Choose the right resource
#Analytics endpoints return the same selected reporting methodology as the platform, with its measurement label. Raw answers retain captured text, collection surface, quality, outcome, observed time, and provenance where available. Their row count can differ from the observations included in a canonical report.
Use /answers for raw evidence, /answers/{answerId} for an observed capture, and its citations or mentions endpoints for supporting detail. CSV export returns raw answer history in the selected window. A latest persisted perception assessment is a Searchly assessment, separate from an observed answer capture.
Read the metric guide for denominators and the answer guide for collection context. Do not treat every empty response as a measured absence.
Traffic and source freshness
#GA4 and Search Console reports require connected, configured sources. Responses retain receipt coverage and the exact connected generation. A successful HTTP response does not imply fresh provider data. GA4 breakdowns expose bounded top lists.
Agent Analytics returns crawler evidence, Page Records, and recheck progress. Retained events describe recorded requests, not a guarantee that every request or crawler identity has been independently verified. Optimization proposals retain their frozen evidence and decision safeguards.
Page Impact evidence
#Call GET /projects/{projectId}/agent-analytics/evidence?date=YYYY-MM-DD for one explicit UTC evidence day. It joins observed owned-page citations, host-bound AI referrals, and Search Console pages using the project's saved website and the current source generations.
sources includes source totals, returned counts, truncation, receipt dates, and answer-cell coverage. absenceEvidenceAvailable identifies when a missing per-page measurement can be read as zero. Incomplete answer coverage or a truncated source list keeps missing cells null.
summaryEvidence.totalCitations.meaning is measured_total, observed_lower_bound, or unavailable. A positive partial count remains a lower bound. Zero citations stay null until the expected answer cells are fully covered. A complete cohort's total can remain measured even if the bounded URL list is truncated; missing per-page cells still remain unknown.
The result's total counts joined available rows; individual source totals remain separate. The url_path association does not establish that a particular answer caused a visit.