A growing part of online shopping originates from a request made to an assistant: "find a coffee maker that accepts reusable capsules, delivered in Porto Alegre and costs up to R$ 400". The responder is no longer a search engine — it is an agent that reads catalogs, compares alternatives, and in some cases, executes the transaction. This phenomenon has a name: agentic commerce.
The practical consequence is direct. The agent does not "see" banners, does not get enchanted by storytelling, and does not wait for a slow page to load. It evaluates data. If the information about your product is not machine-readable, the store is excluded from the list before any human decision.
How the agent chooses a store
The typical flow has four stages: the agent understands the request, searches for candidates, validates whether the product meets restrictions (price, stock, delivery time, region), and only then decides. Each stage requires a specific type of data.
- Discovery: complete product feed, with descriptive title and brand.
- Validation: final price, availability, delivery time, and service area in text.
- Comparison: objective attributes — material, dimensions, compatibility, warranty.
- Execution: simple checkout, without mandatory registration and with predictable steps.
The five technical requirements that decide the choice
1. Complete structured data for each product
Product markup with price, currency, availability, condition, rating, and identifiers (GTIN, MPN, SKU). An empty field is a non-comparable field; a non-comparable product tends to be discarded.
2. Title that describes, not that seduces
"Unmissable super promotion" says nothing to a language model. "Electric coffee maker 1.2 L, stainless steel, reusable capsule, 220 V" responds to the request's restrictions. Write the title thinking about the filter, not the click.
3. Attributes in text, not in images
Dimensions, composition, and compatibility need to be in HTML. Information within a technical sheet in an image or PDF is invisible to the agent — and becomes a reason for exclusion.
4. Clear and unambiguous policy
Delivery time by region, return policy, free shipping from what amount, payment methods. Ambiguity costs sales: the agent prefers the competitor with clear rules.
5. Checkout that does not break automation
Predictable flow, labeled fields, no surprise validations, and immediate email confirmation. Each exotic step reduces the chance of completion.
What changes in commercial strategy
If the decision migrates to data, the competitive advantage migrates along with it. Three movements become more valuable than aggressive discounts:
- Attribute coverage: the catalog with the most complete sheet appears more often.
- Consistency across channels: price and stock equal on the website, marketplace, and feed.
- Verifiable reputation: real reviews and responses to complaints count as a sign of trust.
Errors that take the store out of the competition
- Price only visible after login or in JavaScript loaded on the client.
- Technical sheet in image, without useful alternative text.
- Outdated stock, leading to cancellations and reputation loss.
- Generic product names, repeated among thousands of items.
- Slow page, which expires the agent's time before reading.
Read also:
Where to start this week
Choose the 50 products that generate the most revenue, audit the sheet of each one, and fill in what is missing — attributes, dimensions, warranty, and delivery policy in text. Review the feed next and test the checkout in anonymous mode, without browser history. This is not a months-long project: it is catalog sanitation, and the return appears in the first queries mediated by AI.
Agentic commerce does not eliminate marketing; it changes the target. Continuing to attract human eyes is necessary; being readable for those who execute the purchase has become mandatory.