Why Product Content Is Always Behind Catalog Growth
Merchandising teams don’t lack product content strategy they lack throughput. A new SKU import lands with a title and a spec table; turning that into a page that actually converts takes a writer who understands the category, the target shopper, and the brand voice. That doesn’t scale linearly with catalog size, so the newest and longest-tail products are almost always the ones with the thinnest content.
AI changes the economics of the long tail specifically: it doesn’t write the flagship products your best copywriters already handle well, it writes the other 35,000 SKUs that would otherwise never get real content at all.
Layer 1 – Claude API: Product Copy at Catalog Scale
An App Builder action sends each product’s structured attributes, category context, and brand voice guide to Claude API, generating a description, a short comparison-relevant highlight list, and an FAQ answering the questions that category’s shoppers ask most.
No invented specs: The prompt explicitly restricts the model to stated attributes. A shopper trusting a claimed material or dimension is a return and a support ticket if the model fabricated it.
Layer 2 – Adobe Sensei: Recommendations That Learn Per Visitor
- Session-aware recommendations recent browsing weighs more than a static “frequently bought together” rule
- Cross-sell vs up-sell distinction different placements use different Sensei models tuned for each goal
- Cold-start handling new SKUs with no purchase history still surface via attribute similarity
Layer 3 – AEM + Adobe Commerce Integration: Content and Catalog in Sync
AEM’s Commerce integration keeps AI-generated content and live commerce data on separate refresh paths that converge at render time content updates on a publish cycle, while price and availability pull live from Adobe Commerce on every request.
Where Human Review Stays
AI drafts, merchandising approves: Every AI-generated product description lands as a draft Content Fragment, not a published page. A merchandiser reviews for accuracy especially for regulated categories where an invented or imprecise claim carries real liability. Flag AI-drafted copy unreviewed past a set window so a backlog doesn’t quietly turn into unreviewed publishes under deadline pressure.
Implementation Checklist
- Build an App Builder action that generates product copy strictly from supplied attributes no invented specs
- Write AI-generated copy as draft Content Fragments requiring merchandiser approval before publish
- Flag long-review-window drafts so backlog pressure doesn’t turn into unreviewed publishes
- Enable Adobe Sensei session-aware recommendations, tuned separately for cross-sell and up-sell placements
- Configure cold-start attribute-similarity recommendations for newly added SKUs
- Keep price, inventory, and promotions live from Adobe Commerce never cached alongside AI-generated content
- Require additional review for regulated categories with safety or compliance claims
- Prioritize AI copy generation on long-tail SKUs first
What to Measure
- Catalog content coverage – percentage of SKUs with real description content vs spec-sheet-only
- Draft approval rate and turnaround – how much merchandisers edit AI drafts, and how long review takes
- Recommendation click-through and attach rate – compared against the prior static-rules baseline
- Return rate on AI-copy products – a leading indicator if generated copy is over-promising anything
Final Thoughts
The retailers winning on product content in 2026 didn’t hire bigger copywriting teams. They stopped treating catalog content as something that scales with headcount, and started treating it as something AI drafts and a much smaller merchandising team approves.
Start with your thinnest-content category usually the long tail and measure the coverage gain before expanding into recommendation tuning.
