A new report from SEO software company Search Atlas reveals a significant divergence between how large language models (LLMs) cite sources and how Google traditionally ranks them. The comprehensive analysis, which compared citations from OpenAI's ChatGPT, Google's Gemini, and Perplexity against Google search results across 18,377 matched queries, uncovers a notable gap between traditional search visibility and AI platform citations. This finding has crucial implications for digital content strategies and search engine optimization (SEO).
Perplexity Leads in Search Alignment
As an LLM designed for live web retrieval, Perplexity's citation patterns are expected to closely mirror traditional search results, and the study confirms this. Search Atlas found that Perplexity exhibited a median domain overlap of approximately 25–30% with Google's results. Its median URL overlap was also significant, nearing 20%. Overall, Perplexity shared 18,549 domains with Google, accounting for about 43% of all domains it cited.
ChatGPT and Gemini: A More Selective Approach
In contrast, ChatGPT demonstrated a considerably lower overlap with Google's search results. Its median domain overlap hovered between 10–15%, sharing only 1,503 domains with Google, which represents about 21% of its cited domains. URL-level matches for ChatGPT typically remained below 10%.
Google's Gemini, on the other hand, displayed less consistent behavior. While some of its responses showed minimal to no overlap with search results, others aligned more closely. Despite this variability, Gemini shared a mere 160 domains with Google. This figure represents only about 4% of the domains found in Google's results, even though these specific domains constituted a substantial 28% of Gemini's overall citations.
Implications for SEO and Digital Visibility
The findings clearly indicate that a strong Google ranking does not automatically translate into citations from large language models. This suggests that LLMs retrieve and process information from the web in fundamentally different ways.
For retrieval-based systems like Perplexity, which actively search the web, traditional SEO signals and overall domain authority are likely to remain crucial for visibility. Websites that already perform well in Google's rankings are more prone to appear as sources in Perplexity's answers.
However, for reasoning-focused models such as ChatGPT and Gemini, which lean more on pre-trained knowledge and selective retrieval, the direct influence of current Google rankings on source citation appears to be significantly diminished. These models tend to cite a narrower range of sources, with very low URL-level matches to Google's top results.
Acknowledging Study Limitations
Search Atlas also highlighted several limitations within their study. The dataset was heavily skewed towards Perplexity, which accounted for a dominant 89% of the matched queries, while OpenAI's ChatGPT made up 8% and Gemini only 3%.
Query matching relied on semantic similarity scoring, meaning paired queries expressed similar information needs rather than being identical user searches. A threshold of 82% similarity was used, based on OpenAI's embedding model.
Furthermore, the analysis covered a two-month period, offering only a recent snapshot. Longer timeframes would be necessary to determine if these overlap patterns remain consistent over extended periods.
The Future of SEO in an AI-Driven Landscape
Looking ahead, the study suggests a bifurcated approach to SEO strategy in the evolving AI landscape. For LLMs that primarily function as retrieval-based systems, like Perplexity, traditional SEO best practices, strong domain authority, and high-quality content will likely continue to be paramount for achieving visibility.
Conversely, for reasoning-focused models such as ChatGPT and Gemini, the direct impact of conventional SEO signals on source selection may diminish. This could necessitate new strategies for content creators and marketers to ensure their information is discoverable and cited by these advanced AI platforms.
