An AI shopping agent does not return ten results. It reads a few facts about every product that could answer its shopper's question — name, brand, identifier, price, availability, shipping — and settles on a handful. The rest are not ranked lower; they are ignored. Which products make that shortlist is decided by facts most catalogues are still missing some of.
Which facts does an agent compare?
Shopify's own explainer, published 18 June 2026, describes the mechanism plainly: eligible products are listed in the Shopify catalogue by default, and what that catalogue carries is titles, descriptions, images, pricing, inventory and shipping. Six fields. That is the agent's entire view of your business — no design, no brand, no shopfront. On the same page Shopify reports that AI-driven traffic to its stores grew 8 times year over year, that orders from AI-powered searches rose nearly 13 times, and that new buyers order through AI channels at nearly twice the rate of other channels. Read those with the obvious caveat: they are Shopify's numbers about Shopify stores, and a multiple off a small base is still a small base.
Why is there no page two in agentic shopping?
Agentic shopping does not spread demand the way a results page does. A study of agent purchasing behaviour (arXiv 2508.02630, submitted 4 August 2025, last revised 17 December 2025) found that agents concentrate demand on a handful of products and ignore the rest entirely; that their position biases differ by provider and by model version, and survive even in plain text interfaces; and that how heavily they weigh price, ratings and reviews varies sharply from one model to the next. A model update can redraw the whole distribution. The consequence for a small catalogue is blunt: there is no page two. You are either inside the handful the agent considers or you are absent, and the cheapest way to be absent is a record the agent cannot resolve.
Two prices, one product
Your product page and your product feed are two different databases, and they drift. We check them against each other before anything else on an e-commerce build, because an agent that quotes one price and meets another at checkout cannot finish the job it was sent to do — and a shopper who watches that happen once does not send it back.
What gets a product onto the shortlist?
- Fill every product identifier — brand, barcode or part number, variant.
- Wire the page price and the feed price to one source, not two.
- Keep availability live, not nightly.
- State shipping cost and delivery window per destination.
- Put the return window where a machine can read it, not only in policy prose.
- Mark up each product page with structured data that matches the feed.
How do you know it is working?
Ask the assistants, on a schedule. Once a week, put your customer's actual buying question — with a budget and a deadline in it — to ChatGPT, Gemini and Perplexity, and write down which stores come back and whether yours is one of them. Separately, split assistant referrals out of your analytics so they are not averaged into search. Neither reading is a precise measurement, and the study above is the reason: agents are unstable by design, so a month of absence can be a model update rather than your catalogue. Doing it weekly is what lets you tell those two apart.
Nobody used to write product data for a reader who cannot see the page. That is most of the job now.
What clean product data does not fix
It does not make a weak offer win. An agent that resolves your record perfectly still weighs your price, your delivery window and your ratings against everyone else's, and the study above found that how much each of those counts depends on which model is asked. Anyone selling you a setting that makes agents prefer your store is selling the demo. What clean data buys is narrower and more useful than that: the right to be compared at all.
Your catalogue is the storefront agents see.
Hamed Azimi
Founder, NexusWave
Hamed Azimi is the founder of NexusWave Technologies, a Vancouver-based AI-native digital agency and product studio.
We write these guides from real client work — GEO, SEO, and builds we've shipped. If a piece raised a question about your own site, we're happy to look at it.
- Straight answers, no retainer required
- We reply in your language — EN · فارسی
- Based in Canada, working worldwide
