Abstractions about AI catalog optimization are only useful to a point. At some point you have to look at an actual product listing and walk through what changed, and why. This post does that. We take a single product, show the before state, walk through the changes we made, and explain which AI agent evaluation criterion each change addresses.
The Product: a Trekking Pole Set
We picked this product from a pilot retailer's catalog because it is a useful middle-ground example: not a simple commodity item, not a highly technical piece of gear. A trekking pole set has meaningful specifications that matter to buyers, but its original listing had been written for Google Shopping performance, not AI agent comprehension.
The Before State
Product title: "Trekking Poles Collapsible Hiking Poles Lightweight Anti-Shock Walking Sticks Men Women 2 Pack"
Product description (abridged): "Perfect trekking poles for hikers! Lightweight aluminum construction makes these poles great for all types of terrain. Anti-shock system absorbs vibration. Adjustable length fits most adults. Comfortable cork grips. Great for hiking, backpacking, snowshoeing, and more. Order yours today!"
Attributes listed: None in structured format. The description contained some attribute-like phrases but nothing in a machine-parseable field.
Schema markup: Basic Product schema with name, image, and offers. No itemProperties, no brand entity, no review aggregate.
The After State
Product title: "Trailbound Carbon Fiber Trekking Poles, Flick-Lock Collapsible, 3-Section, Pair"
Product description (rewritten): "Collapsible carbon fiber trekking poles designed for day hikers and weekend backpackers who need a reliable, packable support on mixed terrain. Three-section flick-lock design adjusts from 63cm to 135cm in 5cm increments. Carbon fiber shafts reduce pack weight versus aluminum equivalents without compromising stiffness on rocky descents. Anti-shock spring system absorbs trail vibration; disengages for ascent. Ergonomic cork grip wicks moisture during extended use. Compatible with standard basket systems for snowshoeing conversion."
Structured attributes added:
Material: Carbon fiber (shafts), Cork (grip)
Shaft sections: 3 (flick-lock)
Collapsed length: 63 cm
Extended length: 63 to 135 cm
Weight per pole: 215 g
Anti-shock: Yes, spring-loaded, disengageable
Basket type: Standard 65mm interchangeable
Sold as: Pair (2 poles)
Use case: Hiking, backpacking, snowshoeing (with separate baskets)Walking Through Each Change
Title Change: From Keyword Stack to Specific Identity
The original title was built to capture keyword impressions: "trekking poles," "collapsible," "lightweight," "anti-shock," "walking sticks," "men women," "2 pack." That is seven keyword groups in one title. For Google Shopping's keyword matching model, that approach has merit. For an AI agent trying to understand what this product is and how it differs from similar products, the title is noise with a product name buried in it.
The rewritten title identifies the brand, the primary material (carbon fiber vs. aluminum is a meaningful differentiator), the mechanism (flick-lock vs. twist-lock affects speed of adjustment and failure mode), the configuration (3-section), and the unit of sale (pair). An agent comparing trekking poles can now immediately understand what this product is without parsing a run-on string.
Description Change: From Feature Claim to Evaluative Narrative
The original description makes claims ("perfect for hikers," "great for all types of terrain") without grounding them in specifics. "Lightweight" is not a measurement. "Anti-shock system absorbs vibration" does not tell an agent whether the system is spring-loaded, air-cushioned, or foam-damped. "Comfortable cork grips" is an unverifiable claim.
The rewritten description places the product in a specific use scenario ("day hikers and weekend backpackers on mixed terrain"), makes every attribute claim concrete ("63cm to 135cm in 5cm increments"), and explains the mechanism behind each feature claim ("anti-shock spring system absorbs trail vibration; disengages for ascent"). An agent evaluating this product for a shopper who asked "what trekking poles are best for a weekend backpacking trip in rocky terrain" has enough information to evaluate relevance.
Attribute Addition: From Implicit to Explicit
The most common failure mode in catalog optimization for AI agents is not wrong information, it is missing information. The original listing contained some of these facts embedded in the description text, but not in a structured format that an agent can parse without reading the full description. Adding an explicit attribute table gives agents a scannable signal layer that does not require full description comprehension.
Notice the specifics in the structured attributes: weight per pole (not total weight, not approximate weight), collapsed length (not "compact"), extended length as a range with adjustment granularity. These are the kinds of specifics that allow an agent to answer a comparison question: "Which trekking poles are lightest for ultralight backpacking?"
Schema Markup Extension
The schema update added brand (with @type: Brand and name), weight as a QuantitativeValue, height as a range for collapsed/extended, and material. We also added the numberOfItems field to clarify "sold as pair." The review aggregate was populated from existing review data the retailer had.
The schema additions do not duplicate the description; they provide a parallel machine-readable representation of the key facts. For agents that index schema directly rather than parsing page text, this is the primary data source.
What This Does Not Fix
A rewritten listing does not make a product more competitive on price, does not change its review score, and does not affect whether the retailer has it in stock. If a competitor has a similar carbon fiber trekking pole with 500 reviews at 4.6 stars and this listing has 12 reviews at 4.1, the competitor will likely receive more recommendations even with equivalent listing quality. We are not in the business of claiming otherwise.
The rewrite makes the product legible to AI agents. Whether those agents recommend it depends on the full competitive context. What we have consistently seen in practice is that well-specified listings with good structured attributes perform better against competitors than under-specified listings, even when other factors like review count are similar. The floor for "AI agent can evaluate this product" has to be cleared before anything else matters.
The Broader Pattern
Across the listings we have rewritten, the most common improvements are: title specificity (removing keyword stacking), attribute formalization (pulling implicit facts from descriptions into structured fields), and description reorientation (from feature claims to evaluative scenarios). Schema coverage fills in behind those. In most catalogs we see, getting those three things right for the high-priority SKUs moves the needle more than perfecting schema on every listing in the catalog.
Is your catalog ready for AI shopping agents?
ReFiBuy audits your product feed and rewrites the listings AI agents cannot evaluate. See how it works.
See plans
