Track AI search through three connected records: an identifiable website visit, a completed enquiry and the sales team's qualification decision. Check the assistant-referral data your analytics property actually holds, test the submission event and retain the source and landing page when the request reaches your customer system.
Ask buyers how they found and evaluated you, keeping their account separate from the referral record. Follow suitable opportunities through proposals and won work using the same enquiry record.
Report visits, enquiries, accepted opportunities and sales as different outcomes. Google AI impressions need their own reporting context. This approach traces the evidence you have; it cannot capture every AI influence or prove that an assistant created an additional sale.
Check what your analytics property already records
Google Analytics introduced an AI Assistant default channel in May 2026 for recognized assistant referrals. Check that channel in your property before following an older tutorial that assumes every AI visit must be found through a custom source filter. Google Analytics release notes
Use Traffic acquisition to inspect sessions, then examine source, medium and landing page. Confirm which date range and filters are active. Google documents how default channel classification works; your report should retain the underlying source fields so a channel label can be investigated. Default channel documentation
A recognized referral is evidence of an identifiable visit. It is not evidence that all visits influenced by an assistant appear in that channel. Referrer loss, consent settings and cross-device journeys can leave gaps.
Test the enquiry event before trusting a chart
Submit an authorized test through the actual form. Check that the success state reflects an accepted submission, that the record reaches the intended destination and that the analytics event is captured as designed.
A button click is not necessarily an enquiry. A thank-you page can be loaded more than once. Agree which event represents completion and test errors, retries and repeated visits so they do not inflate the count.
Keep personal information out of analytics event values. Your customer system can hold the enquiry under its normal access controls; the analytics report needs the measurement fields, not the person's message or email address.
The workflow below is instructional. Verify the event and handoff in your own setup before treating the report as a completed measurement implementation.
Neither path proves that AI created an incremental sale. Compare enquiries, qualified opportunities and won work using their own definitions.
Preserve context when the enquiry reaches sales
For each enquiry, retain the landing page and identifiable source where available. Add a separate field for the buyer's own account of discovery. “A colleague recommended you” and “we looked you up in ChatGPT” can both be true.
A Salesforce consultancy needs enough project context to decide whether the request fits its delivery work. An equipment supplier needs enough application or model information to assess suitability. Do not label every submission a qualified lead.
| Field | Why it matters |
|---|---|
| Enquiry date and stable record ID | Lets you follow the same request over time |
| Identifiable source and landing page | Records the website evidence you actually have |
| Self-reported discovery | Preserves the buyer's account without overriding the source record |
| Service or product requirement | Helps assess fit |
| Qualification decision and reason | Separates suitable opportunities from general contact |
| Later commercial stage | Distinguishes proposal, win and open work |
A lightweight customer relationship management system, or CRM, can hold these records. Choose a setup your team will maintain. A complex integration that nobody checks is a weak foundation for a commercial report.
Report the journey without adding unlike totals
Report identifiable AI-assistant sessions, completed enquiries and sales-accepted opportunities as separate counts. Keep booked orders in their own field. A visit is not a lead, and an enquiry is not a sale. When an opportunity wins, add the outcome to the same record rather than rewriting the original enquiry count.
Check whether your system can reliably connect those particular sessions to submissions. If it cannot, report channel sessions and total enquiries separately, with self-reported discovery as another field. Do not create an association the data does not establish.
Keep the counts separate. The arrows show record progression, not a measured conversion rate or proof of AI influence.
Keep Google AI impressions in their own view
Google's generative AI performance report provides impressions for supported Search features and dimensions such as pages and countries. That is a different view from assistant referral sessions. An impression means the site appeared under the report's counting rules; it is not an enquiry. Search Console report documentation
Use it to investigate which pages are being shown, then review the website and commercial records separately. Do not add an AI-impression subtotal to an overall Search figure that already includes it. Record the report type, period and property when sharing a screenshot.
Make the monthly report useful to an owner
Open with the commercial finding you can support. For example: “Identifiable AI visits increased, but none of this month's enquiries has been accepted by sales.” Then explain the questions that need investigation.
Perhaps the landing page attracts the wrong buyer. Perhaps the form loses project context. Perhaps the opportunity is suitable but still early in a long purchasing cycle. The data should help your team choose a task, not force every uncertainty into a positive headline.
Include a short measurement note: what the source covers, which links are reliable and which outcomes remain open. The audit checklist explains how to report answer visibility alongside this commercial view.
Decide whether the next change is content or measurement
If submissions are not recorded reliably, fix measurement first. If the records are reliable but the enquiries are unsuitable, review the audience and page promise. If suitable opportunities are progressing, preserve the evidence that helped those buyers and keep following the sales cycle.
Do not remove a useful article solely because it produces little last-click traffic. Equally, do not claim hidden influence to explain away a complete absence of commercial evidence.
TCC can help review the measurement and implementation scope. Bring the reports you actually have; missing data is a starting point to address, not a result to invent.
Frequently asked questions
Can every AI-influenced enquiry be tracked?
No. Some buyers visit through identifiable referrals; others return later through another channel. Keep tracked referral paths and the buyer’s reported discovery as separate fields.
Should we count AI visits as leads?
No. Report sessions, completed enquiries, accepted opportunities and orders separately. Only connect them when your records support the association.
Can we add AI impressions to Search impressions?
Do not add a subtotal to an overall figure that already includes it. Preserve each report’s definition, property and period before comparing results.