A business can appear in an AI answer without being recommended. It can be cited as a source, mentioned in a long list, or described as one possible option while another company is explicitly favored for the buyer's need.
| AI visibility | AI recommendation |
|---|---|
| Was the business mentioned or surfaced? | Was the business selected, favored, or presented as the best fit? |
| How often did it appear? | For which buyer decisions did it win or lose? |
| Was it cited or represented? | Which competitors won instead, and how consistently? |
| What share of answer presence did it receive? | What candidate, belief, and evidence pattern appears around the decision? |
Inclusion is not selection
Suppose an answer lists five businesses and then says one is the best choice for a buyer with a specific constraint. All five businesses were visible. Only one received the recommendation outcome that mattered.
This is why inclusion rate and recommendation rate should be reported separately. Inclusion can be a useful prerequisite or diagnostic signal, but it should not be described as a commercial recommendation win when no selection occurred.
Visibility: “The system knows or surfaces us.”
Recommendation: “The system favors us for this buyer decision.”
Recommendation depends on the decision context
There is rarely one universal answer to “Who is best?” A recommendation can change when the buyer's situation changes. Relevant variables can include:
- the problem or desired outcome;
- buyer type and constraints;
- product or service category;
- geography and availability;
- price, risk, urgency, or required capability;
- the way the question is framed;
- the AI provider, consumer product, and collection environment;
- normal variation across repeated responses.
A company may be strongly recommended for one customer profile and a poor fit for another. Recommendation Intelligence therefore begins with the commercially important decision, not an undifferentiated visibility score.
A stronger measurement set
A useful recommendation study can preserve several outcomes at once:
- Mention or inclusion: Did the company appear anywhere in the answer?
- Recommendation: Was it selected or favored for the decision?
- Order: Where did it appear, without assuming one answer creates a stable rank?
- Competitor distribution: Which alternatives were mentioned or recommended across trials?
- Context distribution: How did outcomes vary by prompt, buyer need, geography, provider, or environment?
- Visible evidence: Which citations or stated reasons appeared, without inferring hidden retrieval?
Visibility and recommendation are connected layers
AI visibility is not dismissed. A business that never enters the candidate set may have a visibility or entity-understanding problem. But improving visibility alone does not guarantee that the business will be favored when the system makes a choice.
Haenisch connects familiar market language to a broader progression:
GEO / AI Visibility → Recommendation Intelligence → Recommendation Engineering → Measured Recommendation Outcomes
The practical goal is not to appear everywhere. It is to understand and improve the buyer decisions worth winning, while remaining honest about uncertainty and what the evidence can establish.