ChatGPT recommends products by reading a structured merchant feed, not by browsing your store — and OpenAI’s own commerce spec names the ranking signals it wants, including a popularity_score and a return_rate. So the real question isn’t whether ChatGPT can recommend you. It’s whether your products are in that feed and whether the fields OpenAI reads say you’re worth recommending.
How does ChatGPT decide which products to recommend?
Not by crawling your storefront like a shopper. When ChatGPT suggests products, it’s drawing on a structured product feed, and OpenAI has published the exact shape that feed takes. The Agentic Commerce product feed spec lays out the fields a merchant is expected to supply — the basics like title, price, and availability, plus explicit ranking inputs. Two of those inputs matter more than merchants realize: a popularity_score (a 0–5 measure of how well a product performs) and a return_rate. These aren’t scraped or inferred. OpenAI asks you to provide them.
That’s the whole game in one sentence: ChatGPT ranks the feed OpenAI ingests, and some of the strongest ranking fields are values you fill in yourself.
What signals actually move the ranking?
The spec makes the levers concrete:
popularity_score(0–5) — a direct signal of how well the product sells or performs. A product with a strong score is a safer thing for ChatGPT to put in front of a buyer.return_rate— a low return rate tells the model this product satisfies the people who buy it. A high one is a reason to hold back.- Reviews and ratings — social proof the model can lean on when it’s choosing between comparable products.
- Availability and price — recommending something that’s out of stock or mispriced is a bad experience, so stale feed data quietly kills recommendations.
Most of these are things you control at the feed level. If your popularity_score is empty and your return_rate is blank, you’ve handed the model no reason to prefer you over a competitor who filled those fields in.
If I’m on Shopify, am I already in?
Being in the feed and being recommended aren’t the same thing. The demand is real and moving fast — Shopify reports AI-referral sessions up roughly 8x year-over-year and AI-referred orders up about 13x in Q1 2026 (Shopify). That’s the traffic that flows to stores AI engines actually surface. But presence in a feed is table stakes; the ranking fields decide who gets named. A product that’s technically discoverable but has weak or missing signal fields is present and invisible at the same time. And if a competitor keeps getting named instead of you, that has its own diagnosis — see why ChatGPT recommends competitors and not you.
How do I find out where I actually stand?
You can read the feed spec yourself and audit your feed field by field — check that popularity_score and return_rate are populated with real numbers, that reviews are flowing, and that availability and price are current. That’s the honest DIY answer, and for a small catalog it’s doable.
We built AI Visibility to do it at scale and keep doing it. It scores your live product feed against every signal OpenAI reads — the named spec fields plus reviews and structured data — and shows you, product by product, which signals are strong, which are empty, and which are actively hurting you. Instead of guessing why ChatGPT names a competitor, you see the exact fields the model uses and where your feed comes up short. Fix those, and “does ChatGPT recommend my products” stops being a question you have to guess at.