Short answer: to measure AI referral traffic in GA4, first open the default channel report and look at the AI Assistant channel Google added on May 13, 2026, which already groups visits from assistants such as ChatGPT, Gemini, DeepSeek, Copilot and Grok. Then build a custom channel group with an AI channel defined by a session-source regex you control (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai and the rest), place it above Referral, validate it against the source values your property has actually recorded, and report landing pages and key events for it. Finally, accept what analytics cannot see: clicks from Google's own AI Overviews and AI Mode land in Organic Search, stripped referrers land in Direct, and most answers produce no click at all. Search Console's generative AI report covers part of that gap with impressions.
The setup takes under an hour. The judgement around it takes longer, because the number GA4 shows you is a floor, not a total, and treating it as a total leads to bad decisions in both directions.
Key points
- GA4 now has a default AI Assistant channel. Use it as your first read, but you cannot edit it, Google does not publish the referrer list behind it, and it deliberately excludes AI Overviews and AI Mode.
- A custom channel group gives you a definition you can audit. Write the condition on session source, not medium: ChatGPT tags links with
utm_source=chatgpt.combut no medium, so medium-based rules miss it. - Order matters. GA4 puts a session in the first channel it matches, so the AI channel has to sit above Referral or it never receives anything.
- Validate the regex against your own list of session sources before you trust any trend, and re-check it every month.
- Read AI referral traffic next to impression data from Search Console and answer-side measurement. Clicks are the smallest of the three signals.
What counts as AI referral traffic, and what does not
"AI referral traffic" sounds like one thing. In practice it covers several kinds of visit that reach your analytics through different doors, and only some of them carry a label that says where they came from. Before you build anything, decide which of them you are trying to count, because the answer changes the configuration.
Three kinds of AI-influenced visit
The first kind is the labelled click. A person reads an answer in ChatGPT, Perplexity, Gemini, Copilot or Claude, clicks a cited link, and arrives on your page with a referrer header naming the assistant's domain, a UTM tag added by the assistant, or both. These visits are what GA4 can classify, and they are what this guide mostly configures.
The second kind is the unlabelled click. The person clicked a link in an assistant, but by the time the request reached your server the referrer had been dropped: the assistant ran inside a desktop or mobile app that sends none, the link carried a noreferrer instruction, the browser applied a strict privacy setting, or the person copied the URL and pasted it into a new tab. GA4 sees these as Direct, or occasionally as a referral from some intermediate page. Nothing in the hit says "AI".
The third kind is the influenced visit. The assistant named your brand without a link, or the person read the answer, closed it, and searched for you by name a day later. That session arrives as branded organic search or as Direct. It was caused by an AI answer, and no analytics configuration will ever attribute it to one.
A good AI referral setup counts the first kind accurately, estimates the second, and uses other instruments for the third. A bad one counts the first kind with a leaky regex and reports the result as "our AI traffic".
Why Google's own AI features sit in a different bucket
Google's definition of the AI Assistant channel says it excludes Google's AI Overviews and AI Mode, and its Organic Search definition says it includes them. That is a deliberate choice, and it has a practical consequence: a click from an AI Overview arrives from a google domain, exactly like a click from a classic blue link, and GA4 has no field that separates the two. If you hoped to see "AI Overviews traffic" as its own line in GA4, you will not. The closest instrument is Search Console's generative AI performance report, covered further down, and it reports impressions rather than clicks.
Keep this split in mind when someone compares "AI traffic" figures between tools. A report that adds AI Overviews into an AI number is measuring something GA4 cannot see in its standard data; a GA4 AI Assistant figure measures third-party assistants only. Neither is wrong, but they are not the same quantity.
How an AI click reaches GA4: referrer, UTM tag and session source
Every channel decision GA4 makes is based on a small set of values attached to the session: source, medium and campaign, plus derived dimensions such as source platform and, since June 11, 2026, source group. Understanding where those values come from is what lets you write rules that work.
The referrer header
When someone clicks a link, the browser usually sends a Referer header naming the page they came from. How much of that address it sends is controlled by the referring page's referrer policy. When a page sets none, browsers apply strict-origin-when-cross-origin: a link from one site to another sends only the origin, meaning the scheme and hostname without any path or query string, and sends nothing at all if the click goes from HTTPS to plain HTTP. A page can also choose no-referrer, which omits the header entirely.
For AI referral traffic this means two things. First, you will normally see the assistant's hostname, such as chatgpt.com or perplexity.ai, and never the conversation URL, so you cannot tell from the referrer which question produced the click. Second, any assistant or app that chooses no-referrer produces visits that look exactly like Direct. When GA4 receives a referrer from an unknown domain and no campaign parameters, it records the session with the referring hostname as source and referral as medium.
The utm_source=chatgpt.com tag
OpenAI documents a second signal for its own product: ChatGPT automatically adds utm_source=chatgpt.com to referral URLs, and OpenAI points publishers to analytics tools such as Google Analytics to track that traffic. Two details matter. The tag arrives on ChatGPT search clicks for sites that allow OAI-SearchBot, so a site that blocks that crawler should not expect to see much of it. And the tag sets a source but no medium.
Google's own guidance on campaign URLs warns that missing UTM parameters produce (not set) values in reports. In other words, a ChatGPT click that GA4 reads from the tag can show up as chatgpt.com / (not set) rather than chatgpt.com / referral. A rule written as "medium equals referral and source contains chatgpt" quietly misses it. A rule written on source alone does not.
Why source is the field to build on
Because both the referrer path and the tag path end in the same session source value, the source dimension is the one reliable meeting point. Medium varies between referral, (not set) and, for sessions GA4's own AI detection recognises, ai-assistant. Campaign is usually empty, except for GA4's own (ai-assistant) label. Source platform and source group are useful for reading, but you do not control how Google classifies them. Write your conditions on source, and use the other fields to sanity-check.
One more complication applies to any site with a checkout or login on another domain. If a visitor arrives from ChatGPT, leaves to a payment provider and comes back, the return can start a new session credited to the payment provider unless you have listed that domain as an unwanted referral (Admin, Data streams, your web stream, Configure tag settings, List unwanted referrals). That is not specific to AI traffic, but it hits AI traffic harder because the visit counts are small: a handful of misattributed returns can halve the conversions you see for the channel.
Step 1: Read GA4's default AI Assistant channel first
Before you configure anything, look at what Google already gives you. On May 13, 2026, GA4 started setting the medium to ai-assistant and the campaign to (ai-assistant) whenever the referrer matches a recognised AI assistant, and grouping those sessions under an AI Assistant channel in the default channel group. Google's definition names ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples.
Where to look
Open Reports, then Acquisition, then Traffic acquisition. The table's primary dimension is "Session default channel group"; AI Assistant appears as a row once the property has received such sessions. Add "Session source / medium" as a secondary dimension to see which assistants make up the row. For a longer view, build a free-form exploration with "Session default channel group", "Session source" and "Landing page + query string" as dimensions and Sessions, Engaged sessions and Key events as metrics, filtered to the AI Assistant channel.
Two checks are worth doing on the first visit. Compare the AI Assistant row with the Referral row filtered to known assistant hostnames: if perplexity.ai or claude.ai still appear under Referral, the default list did not recognise them in your data. And check the date where AI Assistant sessions begin. Google's announcement does not say whether the channel was applied to data recorded before May 13, 2026, so do not compare a period before that date with one after it and call the difference growth.
What the default channel cannot do
The default channel group cannot be edited. Its definitions are not case sensitive and, in Google's words, may evolve as the market changes. The help page links a downloadable list of sources and categories for the other channels; when we checked it, the list had four categories (search, shopping, social and video) and no AI category, so the referrer list behind AI Assistant is not published there. That is acceptable for a quick read and a problem for reporting: if a number moves, you need to know whether visitors changed or the definition did.
| Question | Default AI Assistant channel | Your custom AI channel |
|---|---|---|
| Who defines which sources count | Google; list not published | You; the regex is in the property and in your notes |
| Can you add a new assistant | No, you wait for Google | Yes, edit the regex |
| Can the definition change without notice | Yes, Google says it may evolve | Only when someone on your team edits it |
| Applied to historical data | Not stated by Google | Yes, custom groups apply retroactively |
| Includes AI Overviews and AI Mode | No, those are Organic Search | No, they arrive from Google domains either way |
| Cost | None | One of the 2 custom groups a standard property allows |
| Best use | First read, cross-check | Reporting, trends, landing-page analysis |
The sensible arrangement is to use both: your custom channel for the numbers you report, and the default channel as an independent second opinion. When they disagree, the disagreement is information. Either your regex is missing a source Google recognises, or Google is classifying something you decided not to count.
Step 2: Build a custom channel group for AI referral traffic
A standard GA4 property can hold 2 custom channel groups alongside the predefined one, and a 360 property can hold 5. Each group can contain up to 50 channels. Custom groups apply to reports retroactively, which is the main reason to build one: you can see how your AI channel looked last year under today's definition. You need the Editor role or above at property level.
Copy the default group instead of starting empty
Go to Admin, then under Data display choose Channel groups, then Create new channel group. GA4 copies the default group as the starting point. Keep that copy: every other channel already has a definition, and you only want to change one thing. Name the group something that will still make sense in a year, such as "Default + AI referrals (v1, 2026-10)", and use the description field to record what changed and why. Version numbers in the name sound fussy until the day someone asks why the AI line jumped in March.
Write the AI condition on session source
Click Add new channel, name it "AI referrals" (avoid the exact name AI Assistant, so nobody confuses it with Google's channel), and add a condition on Source with "matches regex". GA4's regex is a full match by default and case sensitive: without .* or anchors that allow surrounding text, the expression must match the whole value. That is useful here, because it lets you list exact hostnames instead of fragments that over-match.
^((www\.)?chatgpt\.com|chat\.openai\.com|(www\.)?perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|(www\.)?claude\.ai|chat\.deepseek\.com|(www\.)?grok\.com|chat\.mistral\.ai|(www\.)?meta\.ai)$
Read it as a list. Each alternative is one hostname, the optional (www\.)? covers the variants some assistants serve from, and the anchors make sure gemini.google.com matches while gemini.google.com.example.net does not. Add or remove hostnames to match the assistants your audience actually uses; the list above is a starting set, not a standard. If your validation in Step 3 turns up other forms, such as an app identifier or a hostname with a different subdomain, add them explicitly after confirming what they are.
Because ChatGPT's tag and the referrer both resolve to chatgpt.com as source, this one condition covers both paths. Do not add a medium condition. If you want to see the split, add Session medium as a secondary dimension when you read the report.
Why not paste Google's example regex
Google's custom channel group page includes a worked "AI assistants" example with a ready-made regex. It is a reasonable illustration and a weak production rule. It is built from broad fragments such as .*gemini.*, .*gpt.* and .*copilot.*, plus a leading alternative ^.*ai that matches any source ending in "ai". Fragments like these catch things you did not mean: a referral from the gemini.com cryptocurrency exchange contains "gemini"; a newsletter tagged utm_source=copilot-weekly contains "copilot"; any partner whose domain ends in ".ai" matches the first alternative. Claude is mentioned on the page as an example assistant, but the regex has no "claude" alternative, so claude.ai is only caught by the catch-all ending. And because the match is case sensitive, a hand-tagged utm_source=ChatGPT link slips past every lowercase fragment.
None of these problems shows up as an error. They show up as a channel that is a few percent too large or too small, which for a small channel is the whole signal. Exact hostnames take five minutes longer to write and save you from defending a number you cannot explain.
Put the AI channel above Referral
GA4 evaluates channels in order and places a session in the first channel whose definition it matches. A session from perplexity.ai with medium referral matches the Referral channel's rule as well as yours, so whichever comes first wins. Google's own example says to reorder the list so the AI channel appears above Referrals. Use Reorder, drag "AI referrals" above Referral, and click Apply.
It is worth going further and placing it above Unassigned and Direct too. Direct only matches when the source is exactly (direct), so there is no real conflict, but keeping AI near the top of the list makes the intent obvious to the next person who edits the group. Leave the paid channels above it: if you ever run ads that land with an assistant's hostname as source, you want them counted as paid.
Make the group primary, or do not
GA4 lets you set a custom group as the primary channel group, which changes the channel dimension that appears in standard reports and in some integrations. If your team reads standard reports and will not remember to switch dimensions, making it primary is the way to get the AI line in front of them. If other teams depend on the standard reports matching Google's default definitions, leave it secondary and read it through the "Session" version of your custom group's dimension in explorations. Either choice is fine; deciding it explicitly is what matters.
Step 3: Validate the regex against your own session sources
A channel definition is a claim about your data. Test it on your data before you report anything with it. This step is where most AI referral setups fail, because the regex looks right and nobody checks what it actually matched.
Pull the full list of sources
Create a free-form exploration covering as long a date range as your retention allows. Use "Session source" as the only row dimension and Sessions as the metric, and set the row count to show everything. Export it. Then filter the export, in a spreadsheet, for anything that looks AI-related: hostnames containing "gpt", "openai", "perplexity", "gemini", "copilot", "claude", "deepseek", "grok", "mistral", "meta.ai", or ending in ".ai". This is deliberately broader than your regex. The point is to find sources your regex misses and sources it should not match.
Now build a second exploration with two dimensions, "Session source" and your new custom channel, and filter to the AI channel. Every row here is something your regex claimed. Compare the two lists. Each hostname should land in exactly one of three bins: correctly matched, wrongly missed, or wrongly matched.
A test table you can keep
Turn the comparison into a small table and keep it with the channel group's documentation. When you edit the regex later, re-run the same cases.
| Session source value | Should it count as AI? | Exact-hostname regex | Broad-fragment regex |
|---|---|---|---|
| chatgpt.com | Yes | Matches | Matches |
| perplexity.ai | Yes | Matches | Matches |
| claude.ai | Yes | Matches | Only via the catch-all ending |
| gemini.google.com | Yes | Matches | Matches |
| gemini.com (crypto exchange) | No | No match | Wrongly matches |
| copilot-weekly (newsletter tag) | No | No match | Wrongly matches |
| partner-tool.ai (a SaaS referral) | No | No match | Wrongly matches |
| ChatGPT (hand-tagged, capitalised) | Yes | Missed: add it or fix the tag | Missed |
| google / organic from an AI Overview | Not separable in GA4 | No match | No match |
The capitalised case deserves its own note. If your team, or a partner, tags links to your site with utm_source=ChatGPT for a campaign, that value is a different string from chatgpt.com. Decide whether hand-tagged campaign links in AI tools belong in your AI channel. Usually they do not: they measure your own campaign, not organic discovery by an assistant. Keep them in their campaign's channel and fix the tagging convention instead.
Cross-check against the default channel
Once your custom channel looks clean, compare its totals with the default AI Assistant channel for the same period after May 13, 2026. Expect them to be close but not identical. A large gap in one direction tells you where to look: if the default channel is bigger, it recognises a source your regex lacks, and the source / medium breakdown of the default channel will name it. If your channel is bigger, your list includes hostnames Google does not treat as assistants, which is fine if you chose that deliberately.
Re-check on a schedule
Assistants launch new hostnames, merge products and change how links are built. Google's own example warns that the regex should be updated when assistant URLs or the list of assistants change. Put a monthly reminder on the team calendar: export the source list for the past month, filter it with the broad search terms above, and look for anything new. It takes ten minutes and it is the only thing that keeps a hand-maintained channel honest.
Step 4: Build reports that answer real questions
With a validated channel, the temptation is to put one line on a dashboard. A single "AI sessions" line answers almost nothing. The questions people actually ask are which pages AI tools send visitors to, whether those visitors do anything, and how the channel compares with search. Build for those.
Landing pages: what AI referral traffic lands on
Use an exploration with your AI channel as a filter, "Landing page + query string" as rows and "Session source" as columns. This tells you which pages assistants are citing in answers people click. On most sites the list is short and surprising: a few deep pages, often documentation, comparisons, pricing or a single explainer, rather than the pages the marketing team considers important.
Survey data suggests the same unevenness. In WordStream's 2026 survey of small businesses, among businesses that monitor AI referrals, the home page was most often named as the page AI referrals land on, followed by product or services pages and contact pages, while FAQ pages were among the least likely. Self-reported answers are not analytics, and a business's impression of where visitors land is shaped by what it looks at. Treat those answers as a reason to check your own landing pages, not as a benchmark to match; Figure 3 shows the exploration that does the checking.
Two practical notes. Strip query strings in a second view, because the ChatGPT tag adds ?utm_source=chatgpt.com and some assistants append their own parameters, which splits one page into several rows. And look at the page title as well as the path: AI tools often cite pages by their content rather than their place in your navigation, so a page that "nobody visits" can be one of your top AI landing pages.
Engagement and key events
Add Engaged sessions, Average engagement time per session and your key events to the same exploration. Compare them with Organic Search for the same landing pages, not across the whole site, because the two channels land on different pages and site-wide averages mostly measure that difference. If the AI channel converts better on a page, the likely reason is that the answer already did the comparison work for the visitor. If it converts worse, the page may be cited for a fact while the next step is buried.
Keep the absolute numbers in view. On most sites AI referral traffic is still a small channel, and a conversion rate computed from a few dozen sessions moves by large amounts when one person buys. Report counts next to rates, and do not act on a rate until the count behind it is large enough that one conversion does not change the conclusion.
Comparing AI referral traffic with organic search
The comparison that tells you something is page-level: for each page that receives AI referrals, what share of its traffic comes from the AI channel, and how has that share moved over the last quarter? A rising AI share on a page with flat organic traffic suggests assistants are citing it more. A rising AI share because organic fell is a different story, and the two look identical on a site-wide chart. Build the page-level table, sort by AI sessions, and annotate the top rows with what changed on the page and when.
Resist turning AI referral sessions into a percentage of total traffic and calling it "AI share". The denominator includes Direct, which contains some of the AI visits you could not label, so the ratio understates AI by an unknown amount and moves whenever Direct moves.
What GA4 cannot see: dark AI traffic and stripped referrers
Everything above improves the count of labelled clicks. The larger question is how big the unlabelled part is, and the honest answer is that GA4 cannot tell you. What you can do is understand the mechanisms, estimate their direction, and use instruments that do not depend on clicks.
Stripped referrers
Any assistant that serves its answers inside a native app, or sets no-referrer, or opens links in a way that drops the header, produces visits that arrive with no referrer. If those links also carry no UTM tag, GA4 records the session as Direct. You can sometimes see the effect indirectly: Direct sessions landing on deep, rarely bookmarked pages, such as a specific documentation article or a comparison page, on the same days that AI-referred sessions to that page rise. That pattern is suggestive, not proof, and it is worth noting in your report rather than adding to the AI number.
Copy, paste and share
People copy URLs out of answers and paste them into a new tab, a chat message or an email. The browser sends no referrer for a typed or pasted address, and a link shared in a messaging app often loses whatever tag it had. Those visits arrive as Direct or as a referral from the messaging service. This is the same "dark social" problem analytics has had for years; assistants add another source of it.
Unlinked mentions and later searches
Many answers name brands without linking them, and many people read an answer and never click. Pew Research Center's browsing study of 900 U.S. adults in March 2025 found that when a Google search showed an AI summary, users clicked a traditional search result on 8% of visits, against 15% when there was no summary, and clicked a link inside the summary on only 1% of visits. That study measured Google's AI summaries rather than chat assistants, but the mechanism is the same: the answer is read in place. When those people later search for you by name or type your address, the visit is credited to Organic Search or Direct.
How to estimate the gap without inventing a number
You will be asked for a multiplier, such as "true AI traffic is three times what GA4 shows". Decline to give one. No published method supports a general multiplier, and the gap varies by audience, assistant mix and page type. Use triangulation instead:
- Self-reported attribution. Add "How did you hear about us?" with an option for AI assistants to sign-up, demo and checkout forms. Compare the share who pick it with the AI channel's share of the same conversions. The ratio is a rough local measure of under-counting for that conversion type, and it is yours.
- Branded search trend. In Search Console, track impressions and clicks for queries containing your brand name. A rise that coincides with growth in AI mentions, with no campaign to explain it, is consistent with influenced visits.
- Deep-page Direct. As above, watch Direct sessions to pages nobody bookmarks. Report the trend, not a converted number.
- Server logs for user-triggered fetches. When an assistant fetches a page because a user asked, the request comes from an agent such as ChatGPT-User, Perplexity-User or Claude-User. Those requests are not visits and are not in GA4, but counting them per page tells you which pages assistants are reading on behalf of users, which is a leading indicator of the pages that might be cited.
Present these as context around the GA4 number, labelled as estimates. The GA4 figure stays the auditable floor.
Search Console's generative AI report: the impression side
For Google's own AI features, the instrument is not GA4 but Search Console. Google announced its Search Generative AI performance reports on June 3, 2026 and says they were rolled out to all websites worldwide as of August 31, 2026. The Search version counts impressions: how many times links to your site were shown in a generative AI feature on Google Search, such as AI Overviews and AI Mode. Multiple results from your site in one response count as one impression in the chart totals.
What the report gives you, and what it leaves out
The help page lists Pages, Countries, Dates and Devices as dimensions, a search-type filter that separates text-based and multimodal web results, and a URL filter. Dates can be read hourly, daily, weekly or monthly. The report has an export button, is subject to the standard 1,000-row limit, excludes Search Labs experiments, and may not appear at all for a site without enough generative AI impressions. Google also says these impressions are included in the overall Performance report, so they are a subset, not an addition.
It lists no click metric. That is the most important limitation for anyone trying to connect the report with GA4: you can see how often you were shown in AI Overviews and AI Mode, and you can see Organic Search sessions in GA4, but you cannot join the two to compute an AI Overview click-through rate from Google's own data.
| GA4 AI channel (custom or default) | Search Console generative AI report | |
|---|---|---|
| What it counts | Sessions that arrived from third-party assistants with a recognisable source | Impressions of your links in Google's AI Overviews and AI Mode |
| Covers ChatGPT, Perplexity, Claude, Copilot | Yes, when the referrer or tag survives | No |
| Covers AI Overviews and AI Mode | No, those clicks are in Organic Search | Yes |
| Clicks | Yes, as sessions | No click metric listed |
| Conversions | Yes, via key events | No |
| Page-level view | Landing page | Pages dimension and URL filter |
| Main blind spot | Stripped referrers, unlinked mentions | Everything outside Google; no clicks |
Using the two together
Export the generative AI report's Pages table and your GA4 AI landing-page table for the same month, and put them side by side. Pages that appear in both are being surfaced by Google's AI features and clicked from third-party assistants: those are your most-cited pages across engines. Pages with many AI impressions in Search Console and no AI referral sessions in GA4 are visible in Google's answers but not chosen elsewhere, or not clicked. Pages with AI referral sessions and no generative AI impressions are being picked up by assistants that do not use Google's index the same way. Each group suggests a different next step, and none of them is visible from either tool alone.
For a longer treatment of what Google's numbers do and do not show, see our analysis of the figures Google published about its AI search tools.
A monthly AI referral traffic review in ten steps
A configuration nobody reads decays. The review below fits in an hour once a month and keeps the numbers defensible. Run it in the same order each time, so the notes are comparable.
- Export the source list. Sessions by Session source for the past month. Filter for AI-looking values with the broad search terms from Step 3.
- Update the regex if needed. Add any new assistant hostname, record the edit in the group description, and bump the version in the name.
- Compare with the default channel. Custom AI channel against GA4's AI Assistant channel for the month. Note the gap and its cause.
- Check unwanted referrals. Look for payment, login or booking domains appearing as sources of returning sessions, and add them.
- Read the landing-page table. Top pages by AI sessions, with engagement and key events next to Organic Search for the same pages.
- Read Search Console's generative AI report. Export Pages for the month; mark the pages that appear in both lists.
- Read the self-reported field. Share of conversions that chose "AI assistant", against the AI channel's share of the same conversions.
- Check branded search. Impressions and clicks for brand queries, month on month.
- Annotate. Write down what changed on the site, in the regex and in the assistants. A trend without annotations is a guess.
- Report the floor and the context separately. The GA4 count is a measured floor. Everything else is context, and it is labelled that way.
This review is part of a wider measurement framework; our guide to measuring GEO explains how referral data fits with answer-side measures such as mention rate and citations, and measuring change over time covers how to tell a real trend from noise in small numbers.
Many organisations are still at the start of this. In the same WordStream survey, half of respondents said they monitor AI referrals and mentions, and the share rose with business size; among the largest businesses, difficulty tracking and attributing AI referral traffic was named as the top GEO challenge.
Before AI referral traffic exists: make sure assistants can cite you
The setup above measures clicks. Clicks only exist if an assistant can read your pages and chooses to link them. If the AI channel stays close to empty after a few months, the cause is more often upstream than in the regex.
Access. OpenAI ties the utm_source=chatgpt.com tag to ChatGPT search, and ChatGPT search depends on OAI-SearchBot being allowed to read your site. Many sites that meant to opt out of model training blocked every OpenAI agent at once, and with it their own ChatGPT search presence. Check your robots.txt and your CDN rules agent by agent; our guide to AI crawler access walks through the roles and the checks.
Answer-side measurement. Because most answers are read without a click, the most direct measure of AI visibility is the answer itself: run a fixed set of real buyer questions through each assistant on a schedule, and record whether your brand is mentioned and whether your pages are cited. That number moves before referral traffic does, and it explains changes that analytics only records. Referral traffic then becomes what it should be: the confirmation that visibility in answers is turning into visits, on the pages where it matters.
An empty channel. If your validated AI channel shows almost nothing, do not conclude that assistants ignore you. Check, in order: whether OAI-SearchBot and the other search agents can fetch your key pages; whether your brand is mentioned in answers to the questions your buyers ask; whether those answers cite your pages or other people's pages about you. Each "no" points to a different fix, and only the last one is a matter of content. Products such as InsightWonder run the answer-side checks on a schedule, but the order of questions applies whatever tool you use.
Sources
- Google Analytics Help, Custom channel groups (limits, retroactivity, evaluation order, AI assistants example)
- Google Analytics Help, Default channel group (AI Assistant and Organic Search definitions, linked source list)
- Google Analytics Help, What's new in Google Analytics (AI Assistant channel, May 13, 2026; Source Group, June 11, 2026)
- Google Analytics Help, About regular expressions
- Google Analytics Help, URL builders: collect campaign data with custom URLs
- OpenAI Help Center, Publishers and Developers FAQ (utm_source=chatgpt.com)
- MDN Web Docs, Referrer-Policy
- Search Console Help, Generative AI performance report
- Google Search Central Blog, Introducing Search Generative AI performance reports in Search Console (June 3, 2026)
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results (July 22, 2025)
- WordStream by LocaliQ, Small Business Website Trends Report (last updated March 24, 2026)