The math no longer works: a mid-sized catalog times eight channels times three languages produces more text variants than any editorial team can maintain. Shop long descriptions, marketplace bullets in Amazon logic, Google Ads titles with character limits, filter labels, badge copy, category page snippets – anyone who treats each variant as a standalone text has already lost. The answer is an architectural shift for which the term product content supply chain has taken hold: enrich once, publish everywhere.
The principle: texts as views on attributes
At the core of the idea is an inversion: the text is not the asset – the attribute is. Material, application area, compatibility, USPs, certifications – cleanly structured and enriched. Every text is then a generated view on this model: the long description narrates the attributes, the marketplace bullet compresses them, the ads title takes the two strongest. When an attribute changes – a new certification, a corrected limit value – every output changes with it. No searching across eight systems, no forgotten variant.
The output matrix
| Output | Source in the model | Format rules |
|---|---|---|
| Shop long description | all core attributes + application context | SEO-structured, the shop's tone of voice |
| Marketplace bullets | top 5 USPs + mandatory information | character limits and style per marketplace |
| Ads/feed titles | brand + type + differentiating attribute | hard character limit, keyword first |
| Filter labels & badges | normalized individual attributes | controlled vocabulary, no free text |
| Category page snippets | aggregated attributes of the category | programmatic, consistent across the category |
Microcopy: small per element, huge in total
The underestimated half of the supply chain is the small copy. A single filter label looks trivial – but thousands of filter labels, badges and USP snippets determine scannability, filter quality and, ultimately, conversion. This is exactly where generation from normalized attributes with a controlled vocabulary pays off: free-text sprawl in filters (“red”, “Red”, “glossy red”) is a data problem, not an editorial problem.
What this has to do with personalization
Once texts are generated views, segmentation becomes affordable: the same attributes produce the technical description for the procurement buyer and the benefit-oriented one for the end user – without anyone maintaining two texts. The prerequisite, here too, is the governance layer: tone-of-voice profiles per segment, terminology as a hard rule, validation before release. The content factory is only as good as its quality control.
The product content supply chain is the target architecture of our enrichment projects – from the attribute model to channel output. Entry points and pricing: PIM services page.
