Measuring Traffic from ChatGPT, Gemini, and AI Mode: What Changed and How to Check the Data

On October 15, 2026, Google announced link changes in AI Mode and AI Overviews intended to make it easier to move from answers to websites. Two days earlier, Microsoft launched Topics Portfolio in beta within Bing Webmaster Tools. The announcements connect two questions website owners in Israel need to examine separately: where a website appears in AI answers, and how many people actually reach it from them.

The distinction has become more useful following Google's John Mueller's direct observation of UTM tags in Gemini links and the ability to customize brand terms in Clarity. But no single screen provides the full picture. Exposure, a citation, a visit, and an inquiry are different events, and a classification change can alter a report without adding a single customer.

Google highlights links but does not promise growth for a particular website

In its October update, Google described a greater presence of links within answers, including more frequent placement higher in the response. The company also announced a test of a dedicated area highlighting links to official websites when a particular source is needed to complete a task, and the expansion of recent-article carousels to more languages and locations.

These are changes to the interface and ways of reaching content. The announcement does not prove visits increased for every website or guarantee additional traffic for a particular Israeli business. Select relevant pages, save baseline data, and assess them over time. More impressions are one finding; more visits that lead to inquiries are another.

Also distinguish Gemini from the AI experiences in Google search. Different products using AI do not become a single traffic source. For example, a visit from a conversational assistant's link and a visit from a search result may appear in different channels within the same analytics property.

Gemini and UTM tags: what is confirmed and what remains unknown

In the original Reddit discussion about Gemini adding UTM tags to links, Google's John Mueller wrote that he was seeing them too. He later requested reproducible examples of lost referrer information and noted that retaining that information alongside the tags would be useful. This is a direct observation by a Google representative, not a launch document defining a complete rollout.

No official documentation was found detailing every link type, device, or circumstance in which Gemini adds tags. Do not assume that every visit from it is identified. A UTM is a parameter added to a URL to provide attribution information; the referrer is a separate signal. Check both rather than relying only on a report's channel name.

As a suggested test, open a link in an AI answer to a website you own, record the URL you arrive at, and check the source captured. Repeat the test in environments relevant to your customers. This reveals what happens on the tested route, not complete platform coverage.

Start in GA4 with the source of the current visit

GA4's Traffic acquisition report uses session-level dimensions to understand the source of visits, including return visits. A First user dimension concerns the user's initial acquisition, so it does not answer where they came from on the current visit. You can check as follows:

1. Open Traffic acquisition in the reports area. Choose a comparison period appropriate for the test.
2. Change the dimension to Session source / medium, or use Session source. Look for sources associated with ChatGPT and Gemini as they actually appear in your property.
3. Also check Session default channel group. Google's current documentation includes an AI Assistant channel, but it does not include AI Mode or AI Overviews.
4. Compare visits, engagement, and key events. If purchase measurement is correctly configured, examine the relevant revenue data too.
5. Filter by page to understand which content receives traffic and which actions visitors take there. Save the selection for consistent comparisons.

Google classifies organic visits from AI Mode and AI Overviews under Organic Search. A screen showing only AI Assistant is therefore not a report of all traffic influenced by AI. Conversely, do not classify all google / organic traffic as originating from AI answers: it includes other organic search too.

A custom channel group organizes what was measured

For a tailored view, Google publishes instructions for creating a channel group that brings AI assistants together. You can define matching conditions for the desired sources and select the group in acquisition reports. The guidance emphasizes channel order and the need to update matching rules as sources change.

Build the rules from source values present in your property and check examples before expanding them. An overly broad match can include sources you did not intend to measure. If you run paid advertising on the same platform, separate paid campaigns using their campaign data and tags so they are not mistakenly shown as organic traffic.

A custom group can also organize existing historical data, as the documentation notes, but cannot invent a source that was never collected. Do not turn a channel name change into a change in customer history. Record when rules changed so monthly comparisons do not mix two different classification methods.

When Direct rises, inspect the measurement path

According to Google, (direct) / (none) describes traffic without a clear referral source. Missing information, untagged links, or redirects that strip parameters can contribute to it. A decline in an AI source alongside an increase in Direct warrants investigation, but does not prove visits moved from one to the other.

Check whether redirects, shortened URLs, analytics implementation, or campaign links changed. Compare the same pages and inspect the final URL after navigation. If a tag disappeared before measurement, fix the path and record the correction date. Without a source-identifying signal, do not retrospectively move all Direct visits to ChatGPT based on intuition.

Search Console shows AI visibility, not a ChatGPT visits report

Google announced Generative AI performance reports in June 2026 and later updated that they had rolled out to all websites worldwide on August 31. The dedicated search report shows link impressions within AI Overviews and AI Mode. The documentation notes that it may not appear for a website with insufficient impressions.

Open the report in the relevant property, choose a period, and compare pages, countries, and devices. Local businesses should examine Israel separately; businesses exporting services should assess each target market. The documented report focuses on impressions, so do not present them as clicks or use them alone to calculate visit counts.

Also examine the overall search performance report and visit data in GA4. AI data is included in overall performance too, so it should not be added to total impressions as if it were separate. Search Console does not measure referrals from ChatGPT or the standalone Gemini application. For daily comparisons, remember that dates in the impressions report use US Pacific Time.

Bing and Clarity explain where you appear and in what context

Topics Portfolio, launched on October 13 within AI Performance in Bing Webmaster Tools, groups the grounding queries in which a website appears into topics. Microsoft also shows opportunity queries where other websites appear. These queries are used to retrieve sources and are not necessarily the exact wording a person entered.

The figures are estimates for planning. Microsoft warns that automatic categorization can be wrong, some citation volumes are shown as ranges, and websites with limited data may not see the feature. Check whether a topic fits the business before deciding to create content for it. A citation in a system is not a website visit.

In Clarity, a September 29 update allows Brand Terms to be added and edited through Settings, then AI Visibility. An Israeli business should check its Hebrew name, English name, transliterations, and genuine abbreviations. This can refine the distinction between queries mentioning the brand and discovery through a general topic. The documentation separates citation data from the AI assistant traffic metric, and the business report should preserve that separation.

The weekly report should lead to a decision

Organize monitoring in three layers: visibility in answers, identified visits, and business outcomes. For broader context, read about generative answer engines. When a page gains visibility but generates no action, check its fit with the need and the next step. When a traffic source changes sharply, first check what was collected. This helps determine whether to improve content, fix measurement, or change the offer, without assigning a number meaning it does not have.