Five events have different denominators
Citations occur in platform answers; visits occur on the site; inquiries arrive through forms, email or other channels; qualification requires a business judgment; revenue requires commercial verification. Citation counts cannot be treated as potential buyers.
Someone may read an AI answer, search the brand and later visit directly. Direct traffic does not exclude AI influence. A voluntary “AI recommended you” response likewise does not prove a traceable click. Preserve distinct evidence classes and unknowns.
Define a minimal attribution record
| Field | Evidence | Invalid inference |
|---|---|---|
| Source class | Observable referrer or permitted tag | Every AI exposure leaves a referrer |
| Self-report | Voluntary customer response | A corresponding click is verified |
| Session ID | Necessary, permitted association | Identity across all devices |
| Inquiry ID | Deduplicated business record | Qualified or sold |
| Qualification | Criteria, reviewer, status | Pending means unqualified |
| Revenue | Verified business record | AI caused all revenue |
Avoid fingerprinting or unnecessary personal-data joins to improve apparent attribution. Collection, purpose, retention and access must respect applicable requirements and user choices. The public example contains only synthetic records.
Recalculate the three-record fixture
assessment-results.json contains an AI-tagged visit with an inquiry awaiting qualification, a direct visit with self-reported AI influence whose inquiry is unqualified, and a search visit without an inquiry.
There are two inquiries, zero confirmed qualified leads and one pending qualification. Revenue and causal contribution remain null. Pending cannot become qualified, and self-report cannot become a verified AI click. No real customer, traffic or new commercial outcome is represented.
Associate questions and pages cautiously
A question, landing page, content revision and inquiry topic can form a candidate relationship. An inquiry about a newly documented restriction suggests an analysis path, not proof that a specific AI answer caused the deal.
State unknown paths and declare first-touch, last-touch or other attribution rules. Different models can produce different totals; do not present only the favorable one. Causality needs controls or another defensible design because seasonality, sales activity and brand news can also change outcomes.
Define qualification before inspecting results
Relevant criteria might include market, purchasing role, explicit need, budget or delivery fit. They must suit the business and precede result review. Missing budget information is not necessarily disqualification, and competitor research is not automatically buying intent.
Record pending and rejected reasons. Compare review completion and criteria across periods: faster processing can increase confirmed leads without improving acquisition. Revenue also depends on sales-cycle timing and repeat customers, not merely the latest article.
Zhihe Growth's reporting responsibility
Report inspectable implementation and platform observations first. Discuss lead quality or revenue only when clients supply appropriately authorized business data. SuperPDR 17% and ELEREIN 15% are historical official-site citation rates, not inquiry, conversion or revenue-growth rates.
Public reports can show layered counts, unknowns and limitations while personal records remain controlled. Reliable engineering means traceable evidence, not attributing every sale to AI optimization.
Reporting exercise: describe an incomplete business chain
| Observation | Supported wording | Unsupported wording | Needed evidence |
|---|---|---|---|
| Official link in answer | Answer cited the site | Acquired a customer | Visit or business record |
| AI-referral visit | Observed referral | Came from a tested prompt | Verified association |
| Self-reported AI | Customer reported influence | Verified platform click | Keep self-report class |
| Form submitted | Deduplicated inquiry | Qualified opportunity | Business review |
| Qualified lead | Met declared criteria | Sale completed | Contract or revenue |
| Sale after redesign | Occurred afterward | Caused by redesign | Defensible attribution |
Deduplicate repeated email, phone and messaging contacts without collecting unnecessary sensitive data. Unknown associations can remain unknown. Platform observations, website behavior and business review should be separate tables linked only where evidence permits; their totals may cover different populations and periods.
Materials and reference
Download the synthetic attribution records and blank review form. Read controlled evaluation, observability and Google's AI optimization guidance. The material is not measured commercial performance or a revenue promise.