Keep four signals separate
A citation is a link or source reference observed in an answer. A referral session is a visit your analytics system attributes to an external source. A crawler request is a machine fetching a page. A conversion is a defined action such as a completed signup. These events can occur independently: a citation can receive no clicks, and a visitor can arrive after copying a URL without retaining an identifiable referrer.
Start by writing down which question each metric answers. For a creator selling a course, useful questions include whether relevant pages receive visits, whether visitors read the material, and whether those sessions lead to a purchase or enquiry. Reporting a large crawler count as interested human traffic would obscure those questions rather than answer them.
| Signal | Evidence | Does not prove |
|---|---|---|
| Referral session | Analytics source and landing page | Every citation or original prompt |
| Observed citation | Dated answer with a visible source URL | Stable ranking across users or future answers |
| Crawler request | Access log with verified crawler identity where possible | Indexing, recommendation, or a human visit |
| Conversion | A defined completed event | That AI exposure alone caused the decision |
Inspect session sources and landing pages
In Google Analytics 4, start with Reports, Acquisition, Traffic acquisition and the Session source / medium dimension. Look for sources actually present in your data rather than assuming a prewritten list captures every AI service. Sources such as chatgpt.com or perplexity.ai can be useful candidates, but referral domains, mobile apps, redirects, consent, and browser behavior can change what is observable.
Use a landing-page exploration or an appropriate secondary dimension to identify the pages receiving those sessions. Keep the reporting period, timezone, and filters consistent. Separate first-user acquisition from session acquisition: someone first discovered through a social post may later return through an AI answer. Those dimensions answer different questions and should not be mixed into one apparently precise attribution claim.
- Confirm the analytics property and reporting timezone.
- Inspect actual Session source / medium values.
- Group only verified candidate AI referral sources and retain the raw values.
- Break down landing pages, engaged sessions, and defined key events.
- Export a dated baseline and compare like-for-like periods.
Google AI search is part of broader search reporting
Google says sites appearing in AI Overviews and AI Mode are included in Search Console's overall Web search traffic. A change in organic clicks is therefore not, by itself, evidence of a change in AI citations. Use Search Console to inspect pages and queries, and report the measurement scope accurately rather than relabeling all Google organic visits as AI traffic.
The same Google guidance says ordinary SEO fundamentals apply and no special AI schema or new machine-readable file is required. A useful measurement plan consequently begins with accessible, indexable pages and relevant content. Sitemap submission, indexing, a visible answer citation, and a subsequent visit remain separate milestones. Record what is observed without promising the next stage.
Maintain a small, reproducible citation sample
Choose a fixed set of questions your audience genuinely asks, such as a workflow comparison or a platform limitation. Record the exact question, service, date, locale, whether browsing was enabled, and cited URLs. Preserve a screenshot or answer export where available. Mark this as a sample of observed answers, not a universal rank tracker or an estimate of market-wide exposure.
Repeat the same questions at a sensible interval and note wording changes. A single favorable answer can be useful qualitative feedback but is a weak trend. If you use an outside monitoring vendor, ask how it selects prompts, handles localization and personalization, and distinguishes mentions from linked citations. Comparing differently sampled reports can manufacture movement that never happened for a consistent audience.
Connect visits to outcomes without overstating attribution
Choose one or two meaningful key events and confirm they actually fire at completion. A click on Start trial is not the same as an account created, and an account created is not a paid subscription. Review conversion counts alongside their denominator, especially when the source has only a handful of sessions. Use aggregate, consent-respecting reporting rather than collecting chat prompts or personal data you do not need.
Keep an annotation log for article publication, major corrections, pricing changes, and tracking changes. Those events help explain a shift without proving causation. Do not put campaign tags on ordinary internal links to make a channel look attributable; reserve campaign tagging for links you distribute and control. This article is a measurement method, not a report of Ferryman's current AI traffic or a claim that Ferryman includes website-attribution analytics.
Sources & further reading
Written by Ferryman. Sources checked Oct 17, 2026. Platform rules can change; consult the linked documentation for current requirements.