Amazon deploys generative AI to write sales listings

Amazon has launched a new generative AI tool that creates copy listings for users selling items on the company’s e-commerce platform.

Designed to simplify the selling process, the new tool reduces the need for sellers to enter many pieces of specific product data when generating product descriptions. Instead, users can now enter a brief description of the product they are listing for sale – Amazon said this can be a few words or sentences – and the tool will generate the necessary copy, which sellers can then review and refine before uploading their item to the Amazon catalog.

“These new capabilities will help sellers create high-quality listings with less effort and present customers with more complete, consistent, and engaging product information,” Amazon said in a blog post announcing the tool.

The new generative AI tool is fueled by a large language model (LLM) that Amazon has been developing internally, as revealed by CEO Andy Jassy during the company’s first-quarter earnings call in April. Originally built to support its smart assistant, Alexa, Jassy told analysts on the call that Amazon’s LLM model contained “a couple of hundred million endpoints” that were being used across entertainment, shopping, and smart homes. 

That same month, Amazon’s cloud computing division, AWS, launched Bedrock, a foundation model API service that allows small companies who lack the necessary people power to develop their own LLMs to access pre-trained models, including those built by AI21 Labs, Anthropic, and Stability AI.

“With our new generative AI models, we can infer, improve, and enrich product knowledge at an unprecedented scale and with dramatic improvement in quality, performance, and efficiency,” said Robert Tekiela, vice president of Amazon selection and catalog systems, in comments posted alongside the announcement.

“Our models learn to infer product information through the diverse sources of information, latent knowledge, and logical reasoning that they learn. For example, they can infer a table is round if specifications list a diameter or infer the collar style of a shirt from its image,” he said.