How it works
A generative AI model is trained on very large amounts of data. It learns the patterns of language or images, then produces a likely answer to the request it receives.
For text, large language models (LLMs) do the work. They predict the next word in a sentence, one word at a time, which lets them answer a question, summarize a document or rewrite a product description.
On its own, a model doesn't know your catalog, your stock levels or your return policy. It can also make up an answer that sounds right but is wrong (a hallucination). In e-commerce, it has to be connected to the brand's data: product pages, FAQ, shipping terms.
Why it matters in e-commerce
Generative AI can answer a specific question, asked in the shopper's own words, at any time of day. A shopper who asks for "a moisturizer for dry skin that doesn't feel greasy" gets a relevant answer, where a classic search bar only matches catalog keywords.
It also helps behind the scenes: writing descriptions, translating, summarizing reviews. But its biggest effect on sales happens on the site, at the moment a shopper hesitates.
Real-life example
A shopper lands on a travel bag page and asks whether it fits in an overhead bin and holds a 15-inch laptop. The assistant pulls the dimensions from the product data, answers in two sentences and suggests a larger bag if this one won't work. The shopper didn't have to open the specs tab or email customer service.
Common confusion
Generative AI and agentic AI often get mixed up. Generative AI produces content: an answer, a text, an image. Agentic AI chains actions to reach a goal, such as finding a product, checking a detail and adding it to the cart. A modern shopping assistant combines both.