As ‘agentic commerce’ becomes more prevalent, a leading industry expert cautions that businesses should not overly rely on ‘GEO’

Welcome to Eye on AI

In this issue, we cover several major developments in the AI world: Google has introduced a feature that lets users make purchases directly through Google Search’s AI Mode and the Gemini chatbot. Apple has chosen Google technology to enhance Siri, while Meta has unveiled a new team focused on AI infrastructure. Meanwhile, researchers are leveraging AI to discover innovative gene-editing techniques.

The past week was filled with significant AI announcements. One highlight is Google’s rollout of an e-commerce checkout option within its AI-powered search and Gemini app. Walmart is among the first major retailers to adopt this new capability, marking a notable shift in online shopping. This AI-driven checkout is built on a new “Universal Commerce Protocol,” designed to simplify the integration of agentic AI sales for retailers. Additionally, Google Cloud revealed a suite of AI tools to support agentic commerce, including the Gemini Enterprise for Customer Experience, which merges shopping and customer service. Home Depot is one of the initial companies to utilize this new cloud solution.

The Rise of Agentic Commerce and Generative Optimization

Although agentic commerce is still in its early stages, businesses are already concerned about how to ensure their products are recommended by AI agents. A new sector has emerged, offering services known as “generative engine optimization” (GEO) or “generative-AI optimization” (GAIO). While these services draw on traditional search optimization tactics, there are important differences. For now, GEO appears to be less susceptible to manipulation than SEO. AI agents and chatbots seem to prioritize products that have received favorable coverage from respected news sources and high ratings on trusted review platforms, which benefits both consumers and media outlets.

However, the shift to AI-driven commerce introduces significant governance challenges that many organizations may not fully grasp. Tim de Rosen, founder of AIVO Standard, a company specializing in generative AI optimization and AI information tracking, warns that companies need to pay close attention to how AI agents gather and use information.

AI’s Inconsistencies in Financial and Governance Insights

De Rosen explained in a recent conversation that while AI models generally provide accurate descriptions of a brand’s products and features—often citing their sources—they struggle with questions about a company’s financial health, governance, and technical certifications. These inconsistencies can have a major impact on procurement decisions.

For example, AIVO Standard evaluated how advanced AI models responded to questions about Ramp, a rapidly growing business expense management platform. The findings showed that AI models often failed to provide reliable information about Ramp’s cybersecurity credentials and governance practices. This lack of accuracy could unintentionally steer organizations toward established, publicly traded companies, even when newer private firms meet the same standards, simply because the AI cannot verify or cite information about the latter.

In another case, AIVO Standard reviewed AI-generated responses about the risks of competing weight loss medications. The AI models not only listed potential risks but also began to make recommendations about which drug might be safer. Although the responses were generally factual and included disclaimers, they still influenced perceptions of risk and eligibility, according to de Rosen.

These issues were observed across leading AI models and persisted regardless of the prompts used. In some instances, the models would even reinforce incorrect information, insisting on its accuracy.

Generative Optimization Remains Unpredictable

There are several takeaways. First, companies offering GEO services may not be able to consistently influence how their brand appears in AI-generated responses, especially for information beyond product details. Results can vary widely, and there is little evidence on how to reliably guide AI responses for non-product topics.

More importantly, as agentic workflows become more common—even with human oversight—AI-generated information is increasingly forming the basis for critical decisions. De Rosen points out that most organizations lack systems to track which prompts were used, what responses were generated, and how these influenced final recommendations or choices. In regulated sectors like finance and healthcare, this lack of transparency could lead to serious problems if regulators demand detailed records and companies are unable to provide them.

Stay tuned for more updates on the evolving AI landscape.