its contributors: Boston Review, August 6, 2026.
• Browsing-data study of 900 U.S. adults and 68,879 Google searches; 8% versus 15% traditional-result click finding: Pew Research Center, July 22, 2025.
• Google’s account of how AI Overviews and AI Mode surface links, eligibility requirements and measurement: Google Search Central.
• Google’s May 2026 documentation update on optimisation guidance and AI-feature support: Google Search documentation updates.

• What Was the Internet?

Editorial note: This article reports on Boston Review’s symposium and clearly attributes its arguments. It also includes relevant independent research and Google’s stated position to distinguish evidence, company claims and editorial analysis. It does not treat the symposium’s predictions as settled fact.