Which instrument shows traffic arriving from an AI assistant, which one shows mentions, and the Google Analytics channel that now separates them

This is two questions wearing one sentence, and they need two instruments. Traffic arriving from an assistant is answered by your own analytics property, free, and Google now has a named channel for it. Mentions inside an assistant's answer are answered by a monitoring product, because your analytics can only see people who clicked, and most of an answer's influence never produces a click at all. Buy the second one only after the first one is reading correctly, because a mention report beside a misconfigured traffic report produces an argument rather than a decision.

The first fact to have is the one that changes most Singapore dashboards. Read on support.google.com/analytics/answer/9756891 on 29 October 2026, Google's default channel descriptions now include an AI Assistant channel, and the same table puts Google's own AI surfaces somewhere else entirely.

The two definitions, quoted, and the line between them

From the default channel group documentation, read 29 October 2026.

• AI Assistant. "AI Assistant is the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok. It excludes Google's AI Overviews and AI Mode."
• Organic Search. "Organic Search is the channel by which users arrive at your site/app via non-ad links in organic-search results, including Google's AI Overviews and AI Mode."

Read those two together and the consequence is sharp. A click from ChatGPT and a click from an AI Overview are both AI traffic in ordinary speech, and Google counts them in two different channels. So the question "how much of my traffic comes from AI rather than from Google" has two correct answers depending on which one you mean, and a number quoted without saying which is a number nobody can check. Decide the definition before the dashboard, write it down, and keep it the same every month.

Three mechanical notes from the same page, because they decide whether the channel appears in your own reports. The page states "Default channel groups can't be edited in Google Analytics" and that "You have the option to create custom channel groups, which let you create your own rule-based categories of web-traffic sources". It also states "We recently made improvements to default channel grouping for Google Analytics properties", which is the sentence that tells you to re-read this document rather than trust a note from earlier in the year. And the channels surface on three different dimensions, Default channel group, Session default channel group and First user default channel group, each with its own attribution model and scope, so two people reading "AI Assistant" off two reports can legitimately get two numbers.

One thing to be careful about in the definition itself. Google writes "sources like" and then names five, which is an example list and not a closed one. If a particular assistant matters to your category, confirm it is landing in the channel you expect by checking the source and medium behind the sessions rather than assuming from the name.

What each vendor says it measures, read on 29 October 2026

The question asks which tools measure traffic and mentions. Here is what the four vendors in this category served in their own documents on one day, sorted by whether a traffic instrument is named at all.

Vendor and page	Traffic side	Mention side
tryprofound.com, HTTP 200	Names Agent Analytics in its own navigation and states "Prove impact on AI-driven traffic. Understand how bots view, cite, and refer to your pages." Read that carefully: the subject of the sentence is bots, so this is crawler and agent behaviour rather than a count of people who arrived	Answer Engine Insights, Prompt Volumes and citation tracking, with page level citation decay named as a feature
otterly.ai, HTTP 200	Names Agent Analytics in its navigation and publishes a free tool called AI Referral Traffic, which is the only named referral-traffic instrument in this set	Prompt Research and AI Search Analytics, described as getting "your brand mentioned, and your website cited on ChatGPT, Perplexity, AI Overviews, AI Mode, Gemini, and Copilot"
scrunchai.com, HTTP 200	Carries AI website traffic in its own navigation, beside Site Optimization described as "Improve how AI consumes your site"	Monitoring and Citations, described as "Monitor prompt results, surface authoritative citations, and track brand perfo" in the served text
peec.ai, HTTP 200	No traffic-side instrument appeared in the served document on this date	Prompt tracking with tags, tracked across countries, with a competitor comparison view

Two of the four, then, name something on the traffic side, and the two descriptions are not measuring the same thing: one is about crawlers reading your pages and one is about visitors arriving from an answer. A buyer comparing them has to ask which. Scrunch also publishes a results claim on its own homepage, which it states as a "364% increase in brand presence for non-branded prompts" attached to a named customer. Treat that the way you would treat any supplier-published figure: ask for the baseline, the prompt set, the date range and who chose the prompts, because brand presence measured on a prompt list the vendor selected is a different number from brand presence on the list your buyers actually type.

For what each of these vendors publishes about price, and which of them prints a figure in a document a machine can read, the dated reading is in our page on which AI visibility tool a Singapore buyer should shortlist. For which of them connect to your analytics property at all, we read 22 documents from 12 companies in our page on which AI visibility tools integrate with Google Analytics, and that is the question to settle before a procurement form goes out, because a mention monitor that cannot join to your sessions leaves you doing the join by hand.

Why two engines answered this with a CRM

The instrument behind this page is saved assistant answers rather than a keyword tool, and no search volume exists for a question asked in this form. On 28 October 2026 we re-ran the Singapore audit across 21 batches and 126 questions on ChatGPT and Gemini, producing 42 readings with no failures. The brand batch names us inside its own prompt and is excluded from every blind figure, which leaves 40 blind readings.

The batch carrying this question held five analytics and attribution questions and produced two readings. In both of them the engines named HubSpot and Profound, and we were named zero times. One of those two is a customer platform and one is an AI search visibility product, so of the two names a buyer is handed, exactly one is in the category the question asked about.

The citation list says the same thing more plainly. That batch returned 20 distinct hosts and the attribution stack in it is the Google-era one: support.google.com, www.hubspot.com, knowledge.hubspot.com, www.ruleranalytics.com, dreamdata.io, triplewhale.com, www.terminusapp.com, help.siteimprove.com and clutch.co. Six were Singapore hosts, including marketingagency.sg, www.mrconsultants.sg, cognitiv.sg, dupple.com, www.digimau.com and www.therankmasters.com. Not one AI-answer monitoring vendor other than Profound appears anywhere in it.

So the answer a Singapore buyer currently receives is assembled mostly from material written about measuring Google traffic, for a question about measuring assistant traffic. Google's own documentation, which is in that citation list, has already moved further than the vendor commentary around it.

The order to set this up in, and the Singapore detail that changes the prompt list

Free first, in this order, and each step is checkable the day you finish it.

• Look for the AI Assistant channel in your own property. If it is not there, build a custom channel group with the assistant hostnames as rules, because default channel groups cannot be edited and a custom group can.
• Decide, and record, whether your AI number includes AI Overviews. Google's definitions put it in Organic Search. Pick one definition and keep it.
• Read the three channel dimensions once. Default channel group, Session default channel group and First user default channel group answer three different questions, and two people quoting two of them will disagree forever without knowing why.
• Tag nothing that an assistant can strip. The arrival you are counting carries whatever referrer the assistant sent, so build the segment on the hostname and check it against the raw source and medium rather than on a campaign parameter you cannot control.
• Then, and only then, price a mention monitor, and price it against a prompt list you wrote rather than one the vendor supplied.

And one Singapore specific thing that belongs in that prompt list, measured in the same 28 October 2026 re-sweep. The citation volume an answer returns depends on the language the question was asked in. Across those readings, 16 English readings returned 1,053 citations, which is about 66 a reading. Eight Malay readings returned 342, about 43 a reading. Eight Tamil readings returned 316, about 40. Eight Chinese readings returned 248, about 31. Those are four separate measurements of four languages and they are never added. The practical consequence for a Singapore buyer is that a monitoring subscription priced per prompt, measured only in English, is measuring the longest and best sourced answers in the set and missing the ones where fewer sources decide more. Write at least one non-English version of your top five questions into the list before you count prompts for a quote.

Questions a Singapore buyer asks about measuring assistant traffic

Does Google Analytics show me traffic from ChatGPT

Yes. Google's default channel descriptions, read 29 October 2026, define an AI Assistant channel as arrivals "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok", and state that it excludes Google's AI Overviews and AI Mode. Confirm the sessions behind it against the underlying source and medium before you report the number upward.

Where does AI Overview traffic land then

In Organic Search, by the same documentation on the same date, which defines Organic Search as including Google's AI Overviews and AI Mode. That is why an AI traffic figure needs its definition attached.

Do I need to buy a tool to see assistant traffic

No. The traffic half is answered by the property you already own. The paid products in this category earn their place on the mention half, which your analytics cannot see at all, because an answer that names you without a click leaves no session behind.

Which vendor measures both halves

Read on 29 October 2026, Otterly and Profound both name a traffic-side instrument beside their mention tracking, Scrunch carries AI website traffic in its navigation, and no traffic-side instrument appeared in Peec's served document. Ask each one in writing whether it reports crawler activity, visitor arrivals, or both, because two of those descriptions are not the same measurement.

How does this connect to proving the agency produced the leads

It is the same evidence problem one layer up, and the test that settles causation is a pause rather than a dashboard. The ownership list and the pause test for a disputed lead claim are in our page on how to tell whether your Singapore agency is actually responsible for the leads it claims.

The row that does not exist on either instrument

Between the two instruments above there is still a gap, and it is the one worth closing on purpose. Your analytics counts the people who clicked. A mention monitor counts the answers that named somebody. Neither tells you, in one file you can hand to a finance director, which of your buyers' actual questions produce your name and which produce a competitor's, on a fixed list, on a stated date, with named and cited kept apart because they fail for different reasons. That file is what we build at AI Knows Us, and the criterion we ask to be judged on is the same one this page used on four vendors, which is whether the figure and the method are in a document somebody can read. On that test here is our own result from the 40 blind Singapore readings on 28 October 2026: zero namings and zero citations, against HubSpot's two in this batch. A buyer choosing on who the engines currently recommend in Singapore would not choose us, and printing that is the only honest thing to do with it.