How to write product descriptions AI assistants can recommend
When an AI assistant decides whether to recommend your product, the description is its interview with you. It can't pick the product up, can't call your support line — it can only work with what the text answers. Most Shopify descriptions fail that interview not because they're badly written, but because they're written for a shopper who can already see the photos, the price and the vibe. A machine reader starts from zero.
The template
A description that works for both audiences answers five things, in roughly this order:
- What it is and who it's for — one plain sentence. “A 750 ml vacuum-insulated steel bottle for day hikes and commutes.”
- The specifics — materials, dimensions, weight, capacity, compatibility. This is the section models quote when they justify a recommendation, and the section most stores skip.
- What makes it different — the honest one or two claims you'd defend in person: 24 h heat retention, lifetime warranty, made in your workshop.
- Practicalities — care, what's in the box, sizing guidance, shipping constraints.
- The questions customers actually ask — mine your support inbox and reviews for the top three and answer them right in the description. Bonus: this is also perfect FAQ-schema material.
Before / after
Before: “The Alpine Bottle keeps your drinks at the perfect temperature all day long. Stylish, durable, and perfect for any adventure. Get yours today!”
After: “A 750 ml vacuum-insulated steel bottle for winter hikes and daily commutes. Double-wall 18/8 stainless steel keeps drinks hot for 24 hours or cold for 36; weighs 340 g empty; fits standard car cup holders and most bike cages. Powder-coated exterior, bamboo cap with steel core, hand-wash only. Includes a spare gasket. Lifetime warranty against insulation failure.”
Every claim in the “after” version is something an assistant can match against a shopper's question (“keeps drinks hot all day”, “fits a cup holder”, “how heavy is it?”) and quote in its answer. The “before” version contains exactly zero matchable facts.
Notice what the rewrite is not: keyword stuffing. Models are good at spotting filler, and answer engines quote sources that sound credible. Specific, verifiable, plainly-written facts are the optimization.
Common failure modes
- Vibes-only copy. “Elevate your everyday” tells a model nothing. Keep the brand voice — in addition to the facts, not instead of them.
- Specs living only in images. Size charts and spec sheets baked into JPEGs are invisible to most text pipelines. Put them in text (or metafields) too.
- Duplicate manufacturer copy. If thirty stores paste the same paragraph, none of them earns the recommendation. Rewrite in your own words with your own details.
- Leaving the rest of the record thin. A great description under a junk title, no category and no images still loses — the nine signals work as a system.
Scaling it past ten products
Writing one description this way takes fifteen minutes. Writing four hundred doesn't happen — which is why thin catalogs stay thin. The scalable version of this workflow: audit the catalog to find the worst descriptions, let AI draft rewrites from your existing product data (never inventing specs — drafts should only restructure and complete what's true), review, apply, and keep the before/after so you can prove the change moved your scores and your AI-sourced traffic.