San Francisco startup launches proprietary model designed to help AI agents distinguish trustworthy product information from manipulated online content
As AI increasingly takes over product research, recommendations, and purchasing decisions on behalf of consumers, San Francisco-based AI startup Onton has introduced a new model designed to solve one of the internet's growing challenges: determining which product information can actually be trusted.
Onton announced on July 29 the launch of Ontology 1, a proprietary AI model built specifically for the emerging agentic web, where AI agents increasingly act as autonomous decision-makers rather than simple search assistants. Unlike conventional recommendation models, Ontology 1 focuses on evaluating the trustworthiness and authenticity of product information before AI systems make purchasing recommendations.
The company said the model was developed entirely from scratch and outperformed leading product discovery platforms, including Google Shopping and Amazon, in head-to-head benchmarking across every category tested. According to Onton, the most significant performance advantage came in evaluating the veracity of product information—an area the company argues has become increasingly difficult as synthetic content and manipulated recommendations proliferate online.
The challenge has become more urgent as consumers increasingly rely on AI assistants to make purchasing decisions. A request as simple as finding the best sofa under $2,000 may prompt an AI agent to analyze thousands of reviews, comparison sites, influencer recommendations, sponsored rankings, and aggregated content. Yet much of that information may be generated, incentivized, or manipulated, leaving AI systems with little ability to distinguish reliable information from noise.
When poor recommendations result, they are typically viewed as bad purchasing decisions rather than failures of the underlying AI system. Onton argues that what was once considered an edge case has become an increasingly common characteristic of today's internet.
"Everyone is focused on building smarter agents," said Alex Gunnarson, co-founder of Onton. "We're focused on a different question: what should those agents trust? As commerce moves from clicks to conversations and eventually autonomous actions, the quality of information underneath those decisions becomes critical infrastructure."
According to the company, today's internet was built around human judgment. Reviews, recommendations, online communities, and social proof all assumed that people would apply context, intuition, and skepticism when evaluating information. As AI agents assume more of the discovery and decision-making process, however, those human filters disappear, making digital commerce substantially more vulnerable to manipulated information.
Ontology 1 addresses that problem by evaluating not only product characteristics but also whether the surrounding information can be trusted. By interpreting user preferences—including visual inputs—and distinguishing meaningful signals from potential noise, the model aims to provide AI agents with a more reliable foundation for product recommendations.
Alongside the launch, Onton released new research examining how today's leading AI systems evaluate product-related information. The findings suggest that existing models remain ill-equipped to navigate an online environment increasingly shaped by synthetic content and incentivized recommendations.
"We believe the next major internet platform will not be defined solely by who has the best model," said Zach Hudson, co-founder of Onton. "It will be defined by who can provide the most trustworthy foundation for those models to operate on."
Onton argues that a dedicated trust layer is becoming a prerequisite for agentic commerce rather than an optional feature. Without it, AI-powered purchasing decisions remain vulnerable to manipulation. With it, the company believes AI agents can deliver on their promise of helping consumers make faster, better-informed decisions based on information they can genuinely trust.
Ontology 1 is now available through Onton.com and is also being offered on a case-by-case basis to partners building AI applications for the emerging agentic web that require a trusted foundation for product discovery and recommendation.
