Observable behavior, with its limits intact.
What an observation represents
An observation records one approved buyer question, its location context and timestamp in a supported retrieval environment. It does not reproduce every consumer answer product, reveal impression counts, or establish a universal AI ranking. Results may differ across runs and systems.
Site profile sampling
RankGen fetches public HTML from a homepage and at most two linked pages. Profiles are cached for up to six hours. Business identity comes from supported JSON-LD types; services may come from schema or descriptive internal-link labels. Missing fields mean not detected in the sample. We do not claim a comprehensive crawl or robots.txt audit.
How buyer questions are proposed
Suggestions use detected service and page language through a private question-generation method. They are not search-volume estimates. You approve every submitted question and can keep a keyword-style phrase unchanged.
Searches and subsequent searches
RankGen keeps generated wording and the order of exposed search events. Later searches are labeled subsequent searches. A later query may reflect information discovered earlier, but event order alone does not prove why the system generated it. Query text is unavailable when the retrieval environment does not expose it.
Yes, No and Unavailable
Yes means matching evidence was returned. No requires complete applicable data and no match. Unavailable means the relevant data was not exposed or the observation did not complete. A missing source list is never treated as a negative result. Opening evidence is reported only when exposed; otherwise it remains unavailable.
Domain and brand matching
Domain inclusion accepts the exact target domain, its www form and subdomains, with hostname boundaries. Brand mentions use a literal, case-insensitive phrase match for the detected business identity. They do not establish endorsement, sentiment or alias coverage. Citations use explicit citation annotations, not ordinary links found in answer text.
Frequencies and historical comparisons
Counts such as retrieved in 3 of 5 describe only the collected sample. Unavailable states are displayed separately. Source lists are alphabetized rather than presented as rankings. Comparable history uses the same question wording, supplied location and observation environment version. Daily charts may contain different question sets and cannot alone establish improvement.
Recommendations are separate from findings
Each recommendation identifies the observed evidence, a detected site fact and an inference or suggested review. No individual suggestion guarantees retrieval or citations. Repeatedly retrieved sources may be publishers or directories rather than direct business competitors.
Private evidence and future environments
Normalized observations use a versioned schema. Raw responses and usage details are retained privately in encrypted storage for audit and reprocessing. The backend adapter can be replaced without redesigning the report. The current environment is labeled RankGen Retrieval Network; this does not claim a representative distributed panel.
Current limits
Live observations require the observation service to be configured. Free sessions allow up to five attempts, with additional daily abuse limits. Profile extraction and question editing remain available when observations are disabled. Scheduling, representative industry benchmarks and paid plans are not enabled.