A buyer I shipped to last month told me how she found the item: she asked an AI assistant for "a vintage cast iron skillet, 10 inch, restored, under $80, that can ship this week," and my listing was one of three it showed her. She did not search a marketplace. She did not open a browser tab and type keywords. She described what she wanted in a sentence and the assistant went and found it.
That is a different kind of shopper than the one most of us built our listings for. A search engine matches words. An AI shopping assistant reads the whole product page, decides whether the item meets the buyer's description, and either recommends it or moves on. The listings that do well with these assistants are not the ones stuffed with keywords. They are the ones that are specific, structured, and honest.
This post explains how assistants find and judge products, what they need from a page or a feed, and the listing habits that make a small seller's inventory show up. Most of it, you will notice, is just good listing practice with the reasons made explicit.
Two Ways an Assistant Learns About Your Item
AI assistants get product information from two places.
Reading the page
Assistants crawl product pages much as search engines do, but they read more of the page and they lean heavily on the structured product data that sits behind it: the machine-readable block that states the name, brand, price, currency, availability, condition, and images. If that block is present and accurate, the assistant knows exactly what the item is without having to guess from prose. If it is missing, the assistant has to infer, and inference is where listings get skipped.
Your own storefront pages matter here. On APMTSales, every public item page already carries that structured product block, generated from the inventory record: name, brand, price, availability, condition, images, and shipping details where you have set them. You do not have to do anything to get it; you do have to keep the underlying record accurate, which is the subject of most of this post.
Reading a feed
The second path is a product feed: a file or data stream that lists every item with the same fields, delivered directly to the platform rather than crawled. Shopping platforms have used feeds for years (a Facebook and Instagram catalog is one), and AI shopping assistants are adopting the same model, usually through an approval process for merchants and platforms. The fields they ask for are the ones you would expect: a stable item ID, title, description, URL, brand, seller name, image, availability, and price.
Whether an assistant reads your page or your feed, it is reading the same inventory record. Get the record right and both paths work.
What Assistants Are Judging
When a buyer asks for "a 10 inch restored cast iron skillet under $80 that ships this week," the assistant is checking each candidate against each clause. The listing that answers every clause clearly wins. The listing that leaves one ambiguous loses, even if the item would have been perfect.
The things that most often leave an assistant unsure:
- Vague titles. "Skillet" versus "Griswold No. 8 cast iron skillet, 10.5 inch, restored and seasoned." Only one of these answers the question.
- Missing brand. Assistants treat brand as a primary field. A blank brand, or your shop name sitting in the brand field, reads as "unknown maker."
- Unclear condition. "Good vintage condition" is not a condition. New, used, refurbished, and then specifics.
- Stale availability. An item marked in stock that sold last week is the fastest way to get a seller deprioritized. Assistants remember a bad recommendation.
- Price mismatch. The price in the structured data must match the price on the page. If a sale price shows in one place and the regular price in another, the assistant does not know which to trust.
- Descriptions written for keywords instead of readers. Assistants read the description to confirm details. A paragraph of search terms confirms nothing.
The Habits That Make Inventory Findable
None of these are new. They are the habits of a good listing, with the stakes raised because now a machine is reading literally.
1. Write the title as a specification
Brand, product type, size or model, the one or two attributes that matter most, condition if it is not new. In that order, in plain words. The titles post covers the human reasons; the assistant reasons are the same.
2. Fill in the brand field, correctly
If you know the maker, record it in the item's Brand field. If you do not know it, leave the field blank rather than guessing or putting your shop name there. An assistant asked for "a Fiesta pitcher" will find the item whose brand field says Fiesta; it will not find the one that says it in paragraph three of the description. And a wrong brand is worse than none, because the assistant will recommend your item to someone who wanted a different one.
A note from experience: brand names read from a photo of a small label are not reliable, whether a person or software did the reading. Verify against the actual item before you commit the field.
3. State condition in the standard terms, then describe
Start with new, used, or refurbished. Then say what that means for this item, with the flaws named and pointed to photos. The descriptions post gives the structure.
4. Keep availability true to the hour
Mark items sold the moment they sell, everywhere they are listed. If you crosslist from a single inventory record, the record is the source of truth and the listings follow. Nothing damages a seller's standing with an assistant faster than recommending an item that turns out to be gone.
5. One item, one page, one stable address
Each item should have its own page at an address that does not change. Assistants link buyers to that page. If the address changes or the page shows a different item, the buyer lands somewhere wrong and the assistant learns not to trust you.
6. Use real photos in a standard format
A clear main photo in JPEG or PNG, additional photos for details and flaws. Assistants use images to confirm the item matches the description, and some will show the image to the buyer directly. The photo guide applies in full.
7. Describe for a careful reader
Plain text, short paragraphs, specifics. Measurements, materials, what is included, what was tested. No markdown symbols, no walls of adjectives, no keyword lists. A description that would satisfy a careful human buyer satisfies the assistant.
8. Make your shipping and returns findable
"Ships this week" was one of the clauses in that skillet request. If your handling time, shipping cost, and return policy are stated on the page in plain terms, the assistant can answer those clauses. If they are buried or missing, it cannot.
What Not to Bother With
- Stuffing descriptions with phrasings of the same keyword. Assistants read for meaning. Repetition adds nothing and makes the page read worse to humans.
- Writing "for AI" in a special voice. There is no special voice. Clear and specific is the voice.
- Chasing every new assistant individually. They all read the same fields. Fix the record once.
Why This Favors Small Sellers
Large retailers have product data, but it is generic: the manufacturer's title, the manufacturer's description, a stock photo. A reseller's one-of-a-kind item, described specifically by the person who held it, is far richer data. "Griswold No. 8, large block logo, restored, flat on glass, no wobble, ships in two days" answers questions no catalog entry can. When a buyer asks for exactly that, the specific listing wins.
The sellers who have treated listings as a conversation with one careful buyer have been doing this right all along. The assistant is just a very careful buyer who reads every word and never skims.
A Checklist for Your Existing Inventory
Go through your active items, starting with the highest value, and check each one:
- Title names the brand, type, size or model, and key attribute.
- Brand field is filled and verified, or blank.
- Condition starts with new, used, or refurbished and names the flaws.
- Price on the page matches the price in your record.
- Availability is current.
- Main photo is clear and the item has at least one detail shot.
- Description answers what, condition, fit, included, and shipping.
Fix the record, not the individual listings, and let the fix flow everywhere the item appears. If you use a photo-first intake or AI drafting for new items, the same checklist is the review step before you activate them.
The buyer who found my skillet left a five-star review and asked whether I had a lid. I did, and I had never listed it, because I had not gotten around to the description. It is listed now, with the brand field filled in.
