The Platforms Now Describe Your Products For You
Google, Meta, Amazon and Shopify use your catalog to describe your products for you. What they can use depends on one step many stores skip: interpretation.
If you run a store, your catalog already travels further than your website. It goes into Google Merchant Center. It feeds your Meta catalog ads. If you're on Shopify, it can reach ChatGPT, Copilot and Gemini through Shopify Catalog. If you sell on Amazon, its shopping assistant, Alexa for Shopping (formerly Rufus), draws on your listing when shoppers ask it questions.
These systems used to display your data. Increasingly, they interpret it. They decide what your product is, which shopper it suits, which question it answers, and which ad to build from it. Then they describe it, in their own words, to people you'll never talk to.
And the product is only the first thing they have to get right. Before a system can recommend you to anyone, it has to understand three things about your store: what you sell, who stands behind it, and under what rules it can be bought.
So the useful question for a marketer today isn't only whether your products show up. It's what these systems had to work with, and whether you'd sign off on the description they produce. Being visible to AI isn't the same as being understood by it.
The platforms are telling you what they need
You don't have to guess what matters. The platforms have started saying it out loud.
Google asked retailers in September 2026 to send conversational attributes: questions and answers about each product, how products relate to each other, and variant options, so that its AI can understand the nuances of what you sell. In a test with lululemon, details supplied by the brand showed up in AI Mode's product recommendations. The same announcement opened a beta that lets people ask product questions inside YouTube ads.
Shopify tells merchants to group real variants under one product, to use a precise category ("men's insulated winter boots", not "footwear"), to describe products in literal language rather than marketing copy, and to map custom fields so its Catalog can pass them to AI channels. Its own warning: without clean product data, AI platforms may still find you, but they might not find the most accurate information about you.
Meta builds Advantage+ catalog ads directly from your catalog, choosing products and assembling the ad for each person.
Amazon's assistant answers shoppers using your listing together with reviews, customer questions and other sources. The listing is the part of that picture you control.
Four companies, four different systems, and they aren't the only ones. Underneath, one request: tell us clearly what this product is, in fields we can use.
One candle, with and without interpretation
Here is what that request means on a single product.
A store sells a scented candle. The title says Scented Candle No. 4. The category is Candles. The description reads:
A woody blend that opens with cedar, settles into a hint of smoke, and dries down to vanilla and labdanum. Light it in the evening to unwind. Comes gift-boxed.
The rest of what a shopper wants to know is somewhere else on the site. The About page says the candles are hand-poured in small batches in Portland, Oregon, with soy wax from U.S. farms. The shipping page says orders leave in two business days and take two to three more to arrive. The FAQ says returns are accepted within 30 days, unopened.
A person who clicks around knows what they're buying, from whom, and on what terms. A feed that only carries the basic fields moves the title, the category and the paragraph. The rest stays where it was written. Here is that feed next to the same store after interpretation:
| What the platform needs to know | What the store says, and where | Basic feed, without interpretation | With interpretation | |
|---|---|---|---|---|
| Product | What is it? | Title and category: Scented Candle No. 4 · Candles | Category: Candles | Home Fragrance › Scented Candle |
| What does it smell like? | Description: "A woody blend…" | — (empty) | Scent family: Woody | |
| How does the scent develop? | Description: "…opens with cedar, settles into a hint of smoke, dries down to vanilla and labdanum" | — (empty) | Top: Cedar · Heart: Smoke · Base: Vanilla, Labdanum | |
| When would someone use it? | Description: "Light it in the evening to unwind." | — (empty) | Evening · Relaxation | |
| Is it ready to give? | Description: "Comes gift-boxed." | — (empty) | Gift-boxed: Yes | |
| Brand | Who makes it, and where? | About page: "hand-poured in small batches in Portland, Oregon, with soy wax from U.S. farms" | — (stays on the About page) | Made in: USA (Portland, Oregon) · Hand-poured, small batch · Wax: soy, from U.S. farms |
| Transaction | Will it get here by Friday? | Shipping page: "orders leave in two business days and take two to three more to arrive" | — (stays on the shipping page) | Processing: 2 business days · Transit after dispatch: 2–3 business days |
| Can it be returned? | FAQ: "returns accepted within 30 days, unopened" | — (stays in the FAQ) | Returns: 30 days, unopened |
A shopper asks for "a woody candle for a gift." In the basic feed, the scent and the gift box stay inside a paragraph. A filter for woody candles has nothing to filter on, a gift category has nothing to pick up, and an integration that reads fields finds them empty. After interpretation, both answers sit in explicit fields that can be mapped to each platform's requirements.
Then she narrows it: "made in the USA, and I need it by Friday." The store has what it takes to answer both. One answer is on the About page; the pieces of the other are on the shipping page, and nothing connects any of it to the candle. After interpretation, it travels with it.
Nothing in the last column was added that the store hadn't already said. "Opens with", "settles into" and "dries down to" are how perfumers describe top, heart and base notes. A person hears that without thinking. Interpretation is the step that recognizes it, works out what each phrase means, and puts each piece where a machine can use it.
Multiply that by a catalog of five hundred products and every platform you sell on, and you have the work that's still ahead of most catalogs.
The work has a name
We call it Interpretability Engine Optimization (IEO): the practice of making source information machine-interpretable for a declared decision context before downstream reasoning, recommendation, comparison, decision, or action occurs.
The "decision context" part matters. A candle's scent family is essential when someone asks for a woody candle and irrelevant when they ask whether it will arrive by Friday. For that question, the shipping rule is the fact that counts. IEO starts from the question a system has to answer and works back to what it needs to know.
A machine can read a paragraph without understanding it. Reading pulls the words out. Interpretation turns them into meaning a system can act on: this is a candle, it smells woody, it comes ready to give.
That's why IEO comes before the work you already do. SEO optimizes for clicks. Answer engines need to extract an answer; generative engines need to trust a source enough to cite it. IEO optimizes for understanding, the thing all of them depend on. You can rank first for a product the machine never understood. You can even be recommended for something your product doesn't do: described as waterproof when your page says water-resistant.
The rule that keeps it honest is simple: every value has to trace back to something the business actually published. Where the source says nothing, the field stays empty. A catalog filled with confident guesses is worse than one with honest gaps, because the guesses get repeated to customers as facts.
What a machine has to understand
Go back to the candle table. It has three parts, and they are the three things IEO works on.
The product: what is being sold. What it is, what it smells like, when to use it.
The brand: who stands behind it. Who makes it, where, and with what.
The transaction: under what rules it can be bought. Whether it gets there on time and whether it can go back.
Stores tend to work on the first and leave the other two on pages a feed doesn't carry. A recommendation needs all three.
Product tells the machine what is being sold. Brand tells it who stands behind it. Transaction tells it under what rules it can be bought. IEO makes all three interpretable.
What's next in this series
The next nine articles go in three rounds. Each round takes one problem from each of the three: a product, a brand, a transaction. Every article starts with something that goes wrong for a store and shows the part of IEO that fixes it.
Round one
- Product: a shopper asked Google about your product, and the answer was wrong.
- Brand: your About page says who you are, and no machine can use it.
- Transaction: the product page is perfect, and the sale fails at shipping and returns.
Round two
- Product: Meta built your ad, and your best selling point stayed in a paragraph.
- Brand: "organic", "made in Italy", "doctor-founded": which of your claims a machine can trust.
- Transaction: free delivery, and no machine could say whether they'd carry it up three flights.
Round three
- Product: shoppers search your store for things you sell, and find nothing.
- Brand: your brand has three names online, and a machine sees three companies.
- Transaction: "save 15%" was on the page, and nothing said when a customer could cancel.
To close: the method behind all of it, for agencies and practitioners.
Check what's explicit, what's missing and what contradicts itself on one of your products → Run the free audit
To go deeper → The founding paper, Interpretability Engine Optimization (doi.org/10.5281/zenodo.22104280), and the Methodology (doi.org/10.5281/zenodo.22416784). Start at nsolvia.com/ieo.
Sources: Google, "Retailers: Get ready for AI-driven holiday shopping," Sept 16, 2026 · Shopify, "Agentic-Ready Product Data"; Shopify Help Center, "Shopify Catalog and product discovery for agentic storefronts" · Meta for Business, "Advantage+ Catalog Ads" · Amazon, Alexa for Shopping (formerly Rufus).
← Previous · ⌂ IEO · Next: A Shopper Asked Google About Your Product. The Answer Was Wrong. →
Series: Interpretability Engine Optimization · Knowledge Domain: Interpretability (IEO)
Interpretability Engine Optimization (IEO)
The UNDERSTAND Layer
nsolvia.com/ieo