It’s been a busy month for commerce teams. Between Google’s Universal Commerce Protocol (UCP) announcement and OpenAI beginning to test ads and checkout inside ChatGPT, it’s clear that AI-driven discovery is moving fast.
At Productsup, we work closely with global brands and retailers on product data, feeds, and integrations across emerging discovery channels, which gives us a front-row seat to these shifts.
To make sense of what these changes actually mean for brands and retailers, we sat down with Marcel Hollerbach, CIO at Productsup, to get his perspective.
Let’s get started.
1. Right after Google’s UCP announcement, we’re now seeing OpenAI test ads inside ChatGPT. How do you read this moment?
Marcel: What stands out to me is the timing. These announcements aren’t isolated. They’re all pointing in the same direction: commerce is moving away from pages and platforms and toward systems and agents that make decisions on the user’s behalf. Google is rebuilding its commerce infrastructure around that idea, and OpenAI is experimenting with how discovery and purchase can happen inside a conversation. Different approaches, same shift (and the timing is hard to ignore).
2. If ads appear inside conversational interfaces like ChatGPT, what determines whether a product shows up at all?
Marcel: Before anything else, the product has to be eligible in the AI’s reasoning process (which most brands don’t think about yet). That means accurate attributes, clear categorization, and up-to-date information. In traditional commerce, you could sometimes compensate for weak data by spending. In AI-driven discovery, that doesn’t work. If the system can’t confidently recommend your product, it simply won’t surface it.
3. With a reported 4% transaction fee for Shopify merchants using ChatGPT Instant Checkout, is this a reasonable trade-off for AI-driven demand, or the start of a much bigger conversation around margins?
Marcel: The percentage itself matters less than what it represents. That 4% tells us OpenAI isn’t just facilitating discovery; it wants to participate in the transaction. Whether it’s “fair” will depend on performance, but structurally, this is very different from paying for impressions or clicks. Also, looking at the transaction fees that marketplaces like Amazon charge, this looks fairly priced.
4. Some people are framing this as “ChatGPT becoming another marketplace.” Do you agree with that?
Marcel: I think it becomes increasingly hard to put labels on things like this. Everything blends like retailers becoming marketplaces and also adnetworks with retail media offerings. Social Media apps (e.g., TikTok) have become shops. Search Engines are becoming LLMs (e.g., Google AI Mode), etc. So I’d call ChatGPT a Search Engine, an LLM, an Adnetwork and a Marketplace.
5. If a brand asked you today, “Are we ready for ChatGPT ads?”, what’s the first thing you’d want to check?
Marcel: I’d start with their product data. Not their ad setup, not their budget. The question is whether their products are described clearly, consistently, and completely enough for an AI system to understand them. If the system can’t reason about your products, ads won’t fix that. Readiness starts with data, not media.
6. For commerce teams watching these announcements, what’s one action they should take in the next three to six months?
Marcel: They should audit how “AI-readable” their product data really is. That means looking beyond channel requirements and asking whether the data explains the product well enough for a machine to make a decision. E.g., consumers are not searching for a “running shoe” anymore, but they are asking, “What is the best running shoe for rainy conditions?” The product data needs to reflect this. Teams that treat feeds as a strategic asset rather than an operational task will be in a much stronger position as these models mature.
And with Productsup already supporting integrations with ChatGPT and Perplexity, we’re confident that commerce teams can operationalize this shift across emerging AI-driven discovery surfaces as they evolve.
7. Last one: Beyond work, where does AI actually show up in your everyday life?
Marcel: Mostly to save time. I use it to get summaries of books I’ve been meaning to read but never quite get around to, and for some design work like Christmas cards, invitations, and other small creative things 😉
What’s next?
That’s the honest answer; we don’t fully know yet (and that’s okay). Things are moving fast, and a lot of this is still being tested in real time.
We’ll keep watching how this space evolves. And as always, there’s more to come!
About the author
