How it worksReadiness scoreWhy PingPlusLearnBlog
Get started freeLog in
Skip to guide

Practical guide · Part 03

Track AI Referral Traffic in GA4 and Your CRM

To find AI referral traffic in GA4, start with Reports → Acquisition → Traffic acquisition, inspect the session source, then follow the landing page and relevant key event. Use the steps and linked records below to connect observable visits to accepted enquiries, bookings, or orders in your CRM.

PingPlus Editorial TeamReviewed

Find AI referral traffic in GA4 and validate the report

Use Traffic acquisition for the visit's source, inspect landing pages and relevant key events, then check customer quality in the CRM.

  1. Choose the scope. Open Reports → Acquisition → Traffic acquisition in the correct GA4 property. Confirm the date range and property time zone. Google notes that report navigation can differ if a collection was customized.
  2. Inspect the channel. Use Session default channel group and select AI Assistant where recognized traffic is present. Switch to Session source / medium to inspect the underlying rows. Do not substitute First user source for a question about this visit.
  3. Check the destination. Use a report or exploration with Landing page + query string to see which offer or guide people entered through. Preserve the same source filter and period.
  4. Review meaningful actions. Compare sessions, engaged sessions, and the relevant key event, such as a correctly configured lead submission or purchase. Confirm which event the report counts; a page view or form start is not an accepted customer outcome.
  5. Follow outcomes. Reconcile those actions with accepted enquiries, bookings, orders, and later statuses in the business system. Use a supported join instead of expecting a channel report to establish CRM qualification.

Google documents the dimensions and scope in its Traffic acquisition guide. If AI Assistant is empty, inspect the source rows and actual outbound journey before creating custom rules. An empty row can reflect limited traffic or missing source evidence. It is not a reason to classify all Direct, Google, or Bing traffic as AI.

Save the report with its period, time zone, session-source filter, landing-page view, and selected key event. This establishes the analytics view. The next example shows the separate first-party records needed to verify whether an action became an accepted booking.

Follow one visit through submission to a CRM booking

A visit identifier connects the submission to its arrival record. A request event identifier connects the booking to that submission. Check both links before assigning the booking to a source.

Worked record join · fictional identifiers

Follow the IDs from a visit to an accepted booking
  1. 09:00 UTC · Observed visit

    V-041
    referrer
    chatgpt.com
    landing page
    /services/apartment-cleaning
  2. 09:04 UTC · Booking request

    E-041
    visit_id
    V-041
    offer
    CLN-98101
  3. 10:00 UTC · CRM booking

    B-041
    request_event_id
    E-041
    status
    accepted_eligible
?

E-042 → B-042: source unknown. The request-to-booking link exists, but no visit ID or usable source evidence was retained. Keep the source unassigned.

Join visit V-041 to request E-041 and accepted booking B-041 using retained IDs; leave E-042 and B-042 unassigned when source evidence is missing.

These are illustrative identifiers supplied by an integration, not automatic GA4 fields. An accepted booking is not yet a completed job.

This small teaching example uses fictional records from one merchant's own site and CRM. The implementation retains the identifiers with the required permissions. The field names illustrate a record design; they do not describe a PingPlus API or a universal analytics setup.

Three records for one linked booking, November 15, 2026
RecordRecorded identifiers and factsWhat the link establishes
Arrival at 09:00 UTCvisit_id V-041; recorded referrer host chatgpt.com; landing page /services/apartment-cleaningThe merchant observed this visit arriving from the recorded referral source.
Request at 09:04 UTCevent_id E-041; visit_id V-041; event booking_request; offer CLN-98101The request carries the same V-041 identifier as the arrival record.
CRM booking at 10:00 UTCbooking_id B-041; request_event_id E-041; status accepted_eligibleThe business accepted the booking connected to E-041; E-041 connects back to V-041.

The join is V-041 → E-041 → B-041. Under a documented rule that credits the observed arrival for a request in the same visit, the business can assign this accepted booking to that referral. Acceptance is not completion: update B-041 when the job is completed, cancelled, or otherwise changes status. A source label does not reveal the customer's private conversation or prove that the visit created an incremental customer.

Keep a booking with a missing arrival link unassigned

A second submission, E-042, creates booking B-042. The CRM retains E-042, so the team can connect the request to its booking. But no visit identifier or usable source evidence was retained. The source remains unknown; do not attach it to V-041 because the timestamps or selected offer look similar. If the customer separately reports using AI, keep that self-reported evidence in its own field.

Test the two joins independently. If E-041 is sent twice, it should not create two customer outcomes. If the action-to-CRM link exists but the arrival link is missing, retain the valid business outcome while leaving its source unassigned. Use the identifiers your integration actually supplies; this example does not establish tracking across devices or separate visits.

Join source evidence to a business record

Use permitted first-party records and stable outcome identifiers to connect the journey. Avoid storing sensitive customer content in URLs or analytics fields.

A minimal measurement record
Record groupExample fieldsPurpose
ArrivalLanding page, timestamp, observed source/medium, available referral or campaign contextDescribe the visit you can actually observe.
CampaignCampaign ID and supported click or interaction identifierReconcile with the actual campaign integration, when available.
Customer actionEvent name, event time, non-sensitive event ID, selected offerSeparate a visit from a submitted action.
Business outcomeLead, booking, or order ID and current statusCheck completion, qualification, cancellations, and duplicates.
Attribution decisionApplied rule, window, evidence type, and unresolved reasonExplain why a result receives credit or stays unassigned.

These are proposed measurement fields, not a promised PingPlus API schema. Use the identifiers your real integration supplies. Keep access, retention, and permissions aligned with the data collected. Never put names, email addresses, medical details, or private conversation transcripts into tracking URLs.

For links you control, Google documents campaign URL parameters such as source, medium, and campaign. Apply a consistent naming scheme and test that redirects preserve it. Do not add campaign tags to internal navigation or invent tags for organic links you do not control.

Decide what you can observe

Distinguish a visible AI referral, a tagged paid campaign, and a customer who reports using AI. They are different kinds of source evidence.

A shopper might click a link from an assistant, return later through search, and buy directly. Another customer may discover the product through an AI shopping channel and complete checkout without visiting your normal storefront. Your measurement needs to describe these paths without claiming an identifier or browser session that does not exist.

Google Analytics now documents an AI Assistant default channel group for recognized assistant traffic, excluding Google's AI Overviews and AI Mode. Those Google search experiences remain part of Organic Search. Inspect source and medium alongside channel grouping; a channel label is not proof of a particular conversation or prompt.

When no reliable source evidence is present, preserve “unknown” or the observed classification. A direct session is not automatically an AI session. A voluntary “How did you hear about us?” answer can supplement the record, but should stay separate from observed referral data.

Test the whole path before trusting the report

Run a supported test journey from arrival through the final business state, then inspect each record at the same scope.

Follow a real assistant link that can open your destination. Use a fresh test session, inspect redirects, and complete one clearly marked test action. Use Realtime to check activity and DebugView with debug mode enabled to inspect collection. Realtime and DebugView have limited attribution analysis; confirm the final session classification in processed Acquisition reporting.

Test journey: expected evidence at each step
StepSuccess conditionIf it does not pass
ArrivalThe intended page loads and the actual available referral or campaign evidence is retainedInspect the real link and redirects. Missing source evidence stays unknown; do not invent a source.
Action collectionOne intended submission produces one identifiable event with the correct offerInspect form handling, confirmation-page triggers, and retries. Keep tests marked.
Business creationThe request creates the expected CRM lead, booking, or order with its request identifierCheck the form-to-business-system integration before calling the analytics event an outcome.
JoinsArrival → action and action → business record both resolve to the intended recordsFind which identifier was lost; do not substitute timestamps or customer names.
ReportingProcessed session reporting reflects the evidence available for that visit, and the outcome report shows the correct statusCheck scope, time zone, processing state, and attribution rule. Record unresolved differences.

Compare sessions with sessions and outcomes with outcomes. A first-user source describes initial acquisition; a session source describes that visit. An attribution report may assign credit under another scope and rule. Document which question each report answers. Test later cancellations and refunds using the outcome validation cases.

Reconcile a small referral cohort

Show the joins and gaps explicitly. Traffic, attributed actions, and qualified outcomes should remain separately countable.

In this synthetic example, a service business observes 60 sessions from recognizable assistant sources in one review period. Six booking requests have a permitted first-party link to those visits. Four become accepted, eligible bookings; one is a duplicate and one is outside the service area.

Synthetic referral review
MeasureCountInterpretation
Observed assistant sessions60Visible traffic under the selected session definition.
Linked booking requests6A 10% request rate in this small cohort, not a market benchmark.
Accepted eligible bookings4A 6.7% session-to-accepted-booking rate, rounded.
Other bookings with no usable source3Remain unassigned; do not add them to the AI cohort.

The four accepted bookings are attributable under the stated method. This does not prove four incremental customers who would otherwise never have booked. A causal incrementality claim needs a suitable comparison or experimental design.

Keep outcome windows clear. A booking created at the end of the period may be accepted later. Either allow time for the cohort to mature or label incomplete outcomes as pending. Otherwise a weekly comparison can confuse reporting delay with deteriorating quality.

Use gaps to choose the next improvement

Fix missing joins or poor customer fit before increasing acquisition spend. A larger traffic count will not repair unreliable measurement.

If source information disappears after a redirect, inspect the redirect and destination implementation. If requests arrive but cannot be connected to CRM records, fix event identifiers or the agreed reconciliation process. If the link is reliable but many requests are unsupported, correct the offer or qualification path.

For PingPlus, compare the selected recorded campaign event with the business system that owns the outcome. Record any difference and its explanation. A billable click and an accepted booking can both be accurate while describing different stages.

Keep a concise measurement note alongside the report: period, scope, attribution rule, unresolved traffic, and outcome maturity.

Recurring source-to-outcome report
Report layerFields to retainQuestion it answers
AcquisitionPeriod, session source / medium, landing page, sessions, and relevant key eventsWhich observable sources and destinations produce activity?
Business qualityJoined requests, accepted results, pending results, completions, and missing-source countWhich activity becomes a useful result, and where is evidence missing?
Visibility, kept separatelySaved questions, mentions, and citations from the auditHow is the business described in the sampled answers?

An assistant crawler request is not by itself a human referral, citation, or customer. Keep answer observations in the visibility audit worksheet. Review outcome reconciliation before using the customer report to expand a campaign.

FAQ

Should I create my own AI traffic channel in GA4?

First inspect the current default grouping and source/medium data. GA4 now documents an AI Assistant group. A custom group can still be useful for a different reporting requirement, but preserve the underlying evidence and avoid assuming every assistant visit is classified perfectly.

Can I count direct traffic as AI traffic?

No. Direct traffic alone doesn't tell you that a visitor came from AI. Leave the source unknown unless you have supporting evidence. You can record a customer's own account of how they found you separately from the observed session source.

Will browser analytics capture every AI shopping order?

Not always. Some AI shopping checkouts may never load your storefront in a browser. Compare orders with the commerce platform's available records and channel information, and keep those orders separate from measured website sessions.

Does an attributed conversion prove incremental revenue?

No. Attribution applies a credit rule to observed interactions. Incrementality asks what would have happened without the activity and needs an appropriate comparison. Report those conclusions separately.

Does an event appearing in DebugView confirm its final acquisition source?

No. DebugView helps you check that events and parameters arrived; it doesn't confirm the final acquisition source. Check the processed Acquisition report, and keep session source, first-user source, and attribution-report data separate.