Adobe Adds Catalog Agent for LLM Shopping

Adobe's Brand Visibility documentation identifies a common technical limitation in ecommerce: product detail pages often place specifications, variants, and other important information behind JavaScript-rendered tabs, expandable panels, interactive modals, or shopping wizards. According to that documentation, AI agents may not access this deeper information even when human visitors can see it.

The documentation describes the Commerce Catalog Agent as reading catalog data including variants, product relationships, attributes, facets, category metadata, and product characteristics, then comparing it with the content accessible to AI agents on a corresponding product detail page. Adobe Brand Visibility surfaces pages with a catalog-data visibility gap and prioritizes them by agentic traffic volume, according to the documentation.

T2Online reports that Adobe's approach creates a machine-readable layer intended for AI systems rather than human visitors. That distinction matters technically: an LLM-driven shopping assistant needs structured facts to resolve requests such as product compatibility or feature comparisons, while the visual storefront is optimized primarily for people.

A related Product Catalog Enrichment capability is currently in beta for Adobe Commerce customers, according to Adobe LLM Optimizer documentation. The tool evaluates product names and descriptions, then generates intent-oriented alternatives intended to make catalog entries easier for LLMs to interpret. Adobe states that the enrichment focuses on value-driving attributes and intentionally excludes price and inventory.

The documentation uses a coffee-grinder example to illustrate the distinction. Rather than leaving an LLM to infer shopper relevance from a product name and a raw list of specifications, the tool can generate a description connecting technical details, such as grind settings and motor power, to an at-home espresso use case. Adobe states that users can apply the proposed enrichment directly to the Commerce catalog with one click.

That source-of-truth approach differs from generating AI-specific copy only at the storefront layer. In comparable commerce-data systems, centrally managed product attributes can reduce inconsistencies across storefronts, advertising feeds, marketplaces, and assistant-facing discovery surfaces. The quality of that outcome, however, still depends on catalog governance, attribute completeness, and review of generated language before it is published.

Adobe Digital Insights reported that traffic from AI sources to US retail sites grew 125% year over year from April through June 2026, according to FoneArena and T2Online. The outlets also cite Adobe's figure of 693% year-over-year growth in AI-referred traffic during the November-December 2025 holiday period.

T2Online notes that Adobe did not publish the absolute share of retail traffic accounted for by AI assistants, and that high percentage gains can reflect a small starting base. Still, the measurements provide context for why merchants are examining whether product pages expose enough structured, current data to systems that answer product questions before a shopper reaches a retailer's site.

For data and ML teams, the release places product catalogs closer to the retrieval layer of AI shopping. Comparable deployments typically make data lineage, update latency, variant normalization, and validation of generated product language important operational concerns, because stale or inconsistent catalog facts can propagate into recommendations across multiple channels.