Which pages teach AI what to say about you
Every AI sentence about your brand was learned somewhere. Source attribution is the practice of tracing that sentence back to the specific URLs, press pieces, reviews, and structured data that taught it. Without source attribution, fixing perception is guesswork.
Most teams treat AI answers as black boxes. They read the output, dislike it, and try to fix it by rewriting their own website. That rarely works, because the website was not the source.
Models retrieve from the open web, from structured sources, from review sites, from forums, from press archives. The page that taught the model your positioning may be a 2-year-old Crunchbase entry, a single Reddit thread, or an article on a publication you don't even monitor.
Across our scans of tracked B2B brands, a handful of sources usually carry the majority of the model's framing. Typically 5–10 URLs account for 70%+ of the influence on a brand's description.
That's the leverage point. You don't need to influence the whole web. You need to identify and act on the small set of pages that actually shape the answer.
For every tracked prompt, we capture each model's citation set and link patterns. Where models don't disclose sources directly, we use prompt-probing techniques and retrieval signals to identify the most likely source pages.
Sources are then clustered by influence weight, how often they appear, in how many responses, with what framing, so the team can act on the few sources that matter, not chase every backlink.