Our position: honest under-attribution is better than a beautiful fiction. AI discovery often influences a later direct visit or branded search, which means the system should preserve qualitative evidence rather than force every journey into a last-click chart.
What you should leave with
- AI referrals are the visible click subset, not total influence.
- Add self-reported and sales-source evidence.
- Compare high-intent visibility with qualified pipeline trends.
- Do not claim causality from one score increase.
Which AI discovery events are directly observable?
You can observe referrals that include a recognizable source, landing-page sessions, conversions, and downstream CRM outcomes. You can also observe your own controlled prompt results and public citations. Private conversations that produce no click remain outside the site and CRM.
Create a channel grouping for known AI referrals and preserve the landing page and conversion. Avoid treating missing referrers as proof of direct traffic; privacy, apps, and browser behavior can strip attribution.
Store the prompt and citation baseline separately. It explains what the market saw even when the buyer did not click.
How do you capture no-click influence?
Add a short open-text “How did you first hear about us?” field, train sales teams to record AI mentions, tag call transcripts, and ask new customers which tools influenced the shortlist. These signals are imperfect but reveal journeys analytics cannot see.
Do not replace the open text with a leading checkbox that inflates AI. Let the buyer answer naturally, then normalize responses into categories while preserving the original wording.
Review examples monthly. Qualitative evidence can reveal the exact prompt family and competitor that shaped the decision.
What should the attribution model include?
Use four layers: observed AI referrals, self-reported AI discovery, sales-confirmed influence, and market-level correlation between commercial visibility and qualified demand. Keep the layers separate and report overlap instead of collapsing them into a single fabricated total.
For experiments, choose a prompt cluster, complete a small set of fixes, and compare visibility, referrals, self-reports, and pipeline over a defined window. Document concurrent campaigns and seasonality.
The result may support a contribution claim, not exclusive causality. Buyers use several sources before a B2B purchase.
| Layer | Evidence | Main blind spot |
|---|---|---|
| Referral | Known AI source and session | No-click and stripped referrers |
| Self-report | Buyer names AI in open text | Memory and response bias |
| Sales | Call or CRM note | Inconsistent capture |
| Correlation | Visibility and pipeline move together | Confounding factors |
Which KPIs belong in the monthly report?
Report high-intent recommendation coverage, competitor share, citation ownership, factual-error rate, known AI referral conversions, self-reported AI leads, influenced opportunities, completed fixes, and confidence. Separate leading visibility indicators from lagging revenue outcomes.
A visibility score is a leading signal. A qualified opportunity is a business outcome. The report should not imply that one automatically caused the other.
Include the denominator and raw count: three AI-referred demos out of 120 is more informative than a 200% increase from one to three.
Which attribution claims should you refuse to make?
Refuse claims of complete AI traffic, exact no-click reach, guaranteed pipeline from a citation, or causal revenue from one visibility change. Refuse to mix modeled prompt demand with observed site conversions as if both were measured from the same population.
State what the system can see and where the estimate begins. This protects the client and makes future improvements easier to interpret.
A credible conclusion can be modest: “AI became a named discovery source in nine qualified opportunities after recommendation coverage improved across the tracked prompt set.”
Method boundary: Do not report modeled visibility impressions as website sessions, users, or attributable revenue.
Evidence used in this section
Questions that change the decision
Frequently asked questions
Can GA4 show all ChatGPT leads?
No. It can show sessions with recognizable referral data and subsequent conversions. It cannot see private no-click recommendations or every path that later returns through direct or branded search.
Should we add AI to a lead-source dropdown?
Prefer an open-text first-touch question, then normalize the responses. A forced option can introduce bias and hides the actual language buyers use.
What proves an AEO program is working?
A durable improvement in relevant recommendation coverage plus supporting changes in citations, accuracy, known referrals, self-reported discovery, and qualified pipeline. No single metric proves the entire chain.
How long should an attribution test run?
Long enough to cover the sales cycle and normal demand variation. A local service may learn in weeks; enterprise SaaS may require a quarter or longer.
Primary sources and research
Platform documentation supports factual statements. Where we describe an audit method or prioritization rule, that is AnswerMentions' operating judgment and is labeled as such.
- [1]Google Search Central: AI features and your websiteGoogle says AI Overviews and AI Mode can use query fan-out, may surface different links, and include their performance within Search reporting.
- [2]OpenAI: Introducing ChatGPT searchOpenAI says ChatGPT search provides linked web sources and can use third-party search providers and publisher content.
- [3]OpenAI Help: ChatGPT accuracy and citationsOpenAI warns that answers can be wrong or cite nonexistent sources and recommends checking important claims against reliable evidence.
- [4]Don't Measure Once: Measuring Visibility in AI SearchThe 2026 research focuses on the instability of one-time generative-search measurement and supports repeated observations rather than single-answer certainty.