AWS details Amazon Marketplace A+ Content scoring system, reporting 94% accuracy with OCR fusion

AWS’s November 5, 2026 post describes Amazon Marketplace’s approach to evaluating A+ Content on product detail pages. The system renders each complete page as one snapshot so it can assess text, images, and layout together instead of scoring assets separately. A vision-language model evaluates Visual Storytelling, Feature Presentation, A+ Layout, and Conversion Optimization, returning a score from 0 to 10, a confidence value, and a rationale for each dimension.

The training labels covered 5,478 product detail pages across 10 categories. One ecommerce marketing expert and four professional annotators rated the pages as high or low quality, with majority vote setting the consensus label. On a 20% holdout set, the vision-language-model score baseline reached 0.80 accuracy, while CatBoost trained on those scores reached 0.89. Adding OCR measures for text density, character density, spacing, and overall readability lifted the fused model to 0.94 accuracy and 0.95 F1.

For low-quality pages, the system uses SHAP features, product category, best practices, and model rationales to generate prioritized improvement actions. AWS says that, after selling partners revised content, average quality scores rose by 1.4 points on a 10-point scale in a statistically significant sample. It also reports that a worldwide production A/B experiment adjusting the visibility of low-quality A+ Content based on model decisions produced a statistically meaningful improvement in Ordered Product Sales; the post does not give a numerical lift.

The case connects automated scoring to a practical content-review workflow. Amazon’s seller-facing guidance separately says Basic A+ Content can increase sales by up to 8%, while well-implemented Premium A+ Content can increase sales by up to 20%. AWS’s post shows how it combines whole-page visual assessment with text-readability signals to flag content and suggest revisions.