Recommendation Intelligence is the disciplined study of how an AI system handles a decision: which candidates it considers, what it appears to believe about them, which evidence is visible or stated, and which option it ultimately favors under the tested conditions.
In practical terms: Where does a business win? Where do competitors win instead? How stable is that pattern? What evidence may be relevant? Which commercially meaningful changes are worth testing?
The decision-analysis chain
- Decision
- Candidates
- Beliefs
- Evidence
- Recommendation
Decision
What is the buyer actually trying to choose? A recommendation only makes sense in relation to a specific need, constraint, geography, product, service, or use case.
Candidates
Which businesses or options enter the set being considered? A strong business cannot be recommended if it never becomes a candidate for that decision.
Beliefs
What characteristics, capabilities, limitations, or associations does the response appear to attribute to each candidate? These are observed representations in an answer—not direct access to a model's hidden internal state.
Evidence
What public facts, citations, reviews, credentials, structured information, or other visible sources support those representations? Retrievable evidence is not automatically decision-active evidence, so this layer requires careful interpretation.
Recommendation
Which option is selected, favored, shortlisted, or presented as the best fit for the buyer context? Recommendation is a stronger outcome than a mention alone.
How this relates to GEO and AI visibility
Generative Engine Optimization (GEO), AI search visibility, and Answer Engine Optimization (AEO) describe an important emerging market. They commonly focus on whether a brand appears, how often it is mentioned, which citations are used, or how prominently it is surfaced in AI-generated answers.
Those measures can be useful. They are the observation layer. But visibility does not fully explain why one business was favored over another, whether the pattern holds across repeated trials, or whether the gap is commercially worth addressing.
| Layer | Primary question | Typical evidence |
|---|---|---|
| AI visibility / GEO / AEO | Did the system surface or mention us? | Mentions, citations, share of voice, answer presence |
| Recommendation Intelligence | Did the system favor us for the decision—and what may explain the observed outcome? | Repeated recommendations, candidate sets, represented beliefs, evidence, context, variation |
AI Visibility → Recommendation Intelligence → Recommendation Engineering → Measured Recommendation Outcomes
From intelligence to engineering
Recommendation Intelligence measures and diagnoses. Recommendation Engineering begins only when the evidence supports a defined intervention hypothesis.
The advanced chain is:
- 01Observed Recommendation
- 02Decision Decomposition
- 03Evidence Diagnosis
- 04Intervention
- 05Remeasurement
- 06Causal Learning
The last step must be earned. A before-and-after difference can be informative, but causal conclusions require a design strong enough to address volatility, confounding changes, and alternative explanations.
What Recommendation Intelligence can—and cannot—establish
Repeated observations can reveal distributions, provider differences, recurring candidate patterns, recommendation gaps, and hypotheses about potentially relevant evidence. They can support better decisions about what deserves investigation.
They do not expose hidden retrieval or model reasoning. They do not turn one AI answer into a permanent ranking. They do not guarantee that a change will cause a different recommendation. Haenisch reports the collection conditions, denominators, exclusions, and limitations needed to keep those boundaries visible.