By Alexander Jordan, Founder
Published: November 5, 2026
Agentic commerce optimization is the work of making sure AI shopping agents can find your products, read them correctly and choose them when a customer asks what to buy. For a Shopify Plus store, most of that work isn't in your ads or on your homepage. It's in your product data.
Ask ChatGPT, Gemini or Perplexity what to buy and you get a short list, sometimes with a buy button right inside the chat. The store that makes that list isn't always the best store. It's the one whose products the engine could read, check and match to what the buyer asked for.
An agent works from three things.
Your product data. Title, price and availability, plus the attributes that describe the product: material, size, colour, dimensions. It reaches the agent through feeds. OpenAI runs its own product feed for ChatGPT, Google uses Merchant Center, and Shopify Catalog sends your catalogue to AI platforms for you.
Your product page, as a bot sees it. That means the raw HTML your server sends, not the page your browser builds. When we fetched Interior Define's sofa page that way, it was an empty container and three scripts. No name, no price, no dimensions. What AI crawlers actually see
What everyone else says about you. Reviews, "best of" lists and your own pages. In Run 8 of our Source Report (22 October 2026), review sites and editorial made up 44% of the sources the engines cited, and brands' own pages 31%. The engines lean on sources they can check, which is why E-E-A-T now works more like a filter than a nudge.
This is where most stores get it wrong. A feed field isn't crawled and ranked like a web page. It's read and filtered. A material field won't help you rank for "oak coffee table". It decides whether you survive when a buyer says "oak, under forty inches, ships flat." If the field is empty, you're not in the answer, however good the table is.
That's also why the name does so little work. In September we asked the four engines "how do I optimize my product data for AI shopping agents". The pages they cited were product data guides about schema, identifiers and feeds, and most of them never used the phrase agentic commerce optimization (Firon citation audit, 25 September 2026). The engines don't care what you call it. They care what's in the fields.
Nobody has to guess what an agent wants, because OpenAI publishes the spec for ChatGPT's product feed. It requires nine fields: item ID, title, description, product page URL, brand, seller name, main image, availability and price. Most Shopify stores pass those by default.
The fields that decide whether you get picked are the optional ones:
Field | What it tells an agent | Who needs it most |
|---|---|---|
Material | What it's made of | Furniture, apparel, bags |
Dimensions | Whether it fits | Furniture, home, luggage |
Size and colour | Which version the buyer wants | Apparel, footwear |
Product category | Which aisle you're in | Every catalogue |
Variant grouping | That five colours are one product, not five | Anything with options |
Source: OpenAI, Products feed reference, checked 27 October 2026.
Two rules from that spec are worth knowing. Give the product's own dimensions, not the shipping box's. And leave a field empty rather than writing "n/a" or "unknown", because a placeholder is not a value.
We ran this check on Floyd, the Detroit modular furniture brand, for the AI Perception Report on our blog. Floyd isn't a client, so here it is as we found it.
Who makes it: the vendor field, which tells a machine who makes each product, names RIZE-Fedex on 48 products and KUKA on 44. Only 5 say Floyd Home.
The price: The Sofa 2.0 carries a price of 0.0 in its product data. OpenAI's spec asks for a positive price.
The structured data: on The Lift Off Coffee Table, it doesn't parse.
Two of the four engines already describe Floyd as modular and built to move. But asked for the best furniture brands for people who move often, they put Floyd fifth, behind Article, West Elm, IKEA and Burrow. The product isn't the problem. The data is.
Pull your catalogue and compare it with the spec, field by field. Note what's present, what's missing and what share of products has it. The required nine will pass. Look hard at the optional ones.
Weight by what you sell. A furniture store lives or dies on material and dimensions, apparel on size, material and fit. Score what matters in your category, not every field equally.
Fill from the product page, never from a guess. A value goes in only if the product's own page says it. Guess from the product type and you end up with a basket tagged for "organization" on a page that never mentions it. At agent scale, a wrong attribute becomes a wrong purchase.
Write values, not copy. "Oak", not "beautifully crafted solid oak". A filter matches words, not adjectives.
Group your variants. One product in five colours, not five products.
Check the vendor, the price and the structured data on your best sellers. That's exactly where Floyd broke.
Put the facts in the page's raw HTML. The name, price and key attributes should be there before any JavaScript runs.
Read it back. After you upload, check the live values on a sample of products. A field that saved is not the same as a field that's right.
It won't move your Google rankings on its own. Feed fields are read by agents and shopping surfaces, not ranked like pages.
It doesn't change what the engines already believe. In Run 8 of our Source Report, 33 of 80 answers came back with no search at all, so the engine answered from what it had already learned. Clean data helps when an agent looks you up. What it learned before that is built over years, by what the rest of the web says about you. That's the GEO half of the work.
Shopify is closing part of the gap itself. Shopify Catalog already sends your data to AI platforms, and Shopify says it cleans and verifies it on the way. What it can't send is a fact your product page never states.
Who this is for: Shopify Plus brands doing $3M+ whose buyers filter on product facts: furniture, apparel, home, outdoor.
The problem: AI shopping agents filter on attributes, and most stores leave the ones that matter empty or wrong. Floyd's vendor field names other companies on 92 products and Floyd Home on 5.
What fixes it: filling the optional attributes from page evidence, in literal values, then checking them live.
Cost: with Firon it's part of month one's technical foundation, not a separate project.
Proof: the Floyd findings above, from our published AI Perception Report.
How to start: book a free AI Perception Report at fironmarketing.com/scan. It includes this check on your catalogue.
It's making your products findable, readable and choosable by AI shopping agents in ChatGPT, Gemini and Perplexity. Most of it is product data: the fields an agent filters on, filled from what your product page actually says.
No. SEO gets a page ranked. Agentic commerce optimization gets a product through a filter. A material field won't help you rank, but without it you drop out the moment a buyer asks for "oak".
Partly. Shopify Catalog sends your catalogue to AI platforms, but it works from the data in your store. If your product page never states a material, there's nothing to send.
It depends on what you sell. Furniture needs material and dimensions. Apparel needs size, colour and material. Everyone needs a correct brand or vendor, a real price and variants grouped under one product.
Alexander Jordan
Founder, Firon Marketing
15+ years scaling brands. A strategic partner for high-growth founders, focused on sustainable revenue models and long-term equity value.
