ChatGPT sessions are visible in an acquisition report. Someone spots them, exports a screenshot and asks how many times the brand was recommended. GA4 cannot answer that second question.
For analytics teams in Dubai, the UAE and elsewhere, AI referral traffic is still worth tracking. It can show that a visitor clicked from an AI product, which page they reached and whether they took a meaningful action. It is a useful commercial signal. It is not a count of every answer in which your business appeared.
Use GA4 for the part it can actually observe
GA4 can identify many visits from ChatGPT, Perplexity, Gemini, Copilot and similar products when a click passes a recognisable referrer or campaign signal. It cannot reliably measure AI answers that mention, cite or recommend a business without generating a visit. Referrals therefore measure clicked AI-assisted discovery, not total AI visibility, recommendation frequency or AI share of voice.
That distinction prevents two familiar reporting errors. One is dismissing AI influence because referral sessions are small. The other is treating every AI referral as proof that the platform recommended the company for an important buying query.
A prospect might click from ChatGPT straight to an AI visibility audit page after asking a relevant question. That is a valuable session to inspect. Meanwhile, other buyers may see the brand name in an answer, search for it later, ask a colleague, or never click at all. GA4 will not join those events into one neat story. Analytics rarely does.
Checklist: build an AI-referral view you can maintain
- Start with raw acquisition data. In GA4, inspect session source and medium before creating any grouping. Look for referral traffic, campaign-tagged links and unusual source variants. Keep the raw values available in an Exploration or export. A channel called AI referrals is useful for reporting, but it should not replace source-level detail.
- Define classification rules, not a permanent domain list. Create an AI-referral grouping based on observed source hosts, known product-owned referrers and documented campaign conventions. Record the rule, the date added and the reason. Products change domains, use regional hosts, launch app hand-offs and sometimes strip referrers entirely. A hard-coded list will age badly.
- Exclude obvious noise before celebrating. Check that a source is genuinely an AI product rather than a scraper, preview tool, internal test or a referral name that merely resembles one. Review hostname, landing page, engagement and any sudden one-off spike. Referral spam is less glamorous than GEO, but it still exists.
- Preserve the source, medium and landing page together. A grouped total conceals the useful question: which source sent a visitor to which page? Use a report or Exploration with session source, session medium, landing page and your primary conversion event. This is where an apparent AI traffic trend becomes operationally useful.
- Inspect the page the visitor actually saw. If AI referrals land on a generic homepage, the data may tell you little about intent. If they land repeatedly on a service, comparison, audit or contact page, review whether that page answers the likely question cleanly. Check the rendered page too. A page can look complete in a browser while the key service explanation is injected late by JavaScript and absent from a basic fetch.
- Connect visits to commercial actions cautiously. Use your own meaningful events, such as enquiry submission, booked call, qualified lead or document download. Compare AI referrals with other channels, but avoid declaring causation from a handful of sessions. A visit can assist a decision without being the final credited channel, and a conversion can be credited without proving a recommendation occurred.
A practical reporting layout
| Report field | What it helps answer | What it cannot prove |
|---|---|---|
| Session source and medium | Which recognisable referral or campaign sent the visit | How often the brand appeared in AI answers |
| Landing page | Which content attracted the click | The exact prompt or wording used |
| Conversion event | Whether the visit led to a recorded action | That AI alone caused the commercial outcome |
| Grouped AI referrals | Whether clicked AI traffic is changing over time | Total AI visibility across platforms |
| Raw source detail | Whether classification rules need updating | That missing referrers mean no AI influence |
Keep the grouping useful without making it opaque
A maintainable grouping normally has three layers: a labelled AI-referral channel for management reporting, a source-level view for diagnosis, and an exceptions log for new or uncertain referrers. Do not bury the last layer in one analyst’s notebook. Put it in the reporting documentation.
For example, if a previously unseen source sends visitors to a guide page with referral medium, inspect it before adding it to the group. Record whether it is a product domain, an intermediary redirect, an app hand-off or an unresolved source. This avoids quietly inflating the category every time a new domain appears.
Campaign tagging deserves the same discipline. If your team shares links in AI communities, newsletters or partner tools, use consistent UTM conventions. Otherwise a manually promoted link may be mistaken for organic AI referral traffic. The source and medium fields are only as tidy as the routes into the site.
Read landing pages before you read the total
Totals are tempting because they fit a board slide. Landing-page patterns are more revealing. A cluster of visits to a technical guide may indicate informational discovery. Repeated visits to a pricing, audit or service page deserve closer review because the visitor has reached a more commercial part of the journey.
Look for practical signals: page engagement, scroll depth if configured sensibly, form starts, form submissions and assisted paths in your wider measurement setup. Also inspect whether the page has a clear offer, explicit category language, named proof and a visible next action. A referral arriving on a vague page is not much use to the buyer or the analyst.
Put referral traffic beside prompt evidence, not in place of it
AI referrals are one measurement stream. Prompt tracking is another. Prompt tests can observe whether a brand is mentioned, how it is described, whether it is cited and whether it appears for realistic branded and unbranded questions. They still need consistent prompts, dates, locations and scoring rules. A favourable single answer remains an observation, not a market verdict.
Use the two together with commercial evidence. FlareFalcon’s AI visibility baseline framework places referral traffic alongside prompt observations, website conditions and business outcomes before major website changes are made. That makes it harder to mistake a traffic report for an answer about visibility.
Where share of voice is part of the question, use a repeatable prompt set rather than an acquisition chart. The AI share of voice measurement checklist explains why one good answer cannot stand in for a useful comparison over time. Referral data can then show whether any observed visibility also produced clicks.
Document the blind spots in the monthly report
Add one plain-language note to the report: AI referral sessions represent visits that arrived with a measurable signal. They exclude no-click mentions and may miss clicks where referrer information was removed, redirected or unavailable.
This is not a weakness to hide. It is the boundary that makes the report credible. If a business is investing in machine readability, entity clarity, answer-ready content or authority signals, the measurement plan should be able to distinguish a website change from a platform behaviour change and from a simple attribution gap.
An AI visibility audit with measurement priorities can help establish those boundaries before teams rebuild pages or start reporting broad claims from a small referral segment.
Questions teams ask about AI traffic in GA4
Which AI sources can appear in GA4?
Recognisable referrers from AI products can appear when a visitor clicks a link and the hand-off preserves source information. The exact names and domains vary by platform, device, browser, region and product changes. Review your own raw source data regularly rather than relying on a fixed public list of AI domains.
How should ChatGPT and Perplexity traffic be grouped?
Create an AI-referral channel using documented rules based on observed source hosts, referral medium and any controlled campaign conventions. Retain raw session source and medium in the underlying report. This gives stakeholders a usable total while allowing analysts to check whether the grouping still reflects real traffic.
Does GA4 capture no-click AI visibility?
No. GA4 records website activity, not every AI answer shown to a user. A brand can be mentioned, cited or recommended without a click, and some clicks may not retain a usable referrer. Use referral data alongside structured prompt tracking and commercial evidence rather than presenting it as a complete measure of AI visibility.