Retrieval, not memory
The instinct is to read a missing recommendation as a judgment: the AI knows about us and chose someone else. That's almost never what happened. Modern assistants ground their answers by retrieving live pages at question time — the model's training gives it language and reasoning, but the specific names it drops usually come from what it just pulled.
So the real question isn't "why did it rank us low?" It's "was our page even in the set it read before it wrote the answer?" Retrieval comes first; being chosen to be quoted comes second. Miss the first step and the second never happens.
What being invisible actually looks like
This isn't hypothetical, and it isn't only a small-business problem. When we pointed our own five-engine panel at ourselves on day one, we appeared in 0.0% of the vendor-selection answers — while the established tools in the same answers took the lion's share.
| Tool | Share of answers |
|---|---|
| Semrush | 59.4% |
| Profound | 48.4% |
| Otterly.AI | 43.8% |
| The AI Citation Institute (us, day 0) | 0.0% |
Panel-wide share of vendor-selection answers on our frozen five-engine panel, 2026-07-19. The named tools are not our clients. A dated reading, not a standing ranking.
Why it's fixable
Because retrieval and selection reward things you control. Retrieval is on-page work scored against the questions buyers actually ask an assistant. Selection — getting quoted once you're retrieved — rewards a page that answers the question directly, states specifics a model can't invent, and carries citable sources.
The single strongest on-page lever we've measured is prosaic: inbound internal links. In our own audit of 91 posts, pages with zero inbound internal links were cited 4.5% of the time; pages with ten or more were cited about 44%. Same content, different link graph, roughly a tenfold swing.
Our own corpus, one point in time — a correlation on our data, not a guarantee for yours.
Measure first, then move
Don't guess which questions you're invisible on. Run them against the same frozen multi-engine panel a real measurement uses, sampled several times per engine, and read the actual answers. The gap map tells you which queries name competitors instead of you — and those are the pages worth building first.
Common questions
Does the AI just not know my business exists?+
It may not — but "knowing" is the wrong frame. Assistants retrieve pages at question time rather than recalling a fixed roster of companies. If nothing citable about you is retrievable, the model has nothing to pull, so it fills the answer with whoever it did retrieve.
I rank well on Google. Why am I still invisible in AI answers?+
Ranking makes you retrievable, which is necessary but not sufficient. An assistant can retrieve your page and still quote a competitor's because theirs answered the question more directly or carried better sources. Retrieval gets you into the candidate set; selection decides who gets named.
How do I find out where I actually stand?+
Run your real buyer questions across multiple engines, sampled several times each, and read which brands the answers name. A measured baseline beats a hunch — it shows the exact queries where you're missing, which is where the fixable work is.
keep reading
See what AI says about your brand right now — free, on a live engine.
Scan your domain