---
title: AI Shopping Engine Onton Launches Ontology 1 as a Trust Layer for Agentic Commerce
description: Onton's Ontology 1 scores whether product information can be trusted as AI agents take over shopping. Company benchmarks claim wins over Google and Amazon.
author: Darie Nani (Editor-in-Chief)
updated: 2026-07-29T08:24:43.274Z
canonical: https://www.sovereignmagazine.com/article/onton-ontology-1-trust-agentic-commerce
image: https://cdn.nanimediahouse.com/onton-ontology1-search.webp
categories: Artificial Intelligence, Startups
content_type: News
region: San Francisco
publication: Sovereign Magazine
about:
  - type: Organization
    name: Onton
    description: Onton is a next-generation product search engine with an image generation model that lets you visualize and redesign your home in seconds. With our AI interior designer, Imagine, you can create beautiful, high-quality renderings that perfectly match your vision. Try it now at onton.com
    url: https://onton.com/
    industry: AI Product Discovery
    sameAs:
      - https://www.instagram.com/onton.ai/
      - https://www.linkedin.com/company/onton-ai/
      - https://www.tiktok.com/@onton.ai
---

Onton, a San Francisco search and discovery engine for products, has launched Ontology 1, a model designed to judge whether the product information AI agents act on can be trusted, as more consumer purchases are researched and executed by software rather than by people.

The model is available from today to shoppers at [Onton](https://onton.com) and on a case-by-case basis to partners building agent-driven product discovery. Onton says it built the model from scratch and that in head-to-head [benchmarks](https://onton.com/research/ontology-1-benchmarks) against the largest players in the field, including Google Shopping and Amazon, it outperformed them on accuracy across every dimension tested, with the widest margin on interpreting the veracity of product information. The benchmarks and the research behind them are the company's own.

## Agentic Commerce Rollouts Across ChatGPT, Amazon and Google

The infrastructure for agentic commerce has been rolled out over the past year. Stripe and OpenAI published the Agentic Commerce Protocol in September 2025, alongside Instant Checkout in ChatGPT. PayPal added support for the protocol in October. Microsoft launched Copilot checkout in the United States in January 2026 with Shopify, PayPal, Stripe and Etsy. Amazon renamed its Rufus assistant Alexa for Shopping in May and has added agentic features including automatic purchase once an item hits a target price, saying the assistant helped more than 300 million customers in 2025. Google's AI Mode carries an agentic checkout and a universal cart, and Perplexity runs shopping across more than 5,000 merchants with PayPal handling payment.

## AI Shopping Assistants Inherit a Product Discovery Layer Made for People

In the company's own example, a shopper asks an agent for the best sofa under $2,000. The agent returns thousands of results and works through review aggregations, influencer recommendations and sponsored comparisons, an unknown share of which are synthetic, incentivized or manipulated. It has no dependable method of telling those apart from the rest, and it recommends anyway. Nothing in the sequence records a failure, because the outcome looks like an ordinary bad purchase.

The review and social proof layer of the internet was assembled for human readers who bring context and skepticism to it, and an agent working through the same material at volume applies neither of those, which leaves the recommendation open to manipulation by anyone with an interest in the sale.

"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."

## Fake Reviews and the Limits of the FTC Consumer Review Rule

Fake reviews are already regulated in the United States. The Federal Trade Commission's Consumer Review Rule took effect in October 2024 and bans creating, buying or selling fake reviews, including those generated by AI, with civil penalties of up to $53,088 per violation. In December 2025 the commission sent warning letters to 10 companies over potential breaches. The rule sets penalties for publishing fake reviews without giving a system that reads a product page any means of identifying them.

## Ontology 1 and the Case for Trustworthy AI in Product Recommendations

Ontology 1 assesses the information surrounding a product as well as the product itself, including whether that information is incentivized. It also reads the attributes and intent behind a shopper's stated preferences, visual inputs among them, and treats those as data rather than as noise. Onton says the model learns on its own and shows signs of generalizing beyond e-commerce, and it is publishing research alongside the launch on how leading AI systems currently evaluate product information and where they fail once synthetic and incentivized content enters the process. The company's conclusion is that those systems are not equipped for the information environment they are being asked to work in.

"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's Move From Consumer Search to the Agentic Web

Onton, formerly known as Deft, was founded by Hudson, its chief executive, and Gunnarson, after Gunnarson spent 30 hours hunting for a specific mid-century gray couch. Its technology is neurosymbolic, pairing neural search with symbolic reasoning, and the consumer product supports natural-language and image search, a mood-board tool called Canvas and an interior design feature called Imagine. It started in furniture and home decor and has been expanding into apparel and consumer electronics. The company raised a $7.5 million seed round in November 2025 led by Footwork, with Liquid 2 and Parable Ventures taking part, bringing its total raised to around $10 million. It employs about 10 people and says it has guided millions of users since its public launch in late 2023.

Sovereign covered Onton in February, when the case it made to shoppers was speed: [cutting an average 79-day shopping journey to under a day](https://www.sovereignmagazine.com/article/onton-s-ai-cuts-shopping-time-from-79-days-to-just-one).

## FAQ

**Q: What is agentic commerce?**
Commerce in which an AI system does the work a shopper would otherwise do: researching options, comparing them, and in a growing number of cases placing the order and paying. It sits on top of protocols such as the Agentic Commerce Protocol, published by Stripe and OpenAI, which let an assistant complete a checkout with a merchant without the shopper visiting the merchant's site. Onton's position is that the quality of the information those agents read has not been addressed alongside the payment infrastructure.

**Q: What companies are using agentic commerce?**
OpenAI runs Instant Checkout in ChatGPT, Amazon sells through Alexa for Shopping, Google offers an agentic checkout and universal cart in AI Mode, Microsoft launched Copilot checkout in the United States in January 2026 with Shopify, PayPal, Stripe and Etsy, and Perplexity covers more than 5,000 merchants. PayPal supports the Agentic Commerce Protocol. Onton is selling to the companies building on this layer rather than competing with them on checkout.

**Q: Is it illegal to write fake reviews online?**
In the United States, yes, under the Federal Trade Commission's Consumer Review Rule, which took effect in October 2024. It bans creating, buying and selling fake reviews, covers reviews generated by AI, and carries civil penalties of up to $53,088 per violation. The commission sent warning letters to 10 companies in December 2025. Enforcement acts on publishers of the content and does not help an AI agent identify a manipulated review while it is reading one.

**Q: Can AI be trusted to be accurate?**
Accuracy depends on the source material as much as on the model. An agent that reasons correctly over synthetic or paid-for product information still produces a bad recommendation, and as the sofa example above shows, nothing in the process marks it as an error. This is the case Onton makes for a separate model that scores the trustworthiness of the information itself, and the company's benchmark claims against Google Shopping and Amazon are its own rather than independently tested.

**About Onton**

Onton is a next-generation product search engine with an image generation model that lets you visualize and redesign your home in seconds. With our AI interior designer, Imagine, you can create beautiful, high-quality renderings that perfectly match your vision. Try it now at onton.com

[Website](https://onton.com/)