How to See AI Traffic in Google Analytics (and Its Limits)

Yes, you can see some AI traffic in Google Analytics, but only the referrals. GA4 reports the click a human made from an AI answer to your site. It cannot see the crawler fetch that happened before the answer existed, because a crawler never runs the JavaScript tag that GA4 depends on. That blind spot is why "AI traffic" in GA4 and "AI visibility" are two different measurements.

What GA4 can show you

GA4's Traffic acquisition report is the main place to look. It lists sessions by source and medium. When someone asks ChatGPT, gets an answer that links to you, and clicks through, that session appears with a source like chatgpt.com or perplexity.ai. The medium is usually referral. You can filter for these sources and see pageviews, engagement time, and conversions from those visitors.

That is useful. It tells you when AI assistants send you people who act like visitors. They browse, they read, sometimes they buy. The report is a direct measurement of traffic value from AI answers.

The secondary path is the "Pages and screens" report with a secondary dimension of session source. That shows which of your pages get visited from AI referrers. You can stack that with goals or revenue if you have ecommerce tracking on.

So the first answer to "how to see ai traffic in google analytics" is: look at the referral traffic from the AI platforms' domains. Set up a filter or an exploration report for them and you have your baseline.

The limit: a crawler never loads your page

Everything GA4 knows comes from its JavaScript tag. A page loads, the tag runs, an event fires. That is the entire system. AI crawlers do not run that tag. They fetch the raw HTML, the same way a command line tool would. No JavaScript, no cookies, no events.

So when ChatGPT-User fetches your pricing page to answer a question for one person, GA4 sees nothing. It does not register a visit, a pageview, or any signal. The fetch existed, the answer quoted you, and a human clicked through later. GA4 only records the final click.

That means GA4 can tell you about demand that already converted into a visit. It cannot tell you about the exposures that came before. You are flying blind on the part of AI discovery that happens before the click, which is most of it.

What GA4 mislabels

Even the referral data is not clean. AI platforms route some clicks through redirect links, so the session source can appear as something generic instead of the AI assistant. You might see google.com or a service domain instead of chatgpt.com. Perplexity and Claude both use their own redirect schemes. The session is real, the attribution is fuzzy.

You also cannot distinguish between the types of AI interaction in GA4. There is no dimension for "this came from an indexing bot," "this came from a live answer," or "this came from a training crawl." GA4 does not know those categories exist. It only sees the final human session.

The three jobs an AI crawler does

To understand what GA4 misses, split AI bots into the three jobs they actually do. Kymo's bot taxonomy calls them answer, index, and train.

• Answer bots fetch a page live because one person asked a question right now. Examples are ChatGPT-User (OpenAI's live fetcher) and Perplexity-User.
• Indexing bots crawl ahead of time, like a search engine, so they can answer later. OAI-SearchBot, PerplexityBot, and Claude-SearchBot do this.
• Training bots bulk-harvest content for model training. GPTBot, ClaudeBot, and CCBot (Common Crawl) belong here.

GA4 sees none of these. They do not run JavaScript. Every single one of those fetches is invisible to your analytics tag.

How big is the blind spot really?

The table makes the pattern clear. GA4 is a traffic tool. It measures humans who arrive. It does not measure AI exposure that happens off your site, which is where most AI discovery now occurs.

This is the same limitation that affects regular SEO. Search engines crawl you, index you, rank you, and only then do you see a visitor in analytics. But with Google, the crawl is indirectly visible through Search Console. With AI engines, there is no equivalent dashboard. The crawlers are separate from the answer system, and the data is not published.

The "not provided" problem got bigger

Google moved keyword data behind encryption years ago. You saw visit counts but not the queries. The same thing is now happening with AI. You see the click after an AI answer, but you do not see the prompt that produced it, the pages the assistant actually read, or the other sources it compared you against.

If you rely on GA4 alone, you cannot see which of your pages AI assistants fetch most, what prompts drove those fetches, or whether the assistants are citing you at all in areas where you never get a click. A long answer can be surfaced to thousands of people and generate only a few clicks, because the answer already satisfied the user. GA4 records those few clicks and tells you nothing about the thousands of exposures.

What you can add without a new tag

There is one thing you can do today that costs nothing. Kymo offers a free AI visibility check at kymo.in/tools/ai-visibility-checker. It reads a URL from public signals only, so your site does not need Kymo installed. The report arrives by email through a magic link. It will show you which AI engines have fetched the URL and roughly how often. That is a useful complement to what GA4 tells you, and it takes under a minute.

There is also Search Engine Land style manual checking, but it is slow and does not scale. You can paste your URL into ChatGPT and ask what it knows, but that only catches the answer bots that are live right now. It tells you nothing about indexing crawlers like OAI-SearchBot or training crawlers like CCBot (Common Crawl).

The real thing GA4 cannot tell you

The larger gap is categories. An AI answer that fetches your page live is a different signal from an indexing crawl, which is again different from a training crawl. They have different commercial meanings.

A training crawl means your content was used to build a model. It sends no traffic and never will. An indexing crawl means the engine has stored your content for future answers. It can lead to referral traffic later. A live answer fetch means one person's question was answered using your content, right now, and that interaction is always attributable to a prompt someone is actually asking.

The AI visibility, explained doc at Kymo covers this in detail. But the shorthand is: if you only track clicks, you are measuring the last 10 percent of the funnel. The discovery happened before the click. The decision to cite your page happened before the click. The answer that influenced a buyer happened before the click, and GA4 will never see any of it.

What to do about it

Use GA4 for what it is good at. Track the AI referral sessions, measure their conversion rate, and compare them against other channels. That is real data and it has value.

Then close the gap. If you want a complete picture, you need server-side tracking that records crawler hits directly, because a server sees every request that the browser tag misses. Kymo does this with a cookieless identity system and a server-side receiver that classifies bots into answer, index, and training categories. It stores no raw IPs, rotates visitor identifiers daily, and keeps AI crawler user agents only on their own event rows so the classification stays auditable. The cookieless identity docs explain the mechanics.

This is also why the conversation has shifted from SEO to AEO. SEO vs AEO: Do You Need Both in 2026? makes the case that classic ranking still matters, but answer engines decide who gets found. And the distinction between AEO vs AIO matters less than the underlying measurement problem. You cannot improve what you do not measure, and GA4 does not measure the part that happens before the click.

If you want the plain-language version of why this all matters, The Small Business Guide to AEO covers the strategy. The short version is that a small business owner who ranked on Google for years now competes with AI answers that summarize ten sources instead of letting the user pick one. If you are not in that summary, you are invisible, and GA4 will not tell you.

Start with the free check

If you want to see which AI engines already know about your site, run the free AI visibility check on one of your key pages. It needs no Kymo installation because it reads public signals. The report arrives by email, and it will give you a starting list of which bots have visited.

If you want continuous tracking across every page, the paid plans start at $9 a month for 10,000 events, and AI crawler tracking does not count against your event limit. Both plans include every feature, so the choice is just event volume. Each plan starts with 14 days free, no card required. Exceed your limit and the dashboard pauses, but your data is not deleted.

Use both systems

GA4 is not useless, it is just partial. Use it for human traffic and referral conversions. Add a crawler tracker to see everything else. You genuinely need both, because AI engines are now a search channel, a referral source, and a content consumer at the same time, and no single tool built for the old web covers all three.

A mixed approach is the right one

Get your GA4 referral baseline set up first. Filter for the AI source domains, save those explorations, and start watching the trend month over month. Then run the free visibility check on your top five pages to see your current AI footprint. If the numbers matter to your business, do the continuous tracking.

Use GA4 if your traffic is healthy and you only care about clicks that convert. Add crawler visibility if you want to know which pages AI engines read, what they cite, and which prompts bring you exposure. The second choice is the one that tells you why the first one is happening.

Start free → 14-day free trial. No card required.