Measure known-referrer visits, observed citations and received enquiries separately. Missing referrer information cannot reconstruct every origin. Define attribution rules and keep unknowns visible; an estimate of answer visibility does not become a traffic count.
What should be measured first
Start with a simple framework: traffic, visibility, quality, and conversion. Traffic tells you what is already measurable in analytics. Visibility tells you whether your pages are technically eligible and likely to be surfaced. Quality tells you whether those visits show intent. Conversion tells you whether AI-assisted discovery creates business value.
- Traffic: sessions, users, landing pages, source/medium, default channel group, custom AI channel group.
- Visibility: indexed pages, rich-result eligibility, structured data coverage, crawl access, local entity completeness, branded and non-branded presence.
- Quality: engagement rate, scroll depth, key events, assisted conversions, micro-conversions, lead starts.
- Conversion: qualified leads, calls, form submissions, booked demos, revenue, pipeline influence.
The measurement method
1. Create an “AI Assistants” traffic view in analytics
In GA4, create a custom channel group for AI assistants based on source rules, because the default channel grouping is fixed and not designed around every AI referral pattern. Google explicitly allows custom channel groups, and even provides an example for AI assistants. support.google.com
Typical rules include known sources associated with AI platforms, for example ChatGPT, Copilot, Perplexity, Claude or Gemini when they pass a referrer. Microsoft Clarity also classifies dedicated AI traffic channels by checking referral sources from standalone AI platforms. learn.microsoft.com
Use this channel to answer operational questions:
- Which AI platforms send measurable sessions?
- Which landing pages attract those visits?
- Which visits convert into leads or sales?
- Which device, geography or page type performs best?
2. Accept that some AI visits are hidden inside direct or unassigned traffic
Not every assistant click passes a clear referral source. In GA4, (direct) / (none) represents traffic without a clear source, and (not set) can lead to unassigned sessions. This matters because AI assistants inside apps, browser overlays, secure redirects or copied links can reduce attribution quality. Your AI reporting should therefore include a confidence note: measurable AI referrals are the observable minimum, not the full total. support.google.com
A practical control is to monitor whether unexplained direct traffic rises on the same landing pages that gain AI citations or AI referral sessions. That is an inference rather than a native GA4 metric, but it is often useful when patterns are consistent over time. clarity.microsoft.com
3. Track landing pages, not only sources
AI discovery tends to be highly page-specific. One FAQ, service page, comparison page or location page may be cited repeatedly while the rest of the site remains invisible. Build reports around landing page + source, then segment by page template:
- service pages,
- editorial guides,
- FAQ pages,
- case studies,
- product or category pages,
- local pages by city or agency location.
If needed, add content_group in GA4 so you can compare these page families cleanly. Google documents content groups for this type of custom analysis. support.google.com
4. Compare AI traffic quality with other acquisition channels
AI traffic is often smaller in volume but stronger in intent. Microsoft Clarity reported that AI-referred traffic grew sharply and showed stronger conversion or engagement signals than many traditional channels across the sites it studied. That does not guarantee the same result for every business, but it is a useful benchmark for prioritization. clarity.microsoft.com
For agency reporting, compare at least:
- engagement rate,
- average engagement time per session,
- lead form starts,
- lead form completions,
- phone click events,
- newsletter or demo requests,
- assisted conversions and final conversions.
Technical checks to run before trusting the numbers
Check referral integrity
- Review source / medium values manually each month.
- Check unwanted referral exclusions so self-referrals do not pollute acquisition data. support.google.com
- Confirm cross-domain measurement if forms, booking tools or checkout flows use other domains.
- Document naming rules in the custom channel group.
Check bot and crawler access
If you want visibility in AI search experiences, verify that relevant crawlers are allowed where appropriate. OpenAI states that inclusion in ChatGPT search depends on allowing OAI-SearchBot and ensuring hosting or CDN layers allow the published IP traffic. Perplexity documents its crawler behavior and says its bot respects robots.txt disallow rules for indexing full or partial text. help.openai.com
This does not mean “allow every bot by default.” It means reviewing robots.txt, WAF rules and CDN settings intentionally, with legal, security and brand considerations documented.
Check search visibility alongside analytics
Do not interpret low AI referral traffic as low AI visibility. Some assistants cite your content without generating many clicks. Pair analytics with manual answer audits, citation monitoring where available, and page-level search performance reviews. Microsoft Clarity has also introduced AI citation visibility features, which reflects the market’s shift from pure visit counting to citation-level observation. clarity.microsoft.com
Main risks and interpretation mistakes
- Overcounting: classifying ordinary referrals as AI traffic because source rules are too broad.
- Undercounting: ignoring direct or unassigned traffic patterns that may contain assistant-driven visits. support.google.com
- Vanity reporting: celebrating mentions or citations without checking lead quality.
- Template blindness: looking at sitewide averages instead of the few pages that actually surface in AI answers.
- Technical disconnect: trying to improve LLM visibility while pages remain slow, thin, duplicated or poorly linked.
- Local blind spots: forgetting that location pages often drive high-intent AI and map-adjacent discovery.
Unknown origin stays unknown
Fictional example: a visit without a referrer reaches a page often cited by assistants. That coincidence does not establish an AI origin. The team keeps an unknown category and separately analyses sessions with identifiable origins.
Document attribution rules and their coverage. Do not allocate unknown visits among channels through an assumption hidden from the dashboard.