Most teams answer “are we showing up in Google’s AI Overviews?” by opening an incognito window, typing three queries and taking a screenshot. That is what one browser saw on one afternoon. It does not tell you whether you were there last Tuesday, whether you are there in Germany, or whether the answer recommended a competitor two sentences above your link.
To track brand mentions in AI Overviews and AI Mode you need three sources, and together they get you close. Search Console counts how often your links were shown in Google’s generative features. GA4 has the traffic, filed under a channel that surprises most people. Prompt-level tracking is the only view of the answer text itself.
Table of Contents
AI Overviews and AI Mode are two different surfaces
AI Overviews is the generated block that appears above the standard results on some queries. The ten blue links are still underneath. The block carries links to the pages Google drew on, and whether it appears at all depends on the query: plenty of searches still return a plain results page, and the set moves.
AI Mode is the conversational search experience. A user enters it deliberately and then keeps talking, so every turn produces a generated answer and the follow-ups shape what comes next. There is no results page underneath to rank in and no position to report.
That changes the question you can ask. In AI Overviews: did the block trigger here, and were we in it. In AI Mode: when someone asks this and then asks the obvious follow-up, are we in the answer they end up with. They trigger on different things and cite differently, so they have to be measured separately. Google’s own report does not separate them.
Source one: the Generative AI report in Search Console
Search Console has a Generative AI performance report under Search. It is the only first-party data anyone gets about these surfaces, and it is narrower than people expect.
What the report actually contains
- Impressions only. Google defines the metric as how many times links to your site were shown to a user in a generative AI feature on Google Search. No clicks, no CTR, no position.
- Both surfaces together. It covers AI Overviews and AI Mode. The dimensions are pages (by canonical URL), countries, dates and devices, so nothing splits one surface from the other.
- Dates in Pacific Time. You can group by days, weeks or months, and the buckets are Pacific Time. Your local week and this report’s week are not the same seven days.
- The same data is in the main report. It is included inside the Web search type of the main Performance report, so your existing Search numbers already contain it.
- The usual limits apply, including the 1,000 row limit. On a large site the page list you export is the top 1,000 rows, not your catalogue.
- Search Labs experiments are excluded, and a site needs enough impressions before data appears. A small site can see an empty report and still be appearing occasionally.
Google’s documentation for the report is worth reading for the metric definition alone, which a dashboard label will not give you.
What the report cannot answer
An impression here means a link to your site was shown. It does not mean your brand was named in the prose. Those are two different events, and the report only counts the first.
- Whether your brand was named in the generated text, or how it was described.
- Which query produced the impression. There is no query dimension here.
- Who else was in the answer, and whether a competitor was recommended above you.
- How many people clicked, because there is no click metric.
- Which of the two surfaces it came from.
The report is about your pages, and a generated answer is not about your pages. You cannot run an AI visibility programme on this number alone.
Source two: GA4, and where these clicks really land
GA4 has an AI Assistant default channel group, defined as the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. It matches when the medium is exactly ai-assistant or the referrer is on Google’s list of known AI assistant sources.
Clicks from AI Overviews and AI Mode do not go there. In the same documentation, Google defines the Organic Search channel to include Google’s AI Overviews and AI Mode. The channel group definitions say so directly.
Two consequences. A flat AI Assistant channel is not evidence that AI search sends you nothing, because Google’s own surfaces were never counted there. And you cannot isolate AI Overviews clicks inside Organic Search with the default channel group, since they sit in the same bucket as every ordinary result click.
Assistants outside Google are easier to see. ChatGPT appends utm_source=chatgpt.com to the links in its answers, and GA4 derives session source from utm_source when present, falling back to the referrer when it is not, so a visit with no referrer still gets attributed. Perplexity, Claude and Gemini generally do not append campaign parameters. Google’s AI surfaces give you no equivalent marker, which is why an impression count and a click count can never be joined into one honest number. Our GA4 integration reports the human visits arriving from AI assistants and keeps them apart from crawler traffic.
Source three: prompt-level tracking
A prompt-level tracker runs a fixed set of questions against the surface on a schedule and parses each answer: whether your brand was named, where it appeared, which competitors were named alongside or above you, what tone the mention carried, and which URLs were cited.
Building the prompt set decides whether the exercise is useful:
- Write the questions your buyers ask, not the keywords you rank for. “best crm for a 20 person sales team” is a prompt. “crm software” is a keyword.
- Seed it with real query data. Search Console knows what people type. LLM Pulse connects a property and pulls real queries by impressions into prompt generation, readable through the API and MCP server too.
- Set the locale deliberately. The same question in Spain and the United States returns a different answer with different sources.
- Keep the set stable. Changing prompts every month destroys the trend line. Add prompts, retire them rarely, annotate changes.
One metric here is specific to AI Overviews and easy to miss: coverage. Because the block does not trigger on every query, the share of your prompt set that returned a generated answer at all is worth watching on its own. Coverage falling while your mention rate holds is a different problem from losing ground inside the answers you do appear in.
LLM Pulse tracks Google AI Overviews and Google AI Mode alongside ChatGPT, Perplexity and Gemini, weekly by default with daily or monthly available, recording mentions, position, competitors, sentiment and cited sources per answer. If you are comparing options, we keep a working list of AI Overviews tracking tools that includes the ones we compete with.
The honest limitation: a tracked prompt set is a sample you chose. It does not reproduce real query volume and cannot tell you how many humans asked the question this week. That is why Search Console stays in the picture.
Putting the three together
| Question | Source | How often to look |
|---|---|---|
| Are our pages shown in Google’s generative features, and where? | Search Console Generative AI report, pages and countries | Monthly |
| How much traffic is Google search sending us overall? | GA4, Organic Search channel | Weekly |
| Are we named in the answer, and how early? | Prompt tracking | Weekly |
| Who gets recommended instead of us? | Prompt tracking, competitors and share of voice | Reviewed monthly |
| Which sources do these answers cite? | Prompt tracking, citations | Monthly |
| Did a generated answer trigger at all? | Prompt tracking, coverage | Weekly |
How to read a divergence
Impressions up, mentions flat. Your pages are pulled in as links while the prose names somebody else. Check which URLs gained impressions and whether those pages answer the question or merely rank for it. A page that ranks well and reads like a landing page gets paraphrased without your brand surviving into the sentence.
Mentions up, impressions flat. Either the answers name you but link to someone else’s page about you, which is a citation problem rather than a mention problem, or the prompts you track are not the queries that trigger the surface. Compare your prompt set against your Search Console queries.
Impressions up, organic clicks down. A familiar shape that this data cannot turn into causation. There is no click metric for the surface, and GA4 cannot separate those clicks from ordinary organic ones, so the correlation is as far as you can take it.
Everything flat, but your rankings moved. Check coverage first. If those queries are not producing a generated answer, ranking movement has nothing to do with the surface you are reporting on.
It is worth saying out loud in front of stakeholders that these sources cannot be reconciled into one number. Different time zones, different sampling, different definitions of an appearance. Anyone handing you a single blended figure has made a modelling decision they should be showing you.
What actually moves the number
Get cited. The pages that end up in generated answers usually answer a question directly, in the first screen, with facts that can be lifted out intact. Tracking which domains and URLs the answers cite shows where the answer comes from, and the fastest win is often a third-party page you can influence rather than your own. Our guide to tracking citations and sources covers that workflow.
Structure the page for extraction. Question-shaped headings, an answer under each one, tables where a table is the honest format, clean markup, no key fact locked inside an image.
Build off-site presence. You get named partly because other sources name you: review sites, comparison roundups, community threads, documentation. Slow work, and the part most teams underinvest in.
Read Google’s own guidance. Google publishes an AI features optimization guide, and its short version is that the fundamentals have not changed: crawlable, indexable, unique content worth surfacing. For the tactical layer, our guide on ranking in Google AI Overviews goes deeper, and which GEO metrics matter covers what belongs on the report.
One control to know about and leave alone. Search Console offers a Search generative AI control that lets a site opt out of having its content used in generative AI features. It takes you out of the surface altogether and does nothing for your placement inside it. Treat it as a publishing decision.
Common mistakes
Reading impressions as clicks. The report has no click metric. An impression is a link shown. Put it on a slide next to last quarter’s sessions and the room starts doing arithmetic that cannot be done.
Comparing a spot check today with one from last month. Whether the block triggers varies, and a manual search is a sample of one. Two screenshots six weeks apart is an anecdote. Scheduled runs over a stable prompt set make the comparison mean something.
Tracking only branded prompts. Your brand wins queries containing your brand name, which is not a finding. The questions that decide whether you get considered do not include your name, and those are the uncomfortable ones to look at.
Ignoring locale. Country and language change the answer, the sources and the competitive set. A global mention rate for a business selling in five markets hides the market where you are invisible.
FAQ
Does Search Console show AI Overviews separately from AI Mode?
No. The report covers both, and the dimensions are pages by canonical URL, countries, dates and devices. Nothing separates the two surfaces. If you need them apart, prompt-level tracking is the way.
Can I see clicks from AI Overviews?
Not in that report, which is impressions only. In GA4, clicks from AI Overviews and AI Mode land in the Organic Search channel under Google’s default channel group definition, mixed with ordinary result clicks, so there is no clean click figure for the surface.
Why is my GA4 AI Assistant channel near zero when Search Console shows generative impressions?
They measure different things. AI Assistant covers arrivals from ChatGPT, Gemini, Deepseek, Copilot and Grok. Google’s own AI surfaces are defined into Organic Search instead, so a quiet AI Assistant channel says nothing about AI Overviews.
How often should I look at each source?
Prompt tracking weekly, because that is where movement shows up first and where you can act on it. Search Console monthly for the trend, remembering its date buckets are Pacific Time. GA4 on whatever cadence you already report traffic. Reading it daily mostly gives you noise.
Do AI Overviews appear for every query?
No, which is why coverage belongs on your dashboard. Track what share of your prompt set produced a generated answer and watch it separately from your mention rate. A drop in coverage and a drop in mentions have different causes and different fixes.
Should I opt out of generative AI features?
Only if you genuinely do not want your content used, for legal or editorial reasons. The control removes your content from those features, which also leaves the Generative AI report with nothing to tell you. It has no effect on how you appear when you stay in.
