Two announcements this week reveal how fast agentic commerce is moving. On July 23, Google expanded access to Gemini Spark, its agentic AI platform, to all AI Pro subscribers in the United States. The same day, Axios reported that ChatGPT will begin incorporating Yelp reviews, photos, and business information into its local recommendation responses.
These are not incremental updates. They signal that every major AI platform is racing to become the interface through which consumers discover and evaluate products. Google, OpenAI, Amazon, Meta: all of them are building agents that can browse, compare, and recommend.
But there is a gap in their architecture that none of them are talking about. When these agents recommend physical products on Amazon or other marketplaces, they have no way to verify whether the underlying data is trustworthy. They can read reviews, but they cannot tell which reviews are fake. They can compare prices, but they cannot tell if the reference price is fabricated. They can rank products, but they cannot tell if the ranking reflects quality or advertising spend.
The race is on. But nobody brought a trust layer.
What Gemini Spark Actually Does
Google announced Gemini Spark at I/O 2026 as its flagship agentic AI platform. It can plan trips, manage schedules, and complete multi-step tasks across applications. Starting July 23, it is available to millions of US users on the AI Pro tier, with local language support rolling out globally to AI Ultra subscribers.
Gemini Spark is impressive technology. It demonstrates how quickly agent capabilities are maturing. A user can ask Spark to research a topic, compare options, and take action, all within a conversational interface.
But when Spark is asked to recommend a product to buy, it faces the same limitation every AI agent faces. It reads data from the marketplace. The marketplace is manipulated. The agent has no independent verification layer to separate genuine quality from engineered visibility.
Google has an advantage in search infrastructure and product data through Google Shopping. But Google Shopping also relies on merchant-provided feeds and advertising. The conflict of interest is structural: Google profits from ad placements in Shopping, which means its incentives are not aligned with giving users the most honest product recommendation.
The ChatGPT and Yelp Deal: A Pattern, Not a Solution
ChatGPT’s integration with Yelp is revealing. OpenAI recognized that its AI needed third-party data to make trustworthy local recommendations. Relying on its training data or web search alone was not sufficient. By partnering with Yelp, ChatGPT gains access to structured review data, business information, and user-generated photos.
This is the right instinct. Third-party data improves recommendation quality. But the Yelp deal also exposes the scale of the problem.
Yelp covers local businesses: restaurants, services, shops. The review ecosystem for local businesses, while imperfect, is subject to different manipulation pressures than product reviews on Amazon. A restaurant with 500 Yelp reviews is likely to have a reasonably accurate rating because Yelp’s review filters are mature and the review volume relative to customer visits is manageable.
Amazon’s product review ecosystem is fundamentally different. A product can accumulate 10,000 reviews in weeks through incentivized review campaigns. Sellers can ship thousands of units through rebate groups, each generating a “verified purchase” review. The review infrastructure is gameable at a scale that Yelp’s restaurant reviews are not.
The ChatGPT and Yelp partnership shows that OpenAI understands the need for third-party trust data. But it also shows that the gap for physical product recommendations remains unfilled. ChatGPT can now tell you which restaurant to try. It still cannot tell you which headphone brand to trust on Amazon.
Why Every Platform Has the Same Blind Spot
The product trust blind spot exists across every major AI platform for the same structural reason. None of these companies have an incentive to build an independent trust verification layer for physical products.
Amazon’s Project Moonraker, backed by $100 million, is building Alexa into an autonomous shopping assistant. Amazon’s incentive is to drive purchases on Amazon. The company has no reason to highlight fake reviews or manipulated rankings when doing so would undermine consumer confidence in its own marketplace. Amazon’s trust signals, like the Verified Purchase badge, are valuable but insufficient because the platform profits from the volume and velocity that review manipulation enables.
Google’s incentive is to drive ad revenue through Google Shopping. Highlighting product quality issues in sponsored listings would reduce ad spend. Google’s incentive is to make products look findable and purchasable, not to filter them by genuine quality.
OpenAI’s incentive is to provide useful recommendations, but OpenAI does not have a product review database, a price tracking infrastructure, or a review authenticity detection system. The company is building general-purpose AI agents, not commerce-specific trust infrastructure. Partnering with Yelp solves the local business problem. It does not solve the product trust problem.
Meta is experimenting with commerce in messaging. Meta’s incentive is engagement and advertising revenue, not product quality verification.
Every platform has a commercial incentive that conflicts with independent product trust verification. The result is a collective blind spot: agentic commerce capabilities are expanding rapidly, but the trust layer that would make those capabilities reliable for physical products is not being built by any of the major platforms.
What a Real Trust Layer Looks Like
Solving the product trust problem requires infrastructure that no marketplace or AI platform has an incentive to build. It requires an independent layer that sits between the marketplace data and the AI agent, providing verified signals that cannot be gamed by sellers.
Three capabilities are essential:
Review authenticity filtering. The trust layer must analyze reviews for manipulation patterns: textual anomalies, posting velocity spikes, reviewer history gaps, and incentivization signals. Fake reviews must be filtered before any score or rating is computed. Raw review counts and star averages from the marketplace must never be passed through to the agent unfiltered.
Quality-adjusted scoring. A composite score that factors in review authenticity, sentiment depth, seller reputation, price-to-quality ratio, and historical performance. The score must reflect genuine product quality, not visibility engineered through advertising and review farming. Products must be ranked by how well they serve consumers, not by how much sellers spend on manipulation.
Independent product curation. Rather than showing thousands of results ranked by commercial signals, the trust layer must curate a small set of genuinely top-performing products per category. Seven products, not ten thousand. Each earning its position through verified quality metrics over time, not through a temporary review blitz or an aggressive ad campaign.
How GoBuy Fits Into the Agentic Commerce Stack
GoBuy is building this trust layer. It is not a marketplace and it does not sell products. It is an independent verification infrastructure that AI agents can consult before recommending a purchase.
The Smart Score (0-100) is computed from filtered, authenticated review data. Fake reviews are removed before the score is calculated. Products must sustain a score of 80 or higher for 90 days to earn the GoBuy Verified badge, which means short-term manipulation cannot inflate the score. Only the top 7 products per category are surfaced, eliminating the noise of thousands of sponsored and manipulated listings.
Crucially, GoBuy is accessible to AI agents through the Model Context Protocol. Any agent that supports MCP can query GoBuy’s tools at gobuy.ai/api/mcp and receive verified product trust data in real time. The agent does not need to understand review authentication or fake review detection. It calls the MCP tools and gets trustworthy results.
This is the architecture that the agentic commerce ecosystem needs. Not a single platform trying to be both the marketplace and the trust verifier. But separate, independent trust infrastructure that agents consult as part of their decision flow.
The Stakes Are Getting Higher
The Gemini Spark expansion matters because it puts agentic AI in the hands of millions more consumers. The ChatGPT and Yelp partnership matters because it validates the pattern of AI platforms relying on third-party data for recommendations.
Together, they show that agentic commerce is moving from experiment to infrastructure. Consumers will increasingly rely on AI agents to evaluate purchases. Those agents will increasingly make recommendations based on data they process automatically.
If the trust layer is missing, the consequences scale with adoption. A million consumers following AI recommendations that are based on manipulated marketplace data is a million bad purchases. Regulators are already watching. The FTC has been actively pursuing deceptive marketing cases, including the July 22 settlement with Trend Deploy returning $672,000 to consumers harmed by deceptive practices.
The platforms building agentic commerce capabilities need to recognize that model capability is not a substitute for data integrity. A more powerful model processing manipulated data produces more confident wrong answers. The investment in agent capabilities must be matched by investment in trustworthy data sources.
The Opportunity for Honest Sellers
The trust layer is not just about protecting consumers. It is about rebalancing the market for honest sellers.
Under the current system, sellers who invest in product quality but do not engage in review manipulation or aggressive advertising are invisible. Their products are buried under competitors with bigger review farming budgets. The marketplace rewards manipulation over merit.
An independent trust layer changes this dynamic. When AI agents evaluate products based on filtered, authentic review data, honest sellers with genuinely good products become discoverable. The market rewards quality over manipulation budget.
This is the long-term promise of agentic commerce. Not just automation of the shopping process, but a fundamental rebalancing of how products are discovered and evaluated. AI agents that can see through manipulation and recommend based on genuine quality create a market where making good products is the winning strategy.
That future requires trust infrastructure that no major platform has an incentive to build. That is the gap GoBuy fills.
Connect your AI agent to GoBuy’s MCP server at gobuy.ai/api/mcp. Integration documentation is available at gobuy.ai/agent-docs.