What You Can Track Today
You can track recognized referral visits from ChatGPT and other AI assistants. You can track Google AI feature impressions if your Search Console property has the new report. You can track forms, calls, bookings, lead quality, and revenue with first-party systems. You cannot reliably connect every AI answer exposure to a person or sale.
The most important split is between AI assistants and Google's AI search features. Google Analytics currently classifies them differently.
| Surface | GA4 default channel | Search Console visibility | Lead attribution |
|---|---|---|---|
| ChatGPT, Gemini, Claude, Perplexity, Copilot, and similar assistants | AI Assistants when a recognized referrer is present | Not in the Google Search generative AI report | Possible when a visit, action, and lead record can be joined |
| Google AI Overviews and AI Mode | Organic Search | Impressions by page, country, device, and date for eligible properties | Not identifiable from standard GA4 or Search Console alone |
Do not combine these into one “AI traffic” number
Search Console generative AI impressions are exposures. GA4 AI Assistants sessions are visits. Organic Search can contain Google AI clicks and regular search clicks. Adding the three together would mix different units and double count parts of the journey.
The AI Lead Measurement Stack
One platform cannot answer the full question. Give each layer a clear job, then join the layers with landing pages, timestamps, source fields, and pseudonymous IDs.
Layer 1
Visibility
Did a page appear in a Google AI feature? Search Console answers this where the new report is available. Manual citation checks can document examples, but they are samples.
Layer 2
Visit
Did an AI assistant send a session? GA4 and the raw referrer can answer when the referrer survives the click.
Layer 3
Action
Did the visitor submit, book, call, download, or start a trial? GA4 key events and first-party action records cover this layer.
Layer 4
Lead
Which source and session created the form, call, or booking? Hidden fields, session storage, and a server-side lead record preserve the connection.
Layer 5
Quality
Was the lead real, relevant, reachable, and qualified? Only call review and the CRM can answer this reliably.
Layer 6
Outcome
Did the lead become an opportunity, sale, retained client, or program start? CRM and billing data close the loop.
Use Search Console for Google AI Visibility
Google began rolling out a dedicated Generative AI performance report on June 3, 2026. It is still limited to a subset of sites. A property may also lack the report because it has not received enough impressions in supported features.
The report includes AI Overviews and AI Mode. Its current documented metric is impressions. You can break those impressions down by page, country, device, and date. That makes it a strong visibility report and a poor lead-attribution report.
Use it to answer
- Which pages appear in Google generative AI features?
- Is exposure rising or falling over time?
- Which countries and devices account for visibility?
- Did visibility change after a page or template update?
What it cannot establish
- A specific person saw the citation.
- An impression produced a site visit.
- A lead came from an AI Overview.
- Every generative query or prompt is visible.
- 1
Record access and the first available date
The rollout is incomplete, so a missing report does not mean the site had no AI visibility.
- 2
Export by page and month
Store a stable monthly extract so page changes and launches can be annotated beside the visibility line.
- 3
Keep property and page totals distinct
Google notes that chart and table totals can differ because property-level and page-level aggregation are different.
- 4
Join only at an aggregate level
Compare exposed pages and dates with organic landing-page behavior. Call it correlation unless a session source or first-party record supplies the direct connection.
Use GA4 for Recognized AI Assistant Visits
On May 13, 2026, Google Analytics added dedicated AI Assistant traffic measurement. When a referrer matches Google's recognized list, GA4 assigns the ai-assistant medium, the (ai-assistant) campaign, and the AI Assistants channel.
This removes much of the custom-regex work that older AI referral reports needed. Keep source and medium in the report anyway. The channel gives a total; source identifies which assistant sent the visit.
Build this GA4 exploration
- Use Session default channel group and filter for AI Assistants.
- Add Session source / medium and Landing page + query string.
- Add sessions, engaged sessions, key events, and the key event rate.
- Break out form, booking, trial, download, and click-to-call events that represent different intent.
- Compare first-user AI acquisition with session-level AI referrals so new discovery and later assists are not confused.
Google AI traffic is the exception
The current default channel documentation explicitly excludes Google AI Overviews and AI Mode from AI Assistants. Those visits remain Organic Search. GA4 does not provide a default dimension that cleanly separates a Google AI click from a regular Google organic click.
Use Search Console to see which pages earned generative AI impressions. Use GA4 to see what organic visitors did on those pages. The overlap is useful for diagnosis, but it does not prove that each visit came from an AI feature.
Carry the Source Into the Lead Record
GA4 reports behavior in aggregate. Salespeople need a lead record. Capture source context in first-party storage, then pass it into forms, bookings, chats, and calls with a pseudonymous session identifier.
The first touch and the converting touch answer different questions. Preserve both. A visitor may arrive from ChatGPT, leave, then return from a branded search before submitting.
| Field | Why store it |
|---|---|
first_touch_channel | The first observed acquisition group, preserved across later sessions. |
first_touch_source | The first observed source or referrer host, such as chatgpt.com. |
first_landing_page | The first page the visitor reached and its clean campaign parameters. |
latest_session_source | The source attached to the session that produced the form or call. |
latest_landing_page | The landing page for the converting session. |
visitor_session_id | A pseudonymous first-party ID that joins the web session to the lead record. |
self_reported_discovery | The visitor or salesperson answer to how the company was first discovered. |
Implementation rules
- Normalize the source for reporting, but preserve the raw referrer host for troubleshooting and future analysis.
- Write the first-touch values once. Update latest-touch values at each qualifying session.
- Generate the lead ID on the server and attach it to the form or call record.
- Respect the site's consent controls, privacy policy, retention rules, and regional requirements.
- Keep names, email addresses, phone numbers, and other recognizable personal information out of GA4.
- Never put personal information inside URLs or campaign parameters.
Do not invent the missing prompt
Standard GA4 and Search Console reports do not provide the user's ChatGPT prompt or Google AI query. Store the referrer, landing page, and observed action. If a lead voluntarily describes the question they asked, keep that as a CRM note with its own source label.
Connect Calls to CRM Lead Quality
For service businesses, phone calls can be the highest-intent action and the largest attribution gap. Dynamic number insertion should read the current session source, assign a tracking number, and preserve the source, landing page, session ID, and call ID together.
The CRM then has to answer whether the call was useful. Raw call count treats spam, job applicants, current customers, wrong numbers, and qualified prospects as equal. They are not equal.
Observed source
AI assistant, organic search, paid search, referral, direct, or unknown, plus the raw referrer when present.
Lead disposition
Qualified, unqualified, existing customer, vendor, applicant, spam, duplicate, or unreachable.
Business outcome
Appointment, proposal, opportunity, won work, lost reason, revenue, and close date.
Add one self-reported question
Ask, “How did you first hear about us?” Include ChatGPT, Google AI or Gemini, Perplexity, Copilot, another AI tool, Google Search, referral, and other. Keep the answer separate from technical attribution. A self-reported answer can recover a hidden assist, but it should not overwrite an observed source.
Our lead generation SEO approach uses the same principle for every channel: capture the source, then report whether the lead was qualified.
Give Assisted AI Journeys Their Own Label
Last-click reporting will understate channels that start research. AI discovery often behaves that way. Keep the evidence level beside the lead so an assist does not become a fabricated direct conversion.
Directly observed AI lead
A recognized AI assistant referral starts the session and the same session produces the form, booking, or call.
Observed AI assist
A stored first touch came from an AI assistant. The same consenting visitor returned later through search, email, referral, or direct and converted.
Self-reported AI discovery
The lead says an AI product introduced the business, but the technical record does not contain an AI source.
AI visibility only
Search Console shows generative AI impressions for a page. No visitor or lead record can be joined to those impressions.
Inferred influence
AI visibility, organic visits, and leads moved together by page and date. Report the correlation as a hypothesis, not direct attribution.
A common path is ChatGPT referral, useful article, later branded Google search, then a form. GA4 may give the final session to Organic Search. The first-party record preserves the AI assist. If the first click happened inside Google AI Mode, both sessions may look organic, so only Search Console visibility and self-reporting can add context.
Known AI Attribution Blind Spots
A serious dashboard lists what it cannot observe. These gaps are normal:
- Zero-click answers: a person can see the brand and never visit the site.
- Google AI grouping: AI Overviews and AI Mode visits remain inside Organic Search in GA4.
- Missing referrers: apps, privacy controls, redirectors, and copied links can turn an AI visit into Direct or an unknown source.
- Prompt privacy: standard reports do not expose the prompt that produced a citation.
- Cross-device research: a phone discovery and a desktop conversion may never join.
- Consent and blocking: analytics may be limited or absent for some visitors.
- Search Console rollout: many properties still do not have the dedicated generative AI report.
- Impression aggregation: property and page totals can differ under Google's reporting rules.
- Brand search assists: AI can create a later branded search that receives the final credit.
Direct traffic is not an AI estimate
Some hidden AI referrals may land in Direct, but Direct also includes typed URLs, bookmarks, untagged documents, private messages, and other lost referrers. Do not relabel unexplained Direct growth as AI traffic.
A Practical AI Lead Dashboard Template
Build the report in layers and keep the unit visible. The dashboard should let a reader move from exposure to business outcome without pretending the rows are perfectly attributable to each other.
| Layer | Metric | System | Breakdown | Evidence label |
|---|---|---|---|---|
| Visibility | Generative AI impressions | Search Console | Page, country, device, date | Observed exposure |
| Visits | AI Assistants sessions | GA4 | Source, landing page, device | Observed referral |
| Actions | AI key events and calls | GA4 + call tracking | Event, source, landing page | Observed action |
| Leads | Forms, calls, and bookings | First-party lead store | First touch, latest touch, lead ID | Observed or assisted |
| Quality | Qualified leads and opportunities | CRM | Stage, service, owner, reason | Business verified |
| Outcome | Won work and revenue | CRM or billing | Original source, assist, close date | Business verified |
Put these controls above the charts
- Date range and comparison period.
- AI surface: Google generative features, ChatGPT, Gemini, Perplexity, Copilot, other, or unknown.
- Landing page and page group.
- First touch, latest touch, or assisted view.
- Lead status, service, location, owner, and close stage.
- Evidence level: observed, assisted, self-reported, visibility only, or inferred.
The companion guide to what an SEO report should show explains how to separate shipped work, leading indicators, conversions, and business results in the monthly narrative.
Implementation Checklist
A small team can put the foundation in place without buying a new AI visibility platform. Start with the systems already responsible for search, sessions, leads, and sales.
Measurement cannot create AI visibility. If the site is not being cited, start with the practical guide to earning citations in ChatGPT and Google AI Overviews, then use this stack to measure what follows.