A comprehensive new analysis of over 193,000 AI-generated responses from platforms including ChatGPT, Google AI Mode, and Google AI Overviews has provided the most detailed look yet at the factors influencing how artificial intelligence sources information. Conducted by TJ Digital, the study indicates a strong correlation between the structural clarity of a webpage and the likelihood of it being cited as a source by generative AI models. The findings suggest that the "Generative Engine Optimization" (GEO) landscape is shifting toward a preference for direct, front-loaded information, effectively prioritizing user intent over traditional SEO metrics like domain authority alone.
The research, led by TJ Robertson, processed a massive dataset consisting of 193,507 AI responses addressing 4,167 distinct, non-branded questions. These queries spanned 78 client companies across nine diverse business models, with data captured through eight distinct snapshots between August 27 and October 21, 2026. By utilizing the Peec AI application to automate the prompting of AI models, the study systematically mapped how these tools selected sources from the broader web.
The Primacy of the Direct Answer
At the heart of the report is the finding that cited pages are significantly more likely to provide an immediate answer to a user’s query. When comparing pages cited by AI against those that appeared in Google’s Top-10 search results but were ignored by the AI, the disparity in content structure was stark. Cited pages were four times more likely (20% versus 5%) to begin with a direct answer to the question posed.
This pattern persisted even when controlling for visibility. When comparing cited pages to those that the AI identified in its search results but ultimately chose not to cite, the cited pages still outperformed their counterparts, with 23% providing a direct answer up front compared to only 14% of the non-cited pages. This suggests that while AI models are increasingly capable of scanning broad swathes of content, they prioritize pages that minimize cognitive load by offering the "bottom line" information in the opening paragraphs.
Methodology and Comparative Analysis
To ensure the integrity of these findings, TJ Digital employed a two-round blind page rating process. Independent raters were presented with 195 pairs of pages without knowing which page had been cited by the AI. This blind test confirmed that cited pages not only offered more direct answers but also scored higher on subjective metrics of completeness. On a scale of 1 to 5, cited pages averaged a completeness rating of 2.3, whereas "passed-over" pages averaged a significantly lower 1.7.
The study also highlighted a notable lack of differentiation regarding the tone of the content. Approximately 75% of the pages analyzed were commercial in nature—representing businesses promoting their own products or services. Both cited and passed-over pages received nearly identical scores for "promotional tone," suggesting that AI models do not necessarily penalize pages for having a marketing focus, provided the core informational value remains accessible and relevant.
Correlative Factors: Beyond the Direct Answer
Beyond the immediate structural layout, the analysis identified several specific page traits associated with higher odds of being cited. These findings provide a roadmap for digital marketers attempting to navigate the transition from traditional search to generative AI environments:
- Temporal Relevance: Content featuring a specific year in the title saw more than double the odds of being cited. Furthermore, pages with a publish or update date within the previous 12 months experienced an 88% increase in citation probability. This reinforces the AI preference for "fresh" data in an era of rapid information cycles.
- Structured Data and Formatting: The inclusion of tables within the content was linked to a 69% increase in citation likelihood. Additionally, the presence of schema markup was associated with a 13% higher chance of being cited.
- Trust Signals: First-hand experience markers, such as author expertise or original insights, were linked to a 45% increase in citation probability.
- Transactional Data: The inclusion of pricing information saw a 28% increase in odds, likely because pricing is a factual, high-intent data point that AI models are designed to aggregate.
Interestingly, the study identified negative correlations that remain difficult to explain. Pages with an author byline saw a 28% decrease in citation odds, and those featuring a prominent call to action (CTA) experienced a 12% decline. Robertson noted that while these results were statistically present, they do not necessarily imply a causal relationship; removing these elements without further testing would be premature for most site owners.
The Context of Previous Industry Research
The findings from TJ Digital align with, yet expand upon, earlier investigations into the mechanics of AI search. For instance, a May 2024 report from Ahrefs investigated the role of JSON-LD schema markup. While Ahrefs found that cited pages were three times more likely to contain structured data, subsequent testing revealed that retroactively adding JSON-LD did not lead to a statistically significant increase in citations.
This creates a nuanced reality for SEO professionals: correlation is not causation. The fact that high-quality, well-structured pages happen to use schema markup does not mean that schema is the primary "trigger" for AI citation. Instead, it is likely that these elements act as part of a holistic ecosystem of high-quality content that AI models are trained to prioritize.
Broader Implications for Digital Strategy
For companies and content creators, the implications of this study are twofold. First, the data highlights that the traditional "SEO for links" model is being superseded by a "GEO for clarity" model. If a brand wants to be featured in an AI summary, the primary hurdle is no longer just domain authority, but the accessibility of the information. Providing a clear, concise answer at the top of a page—often referred to as the "answer first" approach—appears to be the most effective strategy for capturing citations.
Second, the study serves as a warning against over-optimizing for specific technical markers without considering the user experience. Because the study only captures associations, it cannot guarantee that simply changing the structure of a page will result in more citations. As Robertson emphasized, his company’s findings are intended to be used as a foundation for individual testing. Businesses are encouraged to monitor the specific questions they track in their industry and use these patterns to conduct their own A/B testing rather than blindly adopting every trait identified in the data.
Future Research and the Road Ahead
Despite the breadth of the 193,507-response dataset, the industry remains in the early stages of understanding the "black box" of generative engines. The lack of visibility into the specific AI models used to evaluate some of the pages—and the temporary unavailability of the agency’s internal graphs—underscores the volatility of this research area.
Moving forward, the industry requires more rigorous experimental design, similar to the Ahrefs schema test, to move beyond observation. To truly prove that a specific content trait causes a citation, researchers will need to implement controlled trials where specific variables (such as adding a direct answer) are introduced to one group of pages while another remains unchanged.
For now, the TJ Digital report provides a vital baseline for digital strategists. By focusing on the structural needs of AI—specifically the demand for immediate, accurate, and current information—marketers can begin to align their content strategies with the future of search. The era of generative engine optimization is still in its infancy, but the message from the data is clear: the AI tools of tomorrow are looking for the most direct answers today.
