There is no single public formula for how every AI assistant recommends a business. Different products use different models, indexes, retrieval systems and personalization. But a useful working model is four stages: eligibility, retrieval, confidence and selection.
1. Eligibility
Can the system access and parse enough information to know the business exists? That includes crawlability, indexability where relevant, readable content and a coherent identity.
2. Retrieval
Is information about the business relevant to the user’s question? Google documents query fan-out in its generative search systems, which means a broad question can trigger several related retrievals. Exact-match pages for every variation are therefore unnecessary and can become scaled, repetitive content.
3. Confidence
Do multiple signals agree on what the business is? Consistent first-party information helps. Independent mentions, reviews, profiles and editorial sources can add corroboration. Contradictions do the opposite.
4. Selection
Even an eligible, well-defined business may not be selected. The assistant still has to decide that the company is useful for this particular question, user, location and moment. This is where authority, relevance, recency and the competitive set matter.
What businesses can control
You can control whether your own site is readable, accurate and useful. You can control whether your profiles contradict one another. You can create first-hand content and earn legitimate third-party coverage. You cannot control the model’s final wording or guarantee a fixed recommendation position.
The measurement implication
Because selection is probabilistic, the useful metric is not “did we appear once?” but “how often did we appear across a stable set of buyer questions, and which competitors appeared more often?”
That is why Visible2 separates the readiness scan from the measured visibility protocol.