How AI Models Rank Top Companies: An Analysis of Source Attribution

AI Models and Source Selection: A Deep Dive

A recent investigation into the operational mechanisms of artificial intelligence reveals a significant reliance of AI models on pre-existing compilations when responding to queries involving terms like ‘best,’ ‘top,’ or ‘rankings.’ An analysis of 208 AI responses that included source links demonstrated that 130 of these (62%) directly cited ready-made lists, such as third-party top-10 compilations, rankings, or comparative reviews. This finding suggests that for businesses aiming to appear in AI-generated results, inclusion in such aggregated lists is critically important.

Reliance on Curated Lists and Dominant Resources

The study encompassed 368 responses across 32 diverse projects, ranging from accounting outsourcing to fulfillment services, with 208 responses containing source links. The conclusion that an AI model’s output is largely determined by the existing compilations it discovers is paramount. This underscores that AI algorithms frequently do not generate their own rankings but rather synthesize information from already published lists.

Further examination of sources indicated that when five search-enabled neural networks were queried with 20 questions related to promotional services, 211 unique domains were identified. Notably, 161 of these were referenced by only one neural network. However, a mere 17 websites were cited by three or more neural networks, yet these 17 resources accounted for 79% of all linked responses. More than half of the citations originated from just two domains: Sostav and vc.ru. A follow-up check six weeks later reinforced this pattern, with nine of the 17 dominant sites remaining common across three or more neural networks, highlighting the stable preference of AI models for specific, authoritative resources.