Onton launches AI model to judge product trustworthiness

Onton has launched Ontology 1, an artificial intelligence model for product discovery designed to assess the trustworthiness of product information used by AI agents.

The San Francisco-based group built the model from scratch and tested it against product discovery services including Google Shopping and Amazon. It said the biggest performance gap appeared in judging whether product information was accurate and reliable.

The launch comes as more online shopping journeys begin with AI assistants rather than conventional search or marketplace browsing. In that setting, the quality of underlying information becomes more important because software agents may scan reviews, sponsored comparisons, recommendation lists and social posts without the human judgement shoppers would normally apply.

Onton argues this creates a growing problem for online retail. If an AI agent cannot distinguish between genuine product information and content shaped by incentives, manipulation or automation, it may still return a recommendation that appears credible to the user.

A research paper released alongside the product examines how AI systems assess product-related information and where they struggle when synthetic material enters the process. According to Onton, existing systems remain vulnerable when asked to navigate recommendation environments that mix authentic sources with commercially influenced or generated content.

Trust question

Onton positions Ontology 1 as a model that evaluates not only a product description but also whether the surrounding information can be trusted. It said the system interprets user preferences, including visual inputs, and treats those signals as part of the decision-making process rather than as extraneous data.

That approach reflects a broader shift in eCommerce, as developers try to build agents that do more than retrieve listings. As those systems move towards making or executing purchase decisions, trust in the inputs becomes a separate technical and commercial issue.

Onton said its work may also have uses beyond eCommerce. The company described the model as showing signs that it can generalise to other information environments where systems must distinguish between credible and manipulated material.

Benchmarks cited by Onton put Ontology 1 ahead on accuracy across the dimensions tested in its comparison with established consumer product platforms. While it did not disclose full third-party validation in the material provided, it said its findings were backed by a white paper.

Two founders framed the launch as part of a wider debate over what AI systems should rely on when they act for users.

"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."

His comments point to an issue that extends beyond product search. Much of the internet's recommendation structure was built for people reading reviews, weighing sources and applying scepticism, but that model changes when software acts in their place.

Wider implications

Onton sees a trust layer as a necessary part of agent-led commerce rather than an optional addition. That framing suggests a market opening for specialist models that sit beneath consumer-facing assistants and filter the information those assistants rely on.

For major technology groups and retail platforms, the question is whether trust assessment becomes a distinct market in its own right or is absorbed into broader shopping and search models. Product discovery has long been dominated by large platforms with scale in listings, reviews and ad inventory, but a shift towards autonomous shopping agents could change where value sits.

Onton, which describes itself as a search and discovery engine for products, said it has grown since its public launch in late 2023 to serve millions of users. With Ontology 1, it is seeking to differentiate itself not through the breadth of products indexed, but through the reliability of the information behind 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."