THE SHORT ANSWER
AI referral tracking connects an observed source with a site visit, a delivered inquiry, and a qualification decision. Define each stage separately. Keep the underlying counts and unknowns visible so your team can act without claiming more attribution than the data supports.
Define what AI referral tracking should tell your team
AI referral tracking helps you understand what happens after someone arrives from an assistant. Start with the business question: do these visits lead to relevant conversations? Then define the events needed to answer it.
A page view, a contact-button click, and a delivered inquiry describe different actions. Give each a clear name and meaning. The person reading the report should know exactly what happened when a count increased.
Write this event dictionary before configuring a dashboard. Include the trigger, responsible system, deduplication rule, and owner. Our guide to measuring AI search visibility explains how referral data fits alongside mentions and citations.
Identify observed sources without losing their context
In GA4, begin with the Traffic acquisition report. Its session source and medium dimensions describe the source associated with a session. Keep those fields available when reviewing any aggregated channel total.
Google’s current default channel definitions include an AI Assistant channel. It uses recognized assistant referrers and the ai-assistant medium. Google’s AI Overviews and AI Mode remain within Organic Search rather than that assistant channel.
Inspect your own property’s data and reporting period before building filters. A custom definition or an older export may classify sources differently. Record which rules your report uses, and investigate unfamiliar domains before adding them to an AI category.
Some journeys have no usable referral signal. A person may copy a URL, switch devices, or return later. Keep these unknowns visible. A missing source does not establish that AI influenced the visit.
Measure a completed handoff at the right point
If your website opens an email draft, record that action as a draft or contact-link interaction. The browser cannot confirm that the person sent the email. Use the receiving mailbox or another confirmed record to establish that an inquiry arrived.
For a directly delivered form, define success around the receiving system’s accepted response. Prevent repeat clicks or page reloads from creating duplicate leads. Keep an inquiry identifier in the operational system so delivery and qualification can be reconciled.
Google documents separate generate_lead and qualify_lead events. Map them to your real business stages. Do not trigger both automatically when someone opens a form.
Send only the data your measurement setup needs. Keep names, email addresses, and message bodies out of routine analytics event parameters. Store inquiry details in the appropriate operational system with suitable access controls.
Agree how a delivered inquiry becomes qualified
Ask the team handling new business to define qualification. Useful criteria could include a relevant need, a suitable organization, and an agreed next step. Document the criteria so different reviewers make comparable decisions.
Use a small set of statuses that someone will maintain. For instance: awaiting review, qualified, unsuitable, duplicate, and spam. Record the reason for an unsuitable inquiry. Those reasons can reveal whether a page attracts the wrong audience or sets unclear expectations.
Add an optional discovery question to the inquiry process, such as “Where did you first hear about us?” Store that answer separately from the observed referrer. A buyer’s recollection and a session record describe different parts of the journey.
Calculate rates from visible counts and matching periods
Consider a hypothetical reporting cohort with 200 observed assistant sessions, 10 delivered inquiries, and four qualified inquiries. The session-to-inquiry rate is 5%. The inquiry-to-qualified rate is 40%. The session-to-qualified rate is 2%. These are illustrative numbers, not Citevia client results.
Show the counts beside each percentage. Also state the period, qualification definition, and attribution rule. Four qualified inquiries cannot support the same confidence as hundreds, even when the displayed rate looks strong.
Allow time for qualification. An inquiry received on the last day of the month may still await review. Compare mature cohorts or identify pending records explicitly. Keep the date of the visit and the date of qualification available.
Avoid multiplying conversion rates from unrelated cohorts. A visit-based report and a CRM report may use different time windows or identity rules. Reconcile their definitions before combining them into a funnel.
Use landing-page intent to explain the outcome
Group visits by the kind of page people entered. Someone reading a beginner’s guide may need a related explanation. Someone viewing an implementation comparison may be ready to discuss requirements. Their next steps should reflect those different needs.
When visits produce few inquiries, inspect the offer and contact path. When inquiries rarely qualify, inspect the audience statement and claims. Our comparison-page framework can help a decision page set clearer expectations.
Look at trends alongside sales feedback and known site changes. A change in observed source does not prove that the source caused a change in revenue. Keep channel reporting useful by describing the evidence and the remaining uncertainty.
Turn the report into a monthly action list
Review one question each month: what would improve the next relevant conversation? Choose a specific action, such as clarifying an integration requirement or shortening an unnecessary form. Assign an owner and record when the change goes live.
Keep a short log of tracking changes as well as content changes. Otherwise, a corrected event can look like a sudden growth improvement. Use the same definitions until there is a clear reason to revise them.
Start with reliable source fields, confirmed delivery, and a maintained qualification decision. Then add detail where it changes an action. Our methodology connects this measurement work with content and technical improvements across the site.