Where AI assistants get their information
AI assistants build their answers from four separate information pathways, and each one is a route to being mentioned or cited. Models are trained on a fixed body of data, then topped up with live retrieval from the web, content licensed through partnerships with AI companies, and data users pull in themselves from connected tools and uploaded files.
Training data has a cutoff date, so assistants fall back on live retrieval for anything recent. Crawlers run by AI companies and search engines fetch that material, which means a robots.txt file blocking those crawlers, or pages missing from Google and Bing indexes, keeps content out of reach. Mentions on third-party sites matter too, because assistants ground answers in the wider web and cross-check what they find there against a brand's own pages.
Licensing deals carry particular weight. OpenAI and Google both license content from Reddit and Stack Overflow, and OpenAI also draws on Yelp and Time, so material from those partners tends to surface prominently in answers. Appearing in a partner's content, through a Yelp listing or a Reddit thread, is a direct route into the answers those systems produce.
For brands, the practical upshot is that visibility is not a single-channel problem. Allowing AI crawlers access, confirming indexation, structuring content around the prompts buyers actually use, and building corroborating mentions elsewhere all feed the same result. Semrush's AI Visibility Toolkit tracks mentions, citations and cited pages over time, broken down by platform, so teams can see whether those efforts are landing.
Reported by Semrush Blog ↗
Topics: ChatGPT, Citations, Measurement