- In head-to-head benchmarks against the most dominant platforms in product discovery — including Google Shopping and Amazon — Onton’s system outperformed on accuracy across every dimension tested.
- Proprietary research from scratch is designed to help AI systems cut through synthetic content and manipulated recommendations as purchasing decisions increasingly shift from humans to agents.
Onton, a San Francisco-based startup building the infrastructure layer for trust and authenticity on the agentic web, today announced the launch of Ontology 1 — a foundational model designed to answer a question that grows more urgent by the day: when AI agents research, recommend, and execute purchases on behalf of consumers, what should those agents trust?
The model was built from the ground up. In head-to-head benchmarks against the most dominant platforms in product discovery — including Google Shopping and Amazon — Onton’s system outperformed on accuracy across every dimension tested.
The margin was widest precisely where accuracy has proven most elusive: evaluating the veracity of product information in an environment increasingly saturated with synthetic and incentivised content.
The problem Onton addresses is already reshaping commerce. Consider a consumer instructing an AI agent to find the best sofa under $2,000. The agent searches, surfaces thousands of results, and begins sifting through review aggregators, influencer endorsements, and sponsored comparison pages — much of it generated, gamed, or paid for.
Without a mechanism to distinguish signal from engineered noise, the agent recommends anyway. The consumer purchases based on information that was never trustworthy. Nobody registers this as a system failure, because the outcome masquerades as an ordinary bad purchase. But the underlying collapse of information integrity is real, and it is accelerating.
“What was once perceived as an edge case is now increasingly the default state of the internet,” the company noted, framing the degradation of product information quality as a structural shift rather than a passing anomaly.
The modern internet was architected for human evaluation. Reviews, recommendations, forums, and social proof were designed for people — individuals who could apply intuition, contextual awareness, and a healthy dose of skepticism. As AI agents assume responsibility for discovery and purchasing decisions, those human filters vanish. What remains is a system acutely exposed to manipulation, with no native immunity to the synthetic content flooding digital marketplaces.
“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.”
Onton’s response is a model that evaluates not merely what a product claims to be, but whether the information ecosystem surrounding it warrants confidence. By interpreting the attributes and intentions embedded in a user’s preferences — including visual inputs such as images and design cues — Ontology 1 treats those signals as meaningful data rather than potential noise.
The result is a system capable of producing AI-driven purchasing recommendations grounded in information that has been vetted for authenticity, not just surfaced for relevance.
Systemic vulnerabilities
Alongside the product launch, Onton released new research examining how leading AI systems currently evaluate product-related information and where they fail when synthetic content, paid placements, and manipulated reviews enter the pipeline.
The findings are unambiguous: existing discovery systems were never designed to contend with the information environment they are increasingly being asked to navigate. They lack the structural capability to verify, cross-reference, or distrust — functions that humans once performed implicitly and that agents now need engineered into their architecture.
“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.”
That statement captures the strategic bet at the centre of Onton’s launch. In a landscape where capital and talent are flooding into model development, the company is wagering that the decisive competitive moat will not be intelligence alone, but trusted intelligence — a substrate of verified, reliable information upon which autonomous agents can safely operate.
A precondition, not a feature
Onton’s position is that a trust layer for agentic commerce is not an optional enhancement. It is a precondition. Without it, every AI-powered purchasing decision rests on a foundation that can be systematically exploited — by bad actors generating fake reviews, by merchants optimising for algorithms rather than quality, and by platforms whose incentives sometimes diverge from the end consumer’s best interests.
With such a layer in place, agents can begin to deliver on their core promise: helping people make better decisions, faster, with information they can genuinely rely on.
Ontology 1 is available today to users at Onton.com, and on a case-by-case basis for partners building within the agentic web who require a trustworthy foundation for product discovery and recommendation.
The company frames the launch as the opening move in a broader effort to embed authentication infrastructure into the rapidly expanding ecosystem of autonomous commerce — an ecosystem that, for all its momentum, has yet to answer the most basic question of what, and whom, its agents should believe.