
Start with GA4’s AI Assistant channel, then inspect session source and landing pages to understand the visits behind it. The channel measures recognised traffic reaching your site from supported assistants. It does not count every AI mention, every generated answer or every crawler request.
Google introduced dedicated AI Assistant traffic measurement in May 2026. Older tutorials that tell you to build a custom channel before checking the default reports can therefore miss the current starting point. This workflow uses the native report first, then investigates evidence that may sit outside it.
1. Confirm the property and reporting scope
Open the GA4 property that actually receives data from the website. Check the selected date range and reporting time zone. If the property has no working collection or you cannot access the report, the traffic total is unknown; a missing connection is not evidence of zero visits.
Use a session-scoped acquisition view when the question is how people began visits. First-user acquisition answers a different question about the source of a user’s initial acquisition. Keep that distinction in your notes so a later report does not accidentally compare different scopes.
2. Find the native AI Assistant channel
Open the traffic-acquisition report, or locate it through the report search if your navigation is customised. Use Session default channel group and look for AI Assistant. Google’s current channel documentation distinguishes these visits from Google AI Overviews and AI Mode, whose organic clicks are included in Organic Search.
Review sessions, engaged sessions and relevant key events for the same range. If the row is absent, investigate collection, the period and available traffic rather than assuming there are no AI-influenced customers. The default channel is a classification rule, not a complete measurement of everyone who consulted an assistant before buying.
| Signal | What it describes | Do not count it as |
|---|---|---|
| AI Assistant sessions | Recognised assistant-origin visits to your site | Every answer containing your brand |
| Organic Search sessions | Organic search visits, including Google AI search features | A separate ChatGPT total |
| AI crawler requests in server logs | Automated requests to website resources | Human visits in GA4 |
| Brand citations in an external tool | Mentions within that tool’s observed sample | Website sessions or completed orders |
3. Inspect the source behind the channel
Add or switch to a session-source or session-source/medium breakdown. Look for the actual assistant-related values present in your property. Google’s release notes describe a recognised assistant referrer receiving the ai-assistant medium and the AI Assistant channel classification. Record the actual values rather than assuming all assistants use one domain.
Then examine the landing pages those visitors used. A guide that attracts a handful of relevant visits may be more useful than a homepage row with no meaningful actions. Use the same time window and segment while reviewing the destination and subsequent activity.
4. Investigate missing classifications carefully
Some visits may carry campaign tags, appear under another source or lose a referrer before reaching your site. Inspect the source, medium, campaign and landing-page evidence together. Do not move every Direct session into an AI bucket: absence of an identifiable source does not reveal where the visitor came from.
For an optional custom analysis, maintain an explicit list of observed assistant source values and reviewed campaign rules. Use Google’s supported custom-channel process if it meets the reporting need. Keep the native view available for comparison and document the order and scope of your custom rules.
| Observed case | Investigation | Reporting treatment |
|---|---|---|
| Recognised assistant source | Check native channel and session details | Use the native result with its date range |
| Assistant-looking tagged link | Inspect actual source, medium and campaign | Include only through a documented matching rule |
| Direct visit with no source evidence | Review collection limits and available context | Leave unknown origin unassigned |
| Robot request in access logs | Identify the log event and user agent appropriately | Keep separate from human acquisition reporting |
| A visit matches two analysis filters | Check overlap at the same session scope | Deduplicate rather than adding both totals |
5. Define useful visitor actions before judging quality
Decide which actions matter for the page: a qualified enquiry, a product view followed by a real order, a requested resource or an outbound partner click. Check that the corresponding measurement is working. A session count without a relevant outcome cannot tell you whether the traffic helped the business.
If the page recommends partner tools, track affiliate clicks as a separate event and compare them with confirmed partner outcomes when available. An outbound click is not a commission. For broader channel naming, use the UTM guide without adding artificial campaign tags to internal site links.
6. Use a small evidence worksheet
The following table is an illustrative reporting layout, not Seolign traffic data. Keep the native count, any broader reviewed segment and known limitations visible. If the broader segment overlaps the native channel, do not sum them. Report the combined set only after applying a consistent, deduplicated definition.
| Worksheet field | Illustrative entry | Why retain it |
|---|---|---|
| Property and range | Correct website property; selected reporting dates | Makes comparison reproducible |
| Native segment | AI Assistant at session scope | Preserves the standard starting point |
| Source details | Observed source/medium values | Explains where the visits were classified |
| Destination | Landing page plus relevant actions | Connects traffic to useful content |
| Limitations | Referrer loss, consent or collection gaps where evidenced | Prevents a false complete-total claim |
| Next action | Improve a specific page or verify a specific event | Turns reporting into a decision |
7. Verify without manufacturing traffic
For a legitimate quality check, use a normal authorised visit and inspect available reporting or debugging evidence. Label internal tests appropriately in your measurement process. Do not generate repeated visits to inflate the AI channel or use fabricated engagement as evidence of growth.
After a representative period, compare consistent definitions and similar windows. Review what content received real visitors, whether those visitors completed useful actions and whether the acquisition pattern changed. Small samples can justify a follow-up investigation, but they do not support sweeping claims about the quality of all AI traffic.
Keep visibility and acquisition as connected but separate measures
Search visibility can precede a visit, and many people may read an answer without clicking. Use Search Console’s relevant views for Google search visibility and GA4 for measured website activity. A tool monitoring mentions answers another question again. Preserve these distinctions when presenting results to a business owner.
If no visits are measurable yet, work on useful content, accurate references and legitimate distribution while checking collection. Avoid promising that appearing in an assistant will deliver a particular volume of traffic. A clear report should say what was observed, what remains unknown and what decision the evidence supports.
Frequently asked questions
Does the AI Assistant channel include Google AI Overviews?
Google’s current default-channel documentation places organic clicks from AI Overviews and AI Mode in Organic Search, rather than the AI Assistant channel.
Can GA4 count every ChatGPT mention of my business?
No. GA4 measures activity collected on your website or app. A mention without a measured visit is not a session.
Should I create a custom AI channel immediately?
Check the native channel first. Create a reviewed custom view only when a specific reporting need remains, and preserve a documented definition and deduplication method.
Can I identify the AI source of all Direct traffic?
No. Without supporting source information, do not assign those sessions to an assistant. Report the attribution limitation rather than inventing a source.
Sources and further reading
GA4 default channel definitions | Google Analytics release notes | GA4 custom channel groups
Pingback: ChatGPT Advertising: How ChatGPT Ads Work & How to Launch