AI Session Attribution: Empowering Legal Firms with Accurate Insights

artifact: intake_attribution_capture
label: "Template YAML Policy (TEMPLATE) — AI Referrer + UTM Capture for Intake"
owners:
  businessOwner: "Chief of Staff / Analytics"
  technicalOwner: "IT Director"
  stakeholderApprovers:
    - "Director of Operations"
    - "Managing Partner (Intake Oversight)"
regions:
  dataResidency:
    primary: "US"
    allowed: ["US"]
collection:
  eventName: "intake_first_touch"
  captureOn:
    - "first_pageview"
    - "pre_form_render"
  fields:
    required:
      - visit_id
      - timestamp_utc
      - landing_path
      - referrer
      - user_agent
      - utm_source
      - utm_medium
      - utm_campaign
    optional:
      - utm_content
      - utm_term
      - gclid
      - msclkid
normalization:
  aiReferrerMatchers:
    - engine: "chatgpt"
      contains: ["chat.openai.com", "chatgpt.com"]
    - engine: "claude"
      contains: ["claude.ai"]
    - engine: "perplexity"
      contains: ["perplexity.ai"]
    - engine: "gemini"
      contains: ["gemini.google.com"]
    - engine: "copilot"
      contains: ["copilot.microsoft.com", "bing.com"]
    - engine: "you.com"
      contains: ["you.com"]
    - engine: "kagi"
      contains: ["kagi.com"]
    - engine: "arc_search"
      contains: ["arc.net"]
qualitySLOs:
  utmCompletenessRate:
    target: 0.85
    alertBelow: 0.70
  aiReferrerClassificationRate:
    target: 0.90
    alertBelow: 0.80
  visitIdPropagationRate:
    target: 0.95
    alertBelow: 0.85
security:
  piiPolicy:
    storeIP: false
    redactQueryParamsExcept: ["utm_source","utm_medium","utm_campaign","utm_content","utm_term"]
    storeFreeTextFormFieldsInAnalytics: false
  accessControl:
    rbacRolesAllowed: ["Analytics","IT","Ops"]
    denyRoles: ["Associates","Paralegals"]
audit:
  logRetentionDays: 365
  promptLogging:
    enabled: true
    storePrompts: "hashed_only"
approvalWorkflow:
  steps:
    - name: "Instrumentation review"
      approverRole: "IT Director"
    - name: "Intake process review"
      approverRole: "Director of Operations"
    - name: "Risk & confidentiality review"
      approverRole: "Managing Partner"
thresholdActions:
  onSLOBreach:
    - "create_ticket_in: Jira"
    - "notify_channel: #intake-analytics"
    - "rollback_to_last_known_good: true"
notes:
  - "Adjust thresholds per org risk appetite; values are illustrative."
  - "Use a first-party endpoint (or server-side tag) so referrer/UTMs are captured before form vendors redirect."

{
  "title": "AI Session Attribution: Empowering Legal Firms with Accurate Insights",
  "published_date": "2026-05-14",
  "author": {
    "name": "Matthew Charlton",
    "role": "Founder & CEO",
    "entity": "DeepSpeed AI"
  },
  "core_concept": "AI Search Optimization (GEO, AEO, SEO, SXO)",
  "key_takeaways": [
    "If you can’t see AI-engine referrals and UTMs, you can’t tie GEO/AEO work to consult requests—especially as 40%+ of discovery shifts into AI assistants.",
    "A workable attribution stack is simple: capture referrer + UTMs at session start, persist an anonymous visit ID, then join to intake outcomes (consult form, call, email).",
    "DeepSpeed AI Analytics Dashboard adds citation + competitor monitoring across 12+ AI engines, with owned data (your Firebase, your code, your analytics)."
  ],
  "faq": [
    {
      "question": "Which AI engines should we track for referrals in 2026?",
      "answer": "At minimum: ChatGPT, Claude, Perplexity, Gemini, and Copilot. Most teams expand to DeepSeek, Grok, Meta AI, Kagi, Poe, You.com, and Arc Search to catch long-tail discovery."
    },
    {
      "question": "Do UTMs still matter if referrers identify AI assistants?",
      "answer": "Yes. Referrers tell you the engine; UTMs tell you the campaign and content intent. Together they let you compare ‘thought leadership automation’ pushes vs evergreen pages under the same AI source."
    },
    {
      "question": "How does this connect to delivery outcomes like faster document processing?",
      "answer": "Attribution tells you which inbound topics create document-heavy matters. Delivery automation (document and contract intelligence with human review) is then prioritized where demand and margin pressure are proven."
    }
  ],
  "business_impact_evidence": {
    "organization_profile": "HYPOTHETICAL/COMPOSITE: 75-attorney business law practice (20–200 attorney segment) with web intake + scheduler + lightweight CRM; high volume of document-heavy matters.",
    "before_state": "HYPOTHETICAL: 40–55% of consult submissions attributed to direct/none; AI-engine referrals not visible; inconsistent UTMs across practice pages and thought leadership; limited ability to prioritize content tied to document-heavy work.",
    "after_state": "HYPOTHETICAL TARGET STATE: AI-engine sources normalized and reported; first-touch UTMs captured at intake start; consult outcomes joinable to AI sessions; competitor recommendations monitored across 12+ AI engines.",
    "metrics": [
      {
        "kpi": "Direct/none share of consult submissions",
        "targetRange": "20–40% reduction",
        "assumptions": [
          "first-touch capture runs server-side or before vendor redirects",
          "UTM taxonomy enforced across top 20 landing pages",
          "visit_id propagation ≥ 85%"
        ],
        "measurementMethod": "4-week baseline vs 6–8 week rollout window; compute share of consult submissions with source=direct/none, exclude weeks with site migrations."
      },
      {
        "kpi": "Consult requests per 100 AI-originated sessions",
        "targetRange": "10–25% increase",
        "assumptions": [
          "AI referrer classification rate ≥ 80%",
          "SXO fixes shipped on top 10 AI landing pages (speed + clarity + CTA)",
          "intake response time SLA defined and met"
        ],
        "measurementMethod": "Baseline: 4 weeks after instrumentation is stable; Pilot: next 6 weeks; use first-touch attribution to count AI-originated sessions and consult outcomes joined by visit_id."
      },
      {
        "kpi": "Competitor recommendation incidents captured (Kira, Luminance, CLM, manual review)",
        "targetRange": "Capture 70–90% of tracked prompt clusters",
        "assumptions": [
          "competitor monitoring enabled across 12+ AI engines",
          "prompt cluster definitions agreed with practice leaders",
          "weekly review cadence established"
        ],
        "measurementMethod": "Create a tracked prompt list and compare detected mentions/citations vs the tracked list each week; measure coverage percentage and time-to-response."
      },
      {
        "kpi": "Document-heavy delivery capacity (associate hours available for higher-leverage work)",
        "targetRange": "Target 20–40% more capacity (downstream initiative)",
        "assumptions": [
          "separate pilot of Document & Contract Intelligence initiated for document-heavy matters",
          "human-in-the-loop review adoption ≥ 70%",
          "clause accuracy monitored and improved iteratively"
        ],
        "measurementMethod": "Time study baseline (2 weeks) vs post-pilot (6 weeks) on selected matter types; track review hours logged vs strategy/drafting hours logged; label as delivery KPI, not marketing KPI."
      }
    ],
    "governance": "Acceptability hinges on confidentiality controls: no storage of free-text intake fields in analytics, IP not stored, query-string redaction, RBAC limiting access to analytics, and retention policies. DeepSpeed AI deployments provide audit trails (event logs, change history), prompt logging in hashed form where relevant, and never train public models on firm data. Data residency can be enforced via Managed Cloud or On-Prem/VPC Private Enclaves."
  },
  "summary": "Struggling with AI traffic data? Discover how accurate attribution using referrer and UTM tracking can enhance your legal intake process and improve leadership decision-making."
}