Where did the buyer discover the brand? Which pages and evidence moved the decision forward? Did the contact become a qualified opportunity? SEO/GEO measurement often breaks because search data ends at clicks, analytics ends at forms and CRM begins with a name. The goal is not to award all value to the last click, but to preserve a reviewable journey and lead-quality record.
Capture source/medium, campaign parameters, referrer, first landing page, language and device class within privacy requirements. Match search-query data to landing pages. AI assistants may provide identifiable referrals or appear as direct traffic, so direct traffic is not automatically AI traffic. Keep crawler logs separate from user sessions.
Record service, case and insight views, downloads, email clicks and contact-page arrivals. Scroll depth alone is not purchase intent. Stronger signals are sequences such as moving from a problem article to the relevant service page and reviewing proof before contacting the company. Define the service role and intended next step for every content cluster.
Pass the original source, landing page and key content path into the enquiry record. Sales then adds market, need, budget range, decision window, service fit and next stage. Marketing and sales must agree a qualified-enquiry definition; raw submission count is not a quality metric.
B2B journeys can span brand search, AI answers, Reddit threads, media coverage and direct return visits. Preserve first identifiable touch, last non-direct touch, assisting pages and sales-reported source, while leaving unknowns as unknown. A buyer saying ‘I found you in ChatGPT’ is useful self-reported evidence, but it should remain distinct from verified referral data.
A practical event set includes service_view, case_view, insight_to_service, contact_view, email_click, form_start, form_submit, qualified_lead and opportunity. Define triggers, parameters, owner and QA method. Lead fields should preserve original source, landing page, language, service, content cluster, submission time, quality status and sales notes. Stable naming matters more than event volume.
Weekly reporting checks data quality, anomalies and paths; monthly reporting compares cluster impressions, visits, service-page progression and qualified enquiries; quarterly review combines content, channel and sales feedback. If impressions do not lead to service views, inspect intent and linking. If service views do not lead to contact, inspect proof and CTA. If forms produce weak leads, revisit targeting, promises and qualification.
Can every AI-driven enquiry be tied to a specific answer?
Usually not. Combine identifiable referrals, tagged links, landing paths and self-reported source into evidence levels, and preserve the unknown share.
Is GA4 alone enough?
GA4 can measure site behaviour and some attribution, but search queries, crawler access, CRM lead quality and sales stages require additional sources.
How do we attribute email enquiries without a form?
Track the mailto click and source context, then confirm the source during qualification. A click is not an enquiry; inbox and sales records provide the outcome.
How often should attribution be reviewed?
Check data quality weekly, content and lead quality monthly, and sales-cycle or channel-allocation decisions quarterly.
Minimum viable events and fields
- Acquisition source, landing page and language
- Meaningful engagement with service, case and insight pages
- Email clicks, WeChat copies, form starts and submissions
- Target market, requested service, budget stage and timing
- Qualified lead, meeting-ready lead, opportunity and won status
- Specific sales reason for disqualification
Email and WeChat conversations often move outside the browser, so perfect single-touch attribution is unrealistic. Preserve first touch, last touch, landing page and the buyer's stated source, then let sales validate the need. For AI discovery, classify evidence as tagged direct visit, referral visit, buyer-reported discovery or unverified.
End reporting with a decision, not a traffic chart
A monthly review should identify which topics attract qualified visits, which pages precede contact, which enquiries sales accepts, and what to scale, repair or stop next. When evidence is limited, state the sample constraint rather than presenting correlation as causation.
Should AI discovery still be recorded when the platform cannot be verified?
Yes, but label it as buyer-reported or unverified and keep it separate from confirmed referral traffic. Over time, compare question themes, landing pages and lead quality without overstating causality.
Source：Google Analytics：Traffic acquisition report