It’s clear that at this nascent stage of agentic AI commerce, merchants still hold a lot of power. True, they may have ceded ground to third-party AI agents in the discovery and consideration stages – to the extent that 1 in 5 shopping journeys now start with an AI assistant – but many view it as something that will complement existing channels, rather than replace them wholesale.
Current research underlines that people still prefer to transact on merchant sites, and still value reviews, real-life experience, and social recommendations when it comes to making their minds up.
This kind of content, together with commerce content on publisher sites such as first-look previews, gift guides, comparison articles, and roundups, also forms the backbone of the answers surfaced through Generative Engine Optimisation (GEO).
Indeed, there isn’t much evidence that most people desire to remove too much of the human element from ecommerce. As McKinsey notes, shoppers want to retain control not because AI agents can’t handle the task, but because “human involvement is intrinsic to the value of the experience.”
A clear example of AI tools struggling to complete the full shopping funnel came earlier this year, when OpenAI rolled back its Instant Checkout feature. Promising to allow transactions to be completed automatically within the app, the feature was beset with issues, from the ability to apply coupons and calculate and collect sales tax, through to the accuracy around product pricing and availability.
The latter point is perhaps the most critical to consumers, not least in establishing this as a reliable way of purchasing. But even well before the tool was retired, EMARKETER disclosed that OpenAI’s shopping tools were accurate only 64% of the time on pricing and inventory information (although exactly how much of this is down to the tool’s competence and how much regards the quality of the merchant data it had to work with is unclear).
Where is agentic shopping in 2026?
OpenAI’s dominance means that any missteps will naturally attract a great deal of attention. But the short life of Instant Checkout only tells us so much, and it’s useful to consider this against the backdrop of growing data that shows us where the market stands today and how consumers actually want to use AI tools.
AI platform-driven ecommerce sales are forecast to grow 278% to $20.57bn this year, according to EMARKETER – yet 95% of these sales will still complete on the retailer’s own channel, rather than the AI system used in the process. And while third-party tools, such as ChatGPT and Perplexity, drive some element of around 1.5% of retail sales in the US, this figure is expected to rise to 8.8% by 2029.
Trust is one of the most significant hurdles for AI platforms to overcome – and the data bears this out. A study conducted jointly by ad industry body the IAB and market research agency Talk Shoppe revealed that only 46% of people completely trust recommendations that an AI agent would provide – and 89% of US online shoppers verify AI recommendations via other sources before making a purchase.
While the convenience of using a single platform to perform an entire shopping journey is clear, a December 2025 Bazaarvoice survey showed that more than 67% of shoppers prefer checking out on brand or retailer website, rather than via an AI tool.
What’s also clear is that this preference is not shared across demographics. In fact, Gen Z specifically is even more sceptical on this measure, showing great enthusiasm for using AI to research products and prices, but only 6.2% preferring it to complete a purchase over apps and shopping sites.
How merchants have adopted Agentic AI
Merchants have approached agentic commerce to varying degrees. On their own sites, the most common option has been via AI chatbots, which can advise users on purchases without leaving those environments.
Some more prominent merchants and shopping platforms got involved with Instant Checkout, adopting OpenAI’s Agentic Commerce Protocol (ACP) to facilitate transactions entirely within these environments, while some have opted for Google’s Universal Commerce Protocol (UCP) for an even deeper integration.
Others have taken a different approach. Amazon’s decision to block OpenAI’s crawlers has been widely viewed as a means of protecting ad revenue generated by its own retail media network and the cross promotion of its other services – although, interestingly, it has reportedly bought ads on ChatGPT, effectively tapping into a supply of high-intent shoppers without having to surrender its site to crawlers.
Adding further logic to its decision, the company has also been trialing a ‘Buy For Me’ AI feature that allows users to buy from third-party retailers through Amazon, should it not be able to offer the product itself.
Balancing gains with losses
We’re still in the very early stages of agentic shopping, and consumer confidence remains a key issue.
While trust in AI tools as a whole has a lot of room for progress, research shows that when it comes to recommendations, many more people value the independence of a separate AI platform over one operated by a retailer site, even if (as we’ve seen) that trust is not taken on faith.
Amazon’s stance on AI crawlers points to a separate issue: if other retailers follow Amazon’s lead, these external AI platforms risk showing users an incomplete picture of what’s available to consumers as a whole, further undermining their ability to complete a transaction agentically.
Many other merchants have their own media networks to protect, of course, but their different scale compared to Amazon means they will need to carefully weigh up the value exchange involved in opening themselves up to crawlers.
The obvious gain of being surfaced to shoppers who might not have otherwise visited their site must be balanced with losing the ability to track the customer journey and capture valuable first-party data.
Furthermore, shoppers who transact on an AI platform bypass the carefully curated on-site experience that merchants rely on to upsell products and boost AOV.
Final thoughts
Those who do decide to structure their data for GEO may want to consider what data is actually useful, not just for the AI tool, but for the person behind the query.
Bolstering the basics with experiential information, such as how well a product suits different applications, gives both an AI system and a shopper something richer to work with. This is a space many merchants have already occupied for years, through blogs, first looks, buying guides, and user-generated content.
Given the likelihood of agentic and non-agentic shopping living side by side for the foreseeable future, perhaps the wisest approach is to be mindful of both. Whether machine-readability will meaningfully shift AI discoverability on its own remains to be seen. After all, at least for now, the human element is still very much the one that’s winning the sale.
