Onton launches Ontology 1 trust model for agentic commerce
Onton, a San Francisco-based product search and discovery engine, has launched Ontology 1, a trust and authenticity model designed to help AI agents evaluate the credibility of product information before making purchasing recommendations on behalf of consumers. The company says the model was built from scratch and, according to proprietary benchmarks published alongside the announcement, outperformed Google Shopping and Amazon on accuracy across every dimension tested, with the largest reported gains in interpreting the veracity of product information.
The release coincides with a research paper examining how leading AI systems currently handle synthetic content and incentivised recommendations. Onton's conclusion is pointed: existing systems are not equipped for the information environment they are increasingly asked to navigate. Ontology 1 is available immediately to end users via Onton.com and to partners on a case-by-case basis.
The problem Onton is targeting
The company's pitch rests on a structural argument about agentic commerce. When a consumer delegates a purchasing decision to an AI agent, that agent trawls reviews, influencer posts and sponsored comparisons without any reliable mechanism for distinguishing genuine signal from manufactured noise. A bad outcome is attributed to a bad purchase rather than to a failure of the underlying information layer, which means the problem rarely gets surfaced.
"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."
Co-founder Zach Hudson added that the company believes the next major internet platform will be defined not solely by model quality, but by the trustworthiness of the foundation those models operate on. Onton describes a trust layer as a precondition for agentic commerce rather than an optional feature.
Market context and competitive landscape
The agentic commerce layer is attracting significant attention from infrastructure and AI vendors, but trust and provenance verification remains a relatively underserved niche. Most current investment in AI shopping tools focuses on natural-language understanding, personalisation, and inventory integration rather than on the epistemic quality of the underlying data sources.
The benchmark claims in Onton's release are vendor-issued and have not been independently audited. Comparisons against Google Shopping and Amazon are commercially significant given those platforms' dominant positions in product discovery, but buyers and partners should weigh that the benchmarking methodology, dataset selection and evaluation criteria were defined and executed by Onton itself. The company points to a published whitepaper for the underlying methodology, and independent replication would strengthen the case considerably.
More broadly, the problem Onton is addressing sits at the intersection of AI safety and information quality: as agents act autonomously on behalf of users, the consequences of manipulated training signals or incentivised content extend well beyond a mistaken sofa purchase. Regulatory frameworks including the EU AI Act's transparency provisions and the UK's emerging AI liability standards are beginning to address accountability for AI-assisted decisions, though product-discovery contexts remain lightly regulated compared with financial advice or healthcare.
Onton says its model shows signs of being generalisable beyond e-commerce, which would expand the addressable market considerably if that capability holds up in independent testing. The company has not disclosed funding raised, headcount, or revenue figures. It launched publicly in late 2023 and claims it has grown to serve millions of users, though no audited figure was provided.
The next milestones investors and enterprise partners will be watching for are third-party benchmark validation, named commercial partnerships, and evidence that the model's accuracy advantage holds across product categories beyond those tested in the initial release.