{
  "slug": "ai-referral-source-split",
  "name": "AI referral source split",
  "aliases": [
    "AI assistant mix",
    "Referral split by assistant"
  ],
  "definition": "How assistant-referred sessions divide across the individual assistants, which answer surface is actually sending the traffic.",
  "category": "AI referral traffic",
  "type": "share",
  "unit": "percent of assistant-referred sessions attributed to each traffic-source value",
  "direction": "neutral",
  "verification": "review",
  "aggregation": "A composition of one period's sessions across traffic-source values. The parts sum to the selected set, not to all assistant traffic, because unclassified and referrer-stripped visits sit outside it, and because the sessions measure is a distinct-count sketch, the parts need not add to the whole-domain total either.",
  "grain": "One traffic-source value within one date range, as a share of the selected set.",
  "dimensions": [
    "source/medium",
    "landing page",
    "country",
    "device"
  ],
  "requiredFilters": [
    "date range"
  ],
  "sources": [
    "web-analytics"
  ],
  "reports": [
    "ai-referral-traffic-report"
  ],
  "skills": [
    "search-to-revenue",
    "ai-citation-monitoring",
    "cross-domain-correlator"
  ],
  "verbs": [],
  "rungs": [
    "R5"
  ],
  "levers": [
    "L7"
  ],
  "questions": [
    "Which AI assistant sends us the most visits?",
    "Is our assistant traffic concentrated in one surface?"
  ],
  "caveats": [
    "The denominator is whatever set of traffic-source rows was selected by hand, so two people can produce different splits from the same data.",
    "The default breakdown ranks a top-N by volume, so a small assistant can fall outside it, but a named source can be requested explicitly, and any requested value that matched nothing is echoed back. A missing row is recoverable, not a dead end.",
    "This split is not comparable to the per-engine cut on the AI-visibility side: that one is scoped to a tracked prompt basket, this one to whatever referrers the analytics property recorded."
  ],
  "freshness": "loads nightly, a same-day question gets yesterday's number, labelled as yesterday's.",
  "failureModes": [
    "Reading a missing assistant as zero traffic when it fell outside the top-N, instead of requesting it by name.",
    "Lining this split up against the per-engine citation split and expecting the shares to correspond."
  ],
  "notSameAs": [
    {
      "slug": "ai-citation-category-mix",
      "why": "Both are compositions on the word 'AI', and they describe different populations on different datasets. The citation category mix divides your citations across content categories inside a tracked prompt basket at R3; this divides arriving sessions across referring assistants in web analytics at R5. Neither is a view of the other, and no row in one corresponds to a row in the other."
    }
  ],
  "related": [
    "ai-referral-sessions",
    "ai-referral-conversions",
    "sessions"
  ],
  "workflows": [],
  "lastVerified": "2026-08-04"
}