San Francisco-based product discovery company Onton has launched Ontology 1, a proprietary AI model designed to help shopping agents distinguish trustworthy product information from synthetic content, manipulated recommendations and incentivised reviews.
Announced on 29 July 2026, the model has been developed for the emerging agentic web, where artificial intelligence systems increasingly research products, make recommendations and complete purchases on behalf of consumers.
Onton says Ontology 1 was built from scratch and outperformed established product discovery platforms, including Google Shopping and Amazon, across the accuracy dimensions evaluated in its internal head-to-head benchmarks.
According to the company, the most significant performance difference concerned the model’s ability to evaluate the reliability of product information.
Addressing the Trust Problem in Agentic Commerce
AI shopping agents can analyse thousands of product listings, reviews, social posts, sponsored comparisons and influencer recommendations within seconds. However, much of the information available online may be commercially incentivised, artificially generated or deliberately manipulated.
Existing AI systems can consequently recommend products without determining whether the information supporting those recommendations is reliable.
This problem becomes more significant as commerce moves from traditional searches and clicks towards conversational interfaces and autonomous purchasing. A consumer may ask an AI agent to find a sofa below a particular price, for example, but the resulting recommendation could be based on synthetic reviews or sponsored content that the agent has treated as independent evidence.
“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.”
Evaluating Information, Attributes and Intent
Ontology 1 is designed to assess not only what a product is, but also whether the information surrounding it can be trusted.
The system interprets product attributes alongside the intentions and preferences expressed by a user. These preferences can include visual inputs, which the platform processes as meaningful discovery data.
Onton says this approach enables AI systems to generate recommendations based on more reliable information while reducing their exposure to synthetic and manipulated content.
The launch is accompanied by company research examining how leading AI systems evaluate product information and where they encounter difficulties when presented with incentivised recommendations or artificial content.
Onton argues that many existing platforms were built for an internet designed around human evaluation. People can apply context, intuition and scepticism when reading reviews or comparing recommendations. AI agents do not necessarily possess the same filters, leaving automated decision-making systems vulnerable to manipulation.
Building a Trust Layer for AI Agents
Onton positions Ontology 1 as part of a broader trust layer for the agentic internet rather than simply another ecommerce recommendation system.
“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.”
The company believes reliable information infrastructure will become essential as AI agents assume greater responsibility for commercial decisions. Without mechanisms for assessing authenticity, autonomous purchasing systems may continue to operate on information that can be easily influenced or gamed.
Although Ontology 1 has initially been developed around product discovery, Onton says the underlying research may have applications beyond ecommerce. Its approach could potentially support other AI systems required to evaluate large and complex information environments.
Availability
Ontology 1 is available to users through Onton’s product discovery platform. The company is also making the model available on a case-by-case basis to partners developing services for the agentic web.
Onton was publicly launched in late 2023 and describes its mission as helping consumers make product decisions more quickly. The company says its platform has since guided millions of users through product discovery and selection.
