Yahoo has added generative AI to part of its ad-targeting workflow using Amazon Bedrock, according to a new AWS Machine Learning Blog post. The work focuses on search retargeting, which helps advertisers reach people based on previous search activity.
The old system used Word2Vec embeddings and locality-sensitive hashing to expand advertiser keywords. AWS says that approach could miss current vocabulary, treat phrases too syntactically, and sometimes produce no useful expansion at all.
Yahoo’s newer workflow uses large language models on Bedrock to generate semantically related keyword expansions for audience segments in its demand-side platform. The goal is to connect search intent with display, video, and native advertising more accurately, while still filtering sensitive terms and storing expanded sets for later scoring.
The post is a vendor case study rather than an independent performance audit. Still, it shows a common enterprise pattern: generative AI is being inserted into existing data pipelines where better semantic matching can change operational outcomes without replacing the whole system.