AI visibility audit: how to audit the sources AI answer engines cite

A fixed set of buyer prompts run across the AI engines that matter to you, with every answer recorded: whether your brand is mentioned, which URLs are cited, which competitors appear and the sentiment. The citation audit then explains which sources shape those answers and which of your pages to improve first.

Run the full audit once to set a baseline, then re-run the same prompt set on a regular schedule (for example every two weeks) so changes are comparable. AI answers vary between runs, so look for patterns across several runs rather than reacting to one answer.

No. A citation signals the page contributed to the generated response or was presented as supporting evidence — it doesn’t mean the page is correct, preferred across every prompt, or responsible for a brand recommendation. Keeping the citation connected to the exact prompt, provider, timestamp and response is what prevents a raw URL count from being mistaken for authority everywhere.

Split sources into owned content, competitor-owned content, independent sources like publishers and review platforms, and primary evidence such as standards, datasets and original research. The mix reveals how the category earns trust — a market dominated by primary research needs a different content strategy than one shaped by peer communities and comparison sites.

Open the source and verify it genuinely supports the surrounding answer — check freshness, authorship, specificity, accessibility, and whether the cited passage is primary or just repeats another source. Also track broken URLs and changed content, since an old citation can remain in historical evidence even after the destination page changes.

Not on a single citation. Look for recurring evidence patterns across commercially important prompts instead, then strengthen your own sources with clearer claims, original evidence, transparent methodology and useful structure — copying a page that was cited once tells you almost nothing about why it worked.