The nine attributes agents ask for that your catalog probably lacks
A practical inventory of the fields that decide whether an assistant can recommend you, derived from the constraints that actually appear in buying questions.
When we audit a catalog, the gap is rarely exotic. It is nine or so ordinary fields that exist somewhere in the business, in a spec sheet, a support macro, a sizing image, but not in the product record. Here is the list, roughly in order of how often it costs a recommendation.
1. Constraint-bearing dimensions
Not "medium". The actual measurement, in units, with the tolerance. Buying questions are full of constraints, a shelf that has to fit a space, a part that has to fit a model, a garment that has to fit a body. If the number is inside a JPEG, it does not exist.
2. Compatibility and fitment
What this works with, explicitly enumerated. Merchants tend to publish the positive case and stay silent on exclusions, which forces the agent to either guess or skip. Skipping is safer, so it skips.
3. Material and ingredient composition
Full breakdown with percentages where relevant. This drives an enormous share of constraint queries, allergies, sensitivities, sustainability filters, regulatory categories, and it is very often present only in marketing prose.
4. Substantiated claims
"Waterproof" is a marketing word. "IPX7, tested to 1m for 30 minutes" is a fact an agent can repeat without hedging. Assistants systematically prefer the version they can attribute, because it lowers their own risk.
5. Availability with a horizon
In stock is a snapshot. Agents increasingly want to know lead time, backorder status, and whether the item will still be available at the end of a multi-step task. A stale availability signal is a returned recommendation.
6. Real pricing shape
Including regional pricing, what is bundled, what is not, and where shipping lands. Price surprises are one of the highest-frequency reasons an agent-originated session fails at checkout.
7. Returns and warranty as data
Window, condition requirements, who pays return shipping, and what voids it. This is the single most-asked policy question and it is almost universally trapped in a prose page written by legal.
8. Review provenance
Aggregate ratings are weakly useful. What helps is structured, attributable review data with volume and recency, because it lets the assistant say something specific rather than "reviews are generally positive".
9. Identity of the seller
Who is the authorised seller of record, which listings are grey market, and where the canonical buy path is. Without it, agents route to whichever listing they found first, frequently a reseller undercutting you with your own product.
How to sequence the work
- Start with the attributes that appear in your top twenty buying intents, not with the whole schema.
- Fix the fields that block recommendation before the ones that improve it.
- Make the data generated, not hand-maintained, anything manual drifts within a quarter.
- Instrument it, so you can tell when a claim silently breaks.