AI SEO Study: ChatGPT, Perplexity, and AI Overviews Cite Wildly Different Content

Although most AI tools fall under the umbrella of Large Language Models (LLMs), ultimately serving a similar purpose, there’s no denying that different models deliver notably different answers – even to the exact same queries.

For recreational users, this is just a fun factoid. For individuals who depend on these tools for high-level decision-making, however, knowing which models cite which content types is a critical part of AI Overviews optimization, ChatGPT marketing, and Perplexity SEO strategies.

In March 2026, ZeroClick Labs conducted a study on ChatGPT, Perplexity, and AI Overviews, aimed at determining which content types are most frequently selected as citation sources. This article is a comprehensive review of study results, focusing on each specific model’s behavior.

Methodology:

• Time window: 30 days
• Total URLs analyzed/cited: ~3.3K
• Total citations observed: ~38.5K
• Platforms: ChatGPT, Perplexity, Google AI Overviews (AIO)
• Industries: SaaS, digital marketing, home services
• Overall citations per URL: ~11.67 (38.5K ÷ 3.3K)
  • NOTE: Approximate; topline values are rounded;

Which content types are most commonly cited by AI models?

Listicles, product pages, and homepages were consistently the most frequently cited content types across all three observed AI models, accounting for ~69% (Perplexity) to up to 77.5% (ChatGPT) of total selections. 

This distribution isn’t unexpected, considering how most LLMs share the same RAG selectability criteria: precisely defined entities, modular organization, and explicit information. The reason is that content structured this way is easier for LLMs to extract and reuse in answer synthesis. Since all three dominant page types meet these criteria, it’s natural that they are consistently surfaced across the observed platforms.

However, while all three models prioritize the same top content types, the degree to which each is cited varies significantly across platforms. In addition, as we move further down the citation line, the content-type distribution begins to diverge dramatically, revealing meaningful secondary trends, most notably:

This data shows something very interesting: although the citation selectability criteria coincide, the intensity signals vary between each criterion across all three models, resulting in dramatically different citation behavior:

• ChatGPT → product-page dominant citation behavior, with the highest number of 404 crawls.
• Perplexity → source-agnostic citation behavior (balanced distribution w/ noticeably high discussion footprint).
• AI Overviews → listicle dominant citation behavior, with the highest share of video content.

Why are different platforms rewarding different content types?

While it may appear so, different content type selection patterns are not preferential; they are structural – rooted in system architecture or, rather, design constraints. Essentially, each of the 3 AI platforms solves fundamentally different problems:

ChatGPT: The “answer-first” generator

• Objective: Provide the most direct answer to a query.
  • Requirements for realization: Clean input, a high signal-to-noise ratio, and structured data.
• Hence, favoring product pages with clear definitions, structured information (features, pricing, specs), and high factual density.

Perplexity: Research aggregator + synthesizer

• Objective: Explore breadth and back it up with transparency, rather than just “provide an answer”.
  • Requirements for realization: Diverse viewpoints and an abundance of citations from different sources.
• This is why we’re seeing balanced top and mid distribution, as well as source-agnostic behavior (heavy inclusion of discussion content).

AI Overviews: “At-scale SERPs summarizer”

• Objective: Deliver options, rather than definitive solutions, and do so quickly.
  • Requirements for realization: Pre-organized, immediately usable data.
• AIO heavily favors listicles since they are the easiest and most resource-efficient way to deliver the highest volume of options in the least amount of time.

Attribution philosophy difference: Citations vs URL count

The final metrics that bear comparative analysis are total citations and URL counts, since these patterns reveal meaningfully different philosophies toward source attribution across the three platforms.

• AI Overviews cite a significantly greater number of unique URLs overall, most commonly pulled from the top 10 Google search results. However, they produce fewer citations per response, resulting in a lower total citation count. Notably, AI Overviews use the same URLs across chats way more often than other sources, even if they’re not explicitly citing them, suggesting that AIO prioritize coverage breadth over depth.
• ChatGPT sits on the diametrically opposite end of the spectrum: it has the highest total citation count but the lowest URL count, and it relies on Bing indexing and ranking, rather than Google. This pattern indicates that it explicitly references sources within responses more frequently than other platforms, but draws from a comparatively narrower set of high-trust ones – indicative of the confidence-first approach.
• Perplexity falls right between the two extremes, balancing source diversity and consistent citation visibility. Paired with a substantial presence of discussion-derived citations, the behavior aligns with its source-agnostic, transparency-first philosophy, where information origin clarity matters as much as response quality.

While all three platforms ultimately serve a similar purpose, the way each handles attribution and how it arrives at the end result differs meaningfully – which can influence how users perceive, verify, and trust the synthesized information. 

This is, in part, why private users and decision-makers alike are growing skeptical of search-integrated and generic AI platforms and gravitating toward alternatives like Claude and Perplexity that prioritize response integrity and directness over simple source aggregation. 

When attribution is inconsistent or opaque, trust erodes; and for users making consequential decisions – especially those in high-stakes environments – the cost of that erosion is simply too high, both metaphorically and realistically.

Seth Matthews, an AI SEO content writer within ZeroClick Labs, began his career in civil engineering, but his passion for writing drove his deliberate transition into professional content strategy and SEO. The clash of two worlds produced a writer who can effortlessly combine technical precision and narrative clarity with the emotional and psychological impact of creative storytelling.