Many teams find that AI “knows” them yet still doesn’t recommend them. The usual cause is collapsing three different things into one: mentioned, recommended, and cited.
What the three layers are
- Mentioned: ask AI “describe this product” and it can — so it lives in the model’s parametric memory;
- Recommended: a user asks the category question “what’s the best X?” and the answer includes you;
- Cited: the sources the answer footnotes actually mention your page.
Each layer is harder than the last, and each calls for a different fix.
Why you must separate them
AI knowing you ≠ AI expressing you on the category question; AI citing many sources ≠ those sources mentioning you. Without separating the diagnosis, you might keep editing your website when the real problem is absence from the rankings and reviews AI draws on.
The matching fixes
- Unknown brand → add core brand facts and structured data;
- Retrieved but not recommended → strengthen citability, comparisons, and evidence;
- Absent from the corpus → get into the third-party content pool AI samples.
First locate which layer fails, then decide what to change — that is the value of diagnosis.