AI shopping is no longer a demo. ChatGPT has product cards with live prices, Shopify's catalog syncs into AI storefronts automatically, and Meta's Muse now lets US users check out via Shop Pay inside a conversation.
So I spent the last few weeks digging into one question: when a user asks "help me buy X," how does the AI pick which store to recommend? Here's what I found, with evidence strength marked honestly ([Strong] = vendor/official, [Moderate] = multi-source, [Weak] = single claim).
1. Structured product feeds carry the most weight
Roughly 65% of ChatGPT shopping recommendations come from structured feeds [Moderate]. Shopify and Etsy merchants get auto-synced into the catalog — live prices, inventory, images, variants. Rankings are organic and unsponsored: sorted by query relevance, not paid placement.
If you're not in the feed pool, a beautiful product page won't save you.
2. Your feed and your page must agree
One price in the feed, another on the product page → the model flags a conflict, and the store can get demoted or excluded from recommendations [Moderate]. The classic trap: you change the page price for a sale, the feed sync lags, and one inconsistency sticks in the system's memory.
3. Product data completeness
Per Shopify/OpenAI guidance, the high-leverage fields:
- Titles with material, dimensions, and use case ("100% Cotton 200TC Queen Sheet Set — Fits 18\" Mattress," not "Sheet Set Style A")
- Descriptions in full natural sentences
- Attributes filled: color, size, material, weight
- Image alt text: 8–15 words of real description
- Category at the deepest level of the taxonomy
- 8–15 specific tags
Why: ~70% of AI shopping queries are natural-language questions. Your product data is the script the AI reads from [Moderate].
4. Freshness: one stockout gets remembered
A product that's out of stock or mispriced gets recommended once → the merchant gets demoted. Products untouched for 12+ months are treated as low-relevance [Moderate]. AI's worst fear is recommending something you can't buy.
5. Reviews are weighted above your own site
AI engines lean on review platforms, Reddit, and YouTube — weighted above the brand's own site. Real, specific reviews beat high-star-but-sparse ones [Moderate]. Clear return/shipping policies act as legitimacy signals.
6. JSON-LD structured data
Product / Offer / Review / Organization markup, validated with Google Rich Results Test (free). After feeds, this is the second data source. A page with no structured data is, to an AI crawler, roughly half a blank page.
7. Don't block AI crawlers in robots.txt
Check yourdomain.com/robots.txt for blocks on GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot. Shopify allows them by default, but "block all bots" templates and aggressive security plugins cause collateral damage.
8. llms.txt: monitor it, don't prioritize it
It's a community convention, not an official standard. The vendor best-practice line is verbatim "monitor it, don't prioritize it" [Strong]. Ten minutes of config; don't expect miracles.
9. Authority signals (weak evidence)
One vendor blog claims citations and expert quotes lift AI citation rates 30–40% [Weak — single party]. Interesting, not a basis for action.
The pattern
The connection layer has been made zero-config by the platforms. The question is no longer "how do I plug into AI" — it's "now that I'm plugged in, does my data deserve to be recommended?"
I built a free scanner that automates the checks above (robots.txt, llms.txt, feeds, JSON-LD, price consistency). We ran it against allbirds.com: 93/100 — passed on platform detection, feeds, and structured data; missing review aggregation and Organization schema.
Try it on your own store — takes about a minute:
https://muse.ai/s/free-ai-visibility-checker-qxd5xhdxmnxbbww?utm_source=devto&utm_medium=social&utm_campaign=2026-09-merchant-tutorials
Top comments (0)