What you can actually measure about your visibility in AI search

Search Console's dedicated Generative AI performance report covers impressions in AI Overviews and AI Mode. In GA4's default channel grouping, traffic from those experiences belongs to Organic Search alongside other organic search traffic. The channel called AI Assistant explicitly excludes them. [1][2]

That has a practical consequence: taking AI Assistant sessions from GA4 and dividing them by Google AI impressions will not give you an AI click-through rate. The visits and impressions cover different sources. Sessions are also different from clicks, so changing the channel alone would not fix that calculation. [3]

Before judging AI performance, define what the measurement includes. A dashboard can put related numbers beside each other without making them parts of the same customer journey.

Follow the definition, not the label

“AI visibility” can mean that your business was mentioned in an answer, that a page was cited as a source, or that a platform recorded an impression. Website visits answer a further question: did someone arrive?

Start with the decision you need to make. If you want to know whether a particular page is appearing in Google AI search, inspect the relevant impressions report. If you want to understand visits from recognised AI assistants, inspect the analytics channel and its sources. To judge how an answer describes your business, read the answer.

Google's documented reporting makes the distinction concrete:

Scroll sideways to read all three columns.

These are the documented categories, not interchangeable definitions of AI search. [1][2]

Changing the chart title to “AI traffic” would hide the exclusion. A clearer title preserves the actual channel name, with a note explaining where Google's AI search traffic is classified.

What an appearance report lets you say

Google's dedicated report documents impressions with breakdowns including pages, countries, dates and devices. It does not document clicks or click-through rate in that report. Check whether data is available for your property: absence of the report is not sufficient evidence that your content has never appeared. [1]

Bing measures appearances differently. Its AI Performance public preview reports citations across Microsoft Copilot, Bing AI summaries and selected partner integrations. A citation records a source reference; it does not establish that someone visited the page. Preserve the stated platform coverage when reporting the result. [4]

Bing's preview Citation Share also needs its definition attached. It reports your site's citations as a percentage of all citations across sites for the same grounding query, a grouped phrase describing retrieval context. It is not a share of website visits or an overall measure of market share. [5]

Connect visits to outcomes within the same scope

Recorded visits are useful evidence in their own right. OpenAI documents adding utm_source=chatgpt.com to referral URLs from ChatGPT search results to help publishers identify incoming traffic. That tag describes arrivals, not every answer in which a publisher appeared. [6]

For a session-based GA4 report, use Session default channel group and keep the session scope consistent when examining channels and sources. Google's documentation distinguishes session-scoped channel attribution from event-scoped attribution. Mixing those views can change the question your report answers. [2]

Then examine what those recorded visits did. A completed enquiry form is more useful to a lead-generation business than an undifferentiated visit count, provided the tracking actually records completed enquiries. A contact-button click shows intent; calling it an enquiry would overstate the action.