Recommendation Surface Mapping: Where AI Systems Include or Exclude You
Retrievability and recommendability are not the same. A page can rank well for AI retrieval and still be absent from AI Overviews, chat recommendations, voice answers, and agent discovery. Surface Mapping diagnoses the gap.
AI recommendations do not happen in one place. They happen across at least four distinct surfaces: AI Overviews embedded in search results, chat-based responses from LLM assistants, voice assistant answers, and programmatic consumption by autonomous agents. Each surface has different structural requirements. A page optimized for chat recommendation may be entirely absent from voice retrieval. A page with excellent AI Overviews inclusion may be invisible to autonomous agents. Recommendation Surface Mapping diagnoses your coverage across all four.
Surface 1: AI Overviews
AI Overviews pull from content with featured snippet eligibility: a direct, concise answer in a clean paragraph near the top of the page, preceded by a question-structured heading. They also weight FAQ schema, strong E-E-A-T signals, and high schema completeness. Inclusion probability is modeled from Retrieval Quality Score × Schema Completeness × Citation Probability. Pages that score well on retrieval but lack FAQ schema and featured snippet structure will have low AI Overviews inclusion probability despite good overall visibility scores.
Surface 2: Chat-Based Recommendation
Chat recommendation from LLM assistants is the widest surface and the most strongly correlated with AI Visibility Score. The key differentiator: entity clarity. Chat systems need to confidently identify what your page is about before they will recommend it in a conversational context. A page with low Entity Confidence Score will be retrievable but not recommendable in chat contexts — the model cannot ground the recommendation in a specific, well-defined entity it trusts.
Surface 3: Voice Retrieval
Voice assistants operate under extreme constraints: they return one answer, it must be under 30 words, and it must be a direct response to the spoken query. speakable schema marks content as voice-retrievable. Without it, voice systems fall back to metadata extraction, which rarely produces a clean spoken answer. We check for speakable schema presence and, critically, whether the marked content actually contains sub-30-word direct answers. Schema without suitable content is scored as a voice retrieval blocker, not a voice retrieval signal.
◆For voice retrieval, the most effective structural pattern is a single-sentence direct answer immediately following a question-structured H2 or H3. Mark this structure with speakable schema. The answer must stand alone without context from surrounding paragraphs.
Surface 4: Autonomous Agent Discovery
Autonomous AI agents do not browse websites the way humans do. They look for machine-readable endpoints. SiteNexis probes for three agent discovery signals: /.well-known/ paths (for agent configuration, authentication, and capability discovery), robots.txt agent directives (explicit allowances for AI agent user agents), and OpenAPI endpoint availability (for service capability declaration). A domain with none of these signals has effectively zero probability of autonomous agent discovery regardless of content quality.
Coverage Gaps and Sequencing
The Recommendation Surface Map outputs a coverage gap report that lists missed opportunities per surface, with the required signals to fill each gap and an estimated impact classification. Gaps are sequenced: chat recommendation first (builds on existing AI visibility work), AI Overviews second (requires FAQ schema additions), voice retrieval third (requires speakable schema and answer restructuring), agent discovery fourth (requires technical endpoint work). This sequencing maximizes impact relative to effort.