Agentic commerce
Agentic commerce is a purchase carried out by an AI agent on a person's behalf. The agent finds the product, compares offers against the stated conditions and, in its more advanced form, completes the order. For an online shop that changes something fundamental: the outcome is decided not by the design of the site or the campaign, but by the quality and machine readability of the product data.
In short
| What it is | Search, comparison and possibly checkout performed by an AI agent instead of the user |
| How mature it is | Search and comparison work today, automated payment is still limited and differs by platform |
| Who is affected first | Categories with clear parameters: electronics, spare parts, consumables, travel, insurance |
| What decides | Completeness and accuracy of product data, feed availability, consistent price and stock across sources |
| What loses weight | Visual design, banner creative and part of classic checkout UX work |
How an agentic purchase runs
The user states an intent, not a query. Instead of "men's running shoes size 10" they write "I need running shoes for asphalt, size 10, under 150 euros, delivered by Friday". The agent breaks the intent into conditions, retrieves offers, compares them and returns a shortlist or a direct recommendation.
The data layer decides, not the page
An agent does not read a product page the way a person does. It takes structured data, the feed, and whatever it finds about the product on comparison sites and marketplaces. When the data is missing a material, a dimension or a delivery window, the product drops out of the shortlist even if it is objectively the best option.
Inconsistent data costs more than it used to
A price that differs between the site, the feed and a marketplace used to be an operational annoyance. For an agent it is a reason to discard the offer, because it cannot verify it.
The brand does not disappear from the decision
Agents work with reviews, ratings and how the brand is written about. A brand with no trace in the data carries less weight than one with a consistent public record.
From our own practice: conversational attributes in Merchant Center
The closest practical step we already take in client accounts is conversational attributes in Google Merchant Center. These extend product data with the properties people ask about in sentences rather than parameters: what the product is for, what it suits, what the package contains.
They are filled in as sentences, and that sentence is what the model matches the customer's question against. On large catalogs the attributes cannot be written by hand: they are generated by rules over the feed and templates per category, otherwise the work never finishes.
The second thing we see in day-to-day operations: on accounts where we manage large catalogs, the most frequent reason a product fails to show is a missing or inconsistent attribute, not a low bid. The pattern holds across clients. Agentic commerce will only make that problem larger.
Common mistakes
- Treating it as a new ad placement. It is not a format, it is data preparation. Budget does not solve it.
- Leaving the feed at the minimum. Required attributes are enough to appear, not enough to be selected by an agent.
- Ignoring marketplaces and comparison sites. The agent verifies your offer there and contradictions put it off.
- Waiting for it to become a standard. Good product data pays for itself in today's channels already.
What to do now
- Extend the feed beyond the required minimum. Material, dimensions, compatibility, delivery time, package contents.
- Align price and stock across the site, the feed, comparison sites and marketplaces.
- Add conversational attributes wherever the platform supports them.
- Keep product data reachable. Firewall-level crawler blocks affect this layer too.
- Track the share of products with incomplete data. It is a simple metric and it predicts losses directly.
Related terms
See also ChatGPT Shopping, product feed, AI Mode, RAG and AI visibility.
Frequently asked questions
Do agents actually buy on their own today?
Not at scale yet. They already search and shortlist routinely, and that is the stage which decides whether you make the cut at all.
Does this mean the end of PPC?
No. It means part of the demand moves into an interface where the auction bid matters less than the data. Paid channels stay, they just lose the easiest part of the journey.
How is it measured?
Indirectly for now: the share of products with complete data, sessions from assistant domains, and manual checks of whether your products appear in answers to model queries.
Is it worth it for a smaller shop?
Yes, because data work pays off outside agents too. A better feed means better performance in Shopping, on comparison sites and on marketplaces.
How we can help
Product data, feeds and marketplace visibility are daily work for us. Details are on the E-commerce marketing agency page, and conversational attributes are covered in our article Conversational attributes in Merchant Center.