AI commerce readiness: what shopping agents actually need from your catalogue
ChatGPT, Gemini, and shopping agents are starting to recommend products. Here's what your catalogue needs to be discoverable by them — and what doesn't matter as much as the AI vendors claim.
Shopping agents (ChatGPT plugins, Gemini commerce surfaces, Perplexity Shopping, Adept commerce extensions) consume product catalogues differently from Google Search. The optimisation surface is adjacent to but not the same as SEO. Here's what actually matters for AI commerce discoverability in 2026.
What AI shopping agents read
- Structured product data. JSON-LD Product schema, OpenGraph Product, Microdata Product. They read all three; JSON- LD is the most reliably parsed.
- llms.txt. A flat-file at
/llms.txtdescribing your shop's contents in markdown. Increasing number of agents fetch it as a guided crawl signal. - Product feed APIs. Google Merchant Center feed, Facebook product feed, schema.org Dataset for the full catalogue. Some agents reuse these feeds.
- Long-form content. Buying guides, comparison articles, FAQ pages. Agents cite these when explaining a recommendation to the user.
The fields that matter most
From observing recent agent product mentions in the wild, ranked by impact:
| Field | Why it matters |
|---|---|
| Product description (full, structured) | Agents quote this directly when justifying a pick |
| Price + availability | Agents filter on these — out-of-stock = not surfaced |
| Identifiers (GTIN / MPN / Brand) | Lets agents dedupe across retailers |
| Categorisation (taxonomy + tags) | Lets agents match user intent ("running shoes" → category) |
| Reviews + ratings (real, schema-marked) | Trust signal for the agent's recommendation |
| Image quality + alt text | Agents that produce visual results need both |
| Returns + warranty terms (structured) | Agents increasingly surface this in product cards |
What doesn't matter as much as you'd think
- SEO meta tags. Agents are increasingly ignoring meta description and using the actual page content.
- Page speed. Important for human shoppers; agents fetch + cache. Don't sacrifice content for last-mile speed.
- "AI-optimised" copy. Vendors selling "AI-optimised product descriptions" mostly mean "verbose" — agents don't reward verbose, they reward accurate.
The llms.txt question
llms.txt is a 2024-introduced flat-file standard at https://yourshop.com/llms.txt that gives crawl-time agents a markdown summary of your site structure. Whether it's worth the work depends on:
- Do your top-performing pages have a clear semantic hierarchy? (If so, llms.txt is easy to generate from sitemap + structured page headings.)
- Are your products described well in their detail pages? (If not, fix that first; llms.txt won't paper over thin content.)
For most PrestaShop and WooCommerce shops, the highest-ROI work is better Product schema, not llms.txt. The latter is a secondary signal.
How to audit your catalogue's AI-readiness
- Pick 10 random products. Open Google's Rich Results Test (
search.google.com/test/rich-results) with each URL. The Product schema should validate with no warnings. - For each, verify: name, description, price, availability, image, brand, GTIN/MPN populated. Missing any of these = lower agent discoverability.
- Check your Schema.org Review markup is on real reviews, with Author + datePublished. No fake AggregateRating.
- Test 3 search queries on Perplexity Shopping and ChatGPT Shopping. Does your shop come up? If never, audit the gaps.
The compliance angle
AI shopping agents don't have a Merchant Center to approve you in. There's no single approval gate — discoverability is emergent. But a few things will get you de-prioritised:
- Spammy schema (review markup without real reviews, inflated star counts, fake price-drop badges).
- Cloaking (different content to bots vs humans).
- Geo-blocking AI agents while serving humans. Some sites block by user-agent — that's the opposite of AI- readiness.
What we built for this
NP Feed Doctor (in pilot) scans WooCommerce product data and generates an AI-readiness report — what fields are missing, what schema is broken, what taxonomies need cleanup. It also generates an llms.txt + a read-only agent catalogue from public product data if you want one.
For PrestaShop, the equivalent will land later in 2026. In the meantime, the manual audit + Rich Results Test is the right baseline.
Bottom line
AI commerce readiness is mostly "do classical product schema properly," with a small llms.txt + AI-curated review angle on top. Don't pay for "AI optimisation" plugins; the underlying work is unsexy schema hygiene.