AI Merchandising: What It Is and How to Evaluate It

AI merchandising is the use of artificial intelligence to support the decisions a merchandiser makes: which products to carry, how to present and rank them, what copy and attributes to use, and what price inputs to apply. It spans two surfaces: on-site merchandising (how products appear on a retailer's own site) and merchandising for external AI shopping surfaces (how products appear when an AI shopping agent recommends them).

Traditional merchandising leans on human judgment, seasonal planning, and sales data. AI merchandising adds algorithms that process larger data sets, test more variations, and adapt faster, while still relying on merchandisers to set strategy and approve changes that matter.

The three terms are related but not the same:

• AI merchandising. The commercial decisions about assortment, ranking rules, presentation, copy, and price inputs across surfaces.
• Personalization. Adapting the experience to one shopper based on their history and context. Personalization runs on top of merchandising rules.
• Search ranking. Ordering products for a specific query. Merchandising sets the ranking rules and boost or burial that search applies.

One way to keep them straight: merchandising decides what is eligible and how it is presented; personalization tunes that for one person; search ranking orders it for one query. A platform can do all three, but they are separate jobs.

On-site merchandising optimizes a storefront the retailer controls: layout, banners, category-page ranking, and product detail pages. You can see what you ship and change it directly.

Merchandising for AI shopping surfaces is different. When a shopper asks an agent "what is the best wireless keyboard for programming?", which keyboard gets recommended is a merchandising outcome, but the merchandiser is an AI agent using your structured data. You do not control the agent's layout; you influence it through data completeness, attribute coverage, and feed quality. Feed optimization and enrichment are the levers.

The use cases AI merchandising touches, across surfaces:

• Assortment. Predicting which products sell in which markets or channels.
• Ranking and category-page layout. Deciding product order on category pages and in search.
• Copy and attributes. Generating and testing titles, descriptions, and attribute values for conversion and AI comprehension.
• Price inputs. Providing pricing signals such as sale price and currency, distinct from autonomously setting price.
• Recommendations. Suggesting related products. See AI Product Recommendations.

Not every platform does all of these, and not every retailer needs all of them. The split between "AI proposes" and "human approves" is what separates a governed tool from an autonomous one.

AI merchandising needs four inputs:

• A clean catalog. Complete, structured product data with attributes filled in.
• Attributes that match how people shop. Material, dimensions, use cases, and other fields that answer the questions shoppers ask agents.
• A feed per surface. The right format for each destination, kept current for price and availability. See product feed management.
• Visibility data. How often your products appear, in what position, and for which queries, so you can close the loop. See AI Shopping Visibility.

AI merchandising works best when the AI proposes and a human decides. Governance matters most for changes that touch price, ranking, or public-facing copy:

• Preview before publish. See what a proposed change does before it ships.
• Approve, reject, or edit. A merchandiser stays in control of consequential changes.
• Revert. Roll back a change that did not work.
• Audit trail. A record of what changed, who approved it, and when.

This is the difference between a tool that helps a merchandiser and one that quietly rewrites a storefront. See AI Catalog Management for the workflow Paz implements.

When you evaluate an AI merchandising platform, ask:

• Which surfaces does it cover? Your own site, external AI shopping surfaces, or both. Many on-site tools do not address AI surfaces at all.
• What is the review workflow? Preview, approval, rejection, and revert, or autonomous changes.
• Where does my data live? Keep a system of record you control.
• Does it overstate autonomy? Be cautious of claims that an AI will autonomously set prices or run your category pages with no human in the loop.
• How does it measure results? Clicks and conversions on-site, plus visibility and share of voice on AI surfaces.

For related evaluation criteria, see the feed optimization guide.

Paz supports the part of AI merchandising that connects to AI shopping surfaces:

• Catalog optimization proposals. Suggested changes to titles, descriptions, and attributes, with preview, review, approve, reject, and revert on all runtime tiers.
• Catalog quality scoring with field-level issues to prioritize work.
• Feed distribution to Google Merchant Center and OpenAI Commerce on Growth and Enterprise, plus UCP on Growth and Enterprise when the product surface is enabled.
• AI Shopping Visibility monitoring across ChatGPT, Google AI Mode, Perplexity, and Google AI Overviews.

Teams use this workflow to improve catalog readiness and observe supported AI shopping surfaces. Proposals remain human-reviewed; Paz does not autonomously set prices or change on-site page layouts. See Agentic Commerce Optimization and AI readiness.