How to measure AI search visibility and AI-referred traffic, and what the data cannot show
Being named in AI answers and receiving visits from them are two measurements with different sources. What Google, Bing and your analytics report today, where each one stops, and a plan that fits on one page.
- There are two things to measure. Visibility is whether AI answers name or cite you. Traffic is the visits AI tools send and what those visitors do. They are recorded in different places.
- The platforms report impressions and citations, not outcomes. Google Search Console has a generative AI performance report for AI Overviews and AI Mode, and Bing Webmaster Tools has AI Performance. Neither documents the prompts people typed.
- AI-referred traffic is under-counted. ChatGPT tags its links with utm_source=chatgpt.com, but visits that arrive with no tag and no referrer are filed as direct, so an AI channel in GA4 is a floor.
- Prompt tracking works only as a share over many runs. Research published in January 2026 found less than a 1 in 100 chance of getting the same brand list twice, so report how often you are named, never a position.
“Are we showing up in AI answers?” is two questions. The first is about visibility: whether assistants and AI search features name or cite your company. The second is about traffic: whether those tools send visitors to your website, and what the visitors do once they arrive. Each has its own data source, and neither source is complete.
As of October 2026 the first-party reporting is better than it was a year ago. Google has a dedicated generative AI report in Search Console, Bing has an AI Performance report in public preview, and ChatGPT tags the links it sends. This article sets out what each of those sources reports, what none of them can see, and a plan that fits on one page.
It assumes you already work with Search Console and GA4. If you need the background on how AI search differs from classic search, start with what generative engine optimization is and how it differs from SEO.
Visibility and traffic are two different measurements
Visibility happens inside the answer. A buyer asks an assistant which vendors to consider, your name is in the reply or it is not, and in most cases nobody clicks anything. Traffic happens on your website: a person followed a link from an AI tool and started a session you can observe.
The two do not move together. A brand can be cited often and receive few visits, because the answer resolved the question. A brand can also receive AI-referred visits from a handful of prompts while being absent from most of its category. Reporting one number for “AI search” hides both situations, which is the same reason ChatGPT’s effect on Google traffic cannot be read from a single traffic line.
For a company spending on paid acquisition, the split matters for a practical reason. Visibility is an input you influence slowly and cannot fully observe. Traffic is smaller, but it is the part where you can measure a conversion and a return.
If your analytics shows almost no sessions from AI referrers and your category is rarely researched through assistants, a monthly check of the two platform reports is enough. The full plan below is worth the effort once AI tools are a visible source of sessions or your buyers tell sales they found you through one.
What Google Search Console reports about AI features
Google announced Search Generative AI performance reports on 3 June 2026 and states that, as of 31 August 2026, they are available to all websites worldwide. There is one report for Search and one for Discover. The help documentation says the Search report covers impressions in AI Overviews and AI Mode, and that an impression means a link to your site was shown to a user in one of those features.
The report groups data by pages, countries, dates and devices, with a search type filter for text and multimodal searches. Two results from your site in the same AI feature count as one impression at property level.
What the documentation does not list is as important. There is no query dimension, so the report does not tell you which questions triggered the answers you appeared in. It reports impressions, not clicks. And the documented dimensions do not separate AI Overviews from AI Mode. Google says in the announcement that it may add metrics over time, so check the help page before you rely on this list.
Clicks are handled elsewhere. Google’s guide to AI features (last updated 10 December 2025) states that AI Overviews and AI Mode are reported in the main Performance report within the Web search type. A click on a link inside either feature counts as a click, but it is added to the same total as every other organic click. You can see that AI features showed your pages. You cannot see, in Search Console, how many of your Google clicks came from them.
The same guide repeats a point worth keeping in mind when someone proposes a special AI file or markup: Google says there are no additional requirements to appear in these features. What earns a place is covered in how to optimize for Google’s AI Overviews, and nothing there is a guarantee of being cited.
What Bing Webmaster Tools reports about AI citations
Microsoft introduced AI Performance in Bing Webmaster Tools as a public preview on 10 February 2026. It reports how often your content is cited across Microsoft Copilot, AI-generated summaries in Bing and what Microsoft calls select partner integrations.
The useful addition over Google’s report is grounding queries. Microsoft describes them as a sample of overall citation activity, so treat them as a direction for content work, not a census. They are also the retrieval phrases, which are not necessarily the words your buyer typed.
Microsoft is explicit about limits. Average cited pages “does not indicate ranking, authority, or the role of any page within an individual answer”. The announcement describes citation metrics and does not describe click or session data, so the report tells you that Copilot used your page, not that anyone visited it.
The report was in public preview at launch. If Bing is a small share of your search traffic, the numbers will be small too. They are still the only first-party view of which phrases an AI system used to find your pages.
- Total citations
- Average cited pages
- Grounding queries
- Page-level citation activity
- Visibility trends over time
- Grounding queries are a sample
- Citations do not indicate ranking or role in the answer
How to isolate AI-referred traffic in GA4
Traffic measurement depends on what each click carries. OpenAI’s publishers and developers FAQ states that ChatGPT automatically adds the UTM parameter utm_source=chatgpt.com to referral URLs. No equivalent primary documentation was found for the other assistants when this article was written, so for those you rely on the referrer the browser sends: the domain the visitor came from.
Both signals can be missing. A link opened from a native app, or copied from an answer and pasted into a browser, arrives with no referrer and usually no tag. Analytics files that session as direct. How large that share is for your site is not something any tool reports.
utm_source=chatgpt.com
Clicks from Google’s own AI features are a separate case. They arrive from Google like any other organic click, so GA4 counts them as organic search. The channel below captures assistants, not AI Overviews or AI Mode.
Build the channel once
GA4 has no built-in AI channel. Google’s documentation on custom channel groups shows how to add one to a custom channel group and includes an example for AI assistants. The steps:
- In Admin, open channel groups and create a new group, which starts as a copy of the default one.
- Add a channel named “AI assistants” with the condition: source matches regex
chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai - Move the new channel above Referral and save.
- Use the group as the primary dimension in acquisition reports, with sessions, key events and session key event rate.
The pattern is a starting list of referrer domains, not an official one. Assistants appear and change domains, so check your referral report each quarter for sources the pattern misses. Standard GA4 properties allow two custom channel groups besides the default.
Common mistake: leaving the AI channel at the bottom of the list. Google’s documentation says traffic is placed in the first channel whose definition it matches, so a ChatGPT visit that also qualifies as a referral is counted under Referral and the AI channel stays empty.
Tracking mentions with a fixed prompt set
Neither Google nor Bing shows you the answers themselves, and assistants such as ChatGPT offer site owners no visibility report at all. The practical method is to write down the questions your buyers ask, run them against the assistants on a schedule, and record who was named. The method works, within limits that the research makes clear. A single run of a single prompt is an anecdote, and the share of runs that name you is a metric.
Those figures come from research by SparkToro and Gumshoe, published by Rand Fishkin and Patrick O’Donnell on 28 January 2026. The authors also measured how differently people phrase the same need, and found the wording of human-written prompts had very little in common.
Their conclusion is narrower than “prompt tracking is useless”. They state that visibility percentage “across dozens to hundreds of prompts run multiple times is a reasonable metric”, and that any tool claiming a ranking position in AI should not be trusted.
A single run of a single prompt is an anecdote, and the share of runs that name you is a metric.
In practice that sets three rules. Fix the prompt set and change it only by adding, with the date. Run every prompt several times per assistant on each check. Report the share of runs in which you are named, next to the same share for two or three competitors. For how a brand gets into those answers in the first place, see how to show up in an AI answer when you have never ranked for the question.
What the data cannot show
Three gaps remain after everything above is in place, and no vendor closes them.
- Mentions without a link. When an assistant names your company in plain text, nothing is recorded on your side. Platform reports count impressions and citations of links on their own surfaces only.
- Untagged visits. AI-referred sessions that arrive without a tag or referrer are counted as direct. A rise in direct traffic to deep pages can be a hint, but it is not proof.
- Personalised answers. Your prompt runs are not your buyer’s conversation. Answers vary with phrasing, history, location and the model version, so a prompt set estimates visibility for the questions you chose, not for the market.
A fourth gap is influence. Someone who read your name in an answer on Monday and typed your brand into Google on Thursday appears as branded search. No report connects the two events.
Present AI visibility as a trend against a fixed method, and AI-referred traffic as a minimum. Do not present either as market share, and do not forecast revenue from citation counts.
A measurement plan and what each metric misses
The plan takes about a day to set up and an hour a month to maintain. Its value comes from keeping the method fixed.
- Write 20 to 50 buyer prompts in the words customers use with sales and support. Store them with a version date.
- Run each prompt several times on each assistant your buyers use, on the same day each week. Record the share of runs that name you and each tracked competitor.
- Each month, export impressions and top pages from the Search Console generative AI performance report.
- Each month, export total citations, cited pages and grounding queries from Bing Webmaster Tools.
- Create the AI assistants channel in GA4 and report sessions, conversions and conversion rate for it, next to the site average.
- Put the five numbers on one page with a note on what each one misses, and review the prompt set and referrer list every quarter.
The prompt count in step 1 is a working range, not a research finding. The SparkToro and Gumshoe study supports “dozens to hundreds” of prompts with repeated runs. Fewer than that and one changed answer moves the share too much.
- Re-run the fixed prompt set
- Record the share of runs that name you
- AI impressions and cited pages from Google and Bing
- AI channel sessions, conversions and conversion rate
- Review the referrer list in the channel definition
- Add prompts, never edit existing ones
| Metric | Source | What it tells you | What it misses |
|---|---|---|---|
| AI impressions and pages | Search Console, generative AI performance report | How often and for which pages Google showed your links in AI Overviews and AI Mode | Queries, clicks from AI features on their own, a split by feature |
| Citations and grounding queries | Bing Webmaster Tools, AI Performance | How often Copilot and Bing AI summaries cited you, and sampled retrieval phrases | Clicks, position in the answer, anything outside Microsoft surfaces |
| Share of runs that name you | Your fixed prompt set, re-run on a schedule | Whether assistants name you for the questions you chose, against competitors | Real buyer phrasing, personalised answers, prompts you did not think of |
| AI channel sessions | GA4 custom channel group | Visits identified as coming from assistants, by landing page | Untagged visits filed as direct, Google AI feature clicks filed as organic |
| AI channel conversion rate | GA4 key events, or your on-site analytics | What identified AI visitors do, against the site average | Later visits by the same person through another channel |
What AI-referred visitors do after they land
Everything upstream of the click is partly hidden. Everything after it is yours to measure. Once a visitor from an assistant is on your website, you can see the landing page, what they read, and whether they completed your conversion event: a purchase, a demo request, a booking, an enquiry.
That is the part that decides return. Google notes in its AI features guide that clicks from results pages with AI Overviews are higher quality, meaning people are more likely to spend more time on the site. Pathmonk’s own customer data points the same way. In the AI Search case study, experience gifts retailer Tinggly recorded a 2.4% conversion rate from AI traffic against a 0.3% site average, in the same period in which its AI-referred sessions were compared with the previous 23 days. Across the Pathmonk GEO customers on that page, traffic from AI search rose 28% on average in the first three months.
Two cautions. The page does not state which conversion event the Tinggly rate refers to, and one retailer’s ratio is not a benchmark for your site. Run the comparison on your own data: AI channel conversion rate next to the site average, for one named conversion event.
Pathmonk’s AI Search analytics reports sessions and conversions per assistant and the conversion rate against the whole site, and its report states that AI referral is under-counted, so the figures are a floor. Because Pathmonk reads behaviour in the session with first-party, cookieless data, an AI-referred visitor with no history can still be shown the next step that fits their intent.
With a few dozen AI-referred sessions a month, a conversion rate for the channel is noise. Report the session count and the landing pages, and wait for volume before drawing conclusions about quality.
Key takeaways
- Measure visibility and traffic separately. One is recorded by the platforms and your prompt checks, the other by your analytics.
- Search Console has a generative AI performance report for AI Overviews and AI Mode. It documents impressions by page, country, device and date, not queries or clicks.
- Clicks from Google AI features are counted inside total Web search clicks and cannot be separated in Search Console.
- Bing Webmaster Tools AI Performance reports citations and sampled grounding queries for Copilot and Bing AI summaries, not visits.
- ChatGPT tags links with utm_source=chatgpt.com. Visits with no tag and no referrer are filed as direct, so an AI channel is a minimum.
- Track a fixed prompt set with repeated runs and report the share of runs that name you. A position in an AI answer is not a stable metric.
- Mentions without a link, untagged visits and personalised answers cannot be measured with any tool today.
- What AI-referred visitors do on your website is fully measurable, and it is what decides the return on being cited.
FAQs on measuring AI search visibility
How do I know if my company shows up in AI answers?
Use two sources. For Google and Bing surfaces, read the Search Console generative AI performance report and the Bing Webmaster Tools AI Performance report, which count impressions and citations of your pages. For assistants, run a fixed set of buyer prompts several times on a schedule and record the share of runs that name you.
Can Google Search Console show AI Overviews and AI Mode traffic separately?
Not clicks. Google documents a generative AI performance report that shows impressions for AI Overviews and AI Mode by page, country, device and date. Clicks from those features are counted in the main Performance report inside the Web search type, together with all other clicks. The documented dimensions do not include queries or a split between the two features.
How do I track ChatGPT traffic in GA4?
OpenAI states that ChatGPT adds utm_source=chatgpt.com to referral URLs, so those sessions appear with chatgpt.com as the source. Create a custom channel group with an AI assistants channel that matches the source against a list of assistant domains, and place it above Referral so the sessions are not counted there first.
Why does AI traffic show up as direct?
A visit is labelled by the tag in its URL or the referrer the browser sends. Links opened from a native app or copied and pasted into a browser can arrive with neither, and analytics files those sessions as direct. The size of that share cannot be measured, which is why AI-referred traffic figures are a minimum.
Are AI visibility tracking tools accurate?
They are useful for one thing: the share of repeated runs in which a brand is named across a fixed prompt set. Research by SparkToro and Gumshoe published in January 2026 found less than a 1 in 100 chance of two runs returning the same brand list, so a single run or a ranking position in an AI answer is not reliable.
What is the most useful AI search metric for a company that runs paid ads?
The conversion rate of identified AI-referred sessions for one named conversion event, next to the site average. Visibility figures are partial and slow to move. What visitors do after they land is fully measurable and is the figure that connects AI search to return.
- Google Search Central Blog: Introducing Search Generative AI performance reports in Search Console, 3 June 2026 (developers.google.com)
- Search Console Help: Generative AI performance report (Search), accessed 5 October 2026 (support.google.com)
- Google Search Central: AI features and your website, last updated 10 December 2025 (developers.google.com)
- Search Console Help: What are impressions, position and clicks, accessed 5 October 2026 (support.google.com)
- Bing Webmaster Blog: Introducing AI Performance in Bing Webmaster Tools public preview, 10 February 2026 (blogs.bing.com)
- OpenAI Help Center: Publishers and developers FAQ, accessed 5 October 2026 (help.openai.com)
- Google Analytics Help: Custom channel groups, accessed 5 October 2026 (support.google.com)
- SparkToro and Gumshoe: AIs are highly inconsistent when recommending brands or products, 28 January 2026 (sparktoro.com)