AI Agents Struggle With Website Pricing Information

AI agents are increasingly interacting with websites, with Google introducing agentic tasks in Search and Salesforce reporting 20% of sales originating from agents. A recent analysis by the founder of Siteline, David Kaufman, and the author examined how AI agents navigate B2B websites and identified a significant breaking point: pricing information. The study found that while most sites are generally agent-ready, pricing pages frequently impede an agent's ability to extract and cite necessary data.

AI agents process information differently than humans. They receive tasks, search the web, retrieve pages, extract facts, and cite their sources. A website that might effectively persuade a human visitor can fail an agent if the required facts are opaque, difficult for machines to read, or subject to access friction that prevents citation. This transforms websites from visual showrooms into data-driven barcodes for agents.

The analysis involved giving AI agents three buyer-related tasks—finding pricing/features, integrations, and security/compliance—for 100 B2B products. Agents were tasked with finding the official website independently, without starting links. Each task was executed five times to account for the probabilistic nature of large language models. The focus was not on whether information existed online, but on whether agents could reliably retrieve it directly from the vendor's site.

Pricing pages emerged as the most challenging area. The study highlights that pricing is a critical juncture for prospects, representing high buyer intent. However, pricing pages must balance three competing demands: companies wish to control pricing disclosure, buyers desire rapid comparison, and agents require clear, retrievable, and citable facts. When AI agents attempt to access pricing details, they encounter significantly more difficulties than with other types of information, such as security and compliance data.