How to Measure AI Traffic: Making ChatGPT, Perplexity & Co. Visible in GA4 and with Peec

June 12, 2026 #AEO#Analytics#GA4#AI Search#Tracking#Measurement

In 2026, you don’t measure AI traffic from ChatGPT, Perplexity, Claude and the rest with a single number, but with a combination of four methods that back each other up. First, the native AI channel in GA4 (live since around May 2026), which recognizes referrals from ChatGPT, Claude and Gemini — but not Perplexity and Copilot reliably. Second, a specialist tool like Peec AI or OtterlyAI that checks for your brand directly in the engines’ answers. Third, monthly manual prompt baseline testing, where you ask the same industry questions in the chatbots and document whether and where you get cited. And fourth, analyzing your server logs to see whether the AI crawlers are fetching your pages at all. None of these methods is complete on its own. Together, they give you a reliable picture.

TL;DR

Measuring AI traffic precisely is still immature in 2026 — plan for a mix of methods, not a single number.
Since around May 2026, GA4 has had a native channel for AI assistants (recognizes ChatGPT, Claude, Gemini and others), but Perplexity and Copilot often slip through the cracks.
Specialist tools: Peec AI (multiple engines, near real-time), OtterlyAI (affordable entry point, multiple engines), HubSpot AEO Grader.
Run monthly prompt-pool baseline testing: check relevant purchase-intent prompts in ChatGPT, Perplexity and Claude — do you get cited, where, and which competitors show up?
Add referrer segmentation (hosts like chatgpt.com, perplexity.ai) and server log analysis of the AI crawlers (GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot).
AI traffic is smaller in volume than classic organic clicks, but it tends to convert higher — and it’s harder to attribute cleanly.

Why is AI traffic so hard to measure?

Because a large share of the value gets created without a click. When someone asks ChatGPT “Which agency in Munich does GDPR-compliant AI integration?” and you’re named in the answer, that person has seen you — but they won’t show up in any of your analytics reports unless they click a link. That’s the core problem: visibility in AI answers and measurable traffic to your site are two different things.

On top of that come three practical hurdles. First, many AI interfaces deliver messy referrer data or none at all; sometimes the click arrives with chatgpt.com in the referrer, sometimes via a redirect, sometimes anonymized. Second, the answers are non-deterministic — the same question gives a different answer today than it will next week, depending on the model version and context. A single sample tells you almost nothing. Third, the providers keep changing their link structure and crawling behavior, so every measurement stays a snapshot in time.

The conclusion is uncomfortable but important: anyone who promises you a single, exact “AI traffic figure” is oversimplifying. What’s realistic is a triangulated estimate from several sources — and that’s exactly what we’re assembling here.

What can the new GA4 AI channel do — and where does it stop?

Since around May 2026, Google Analytics 4 has had a native channel dedicated to AI assistants. It’s the most convenient starting point, because it works without any extra tool and automatically sorts referrals from the major chatbots into their own channel, instead of hiding them under “Referral” or “Direct” as before. Specifically, the channel recognizes referrals from services like ChatGPT, Claude and Gemini fairly reliably.

The limitation matters just as much as the feature: Perplexity and Microsoft Copilot are not reliably captured by the GA4 AI channel. Sometimes their clicks land in the AI channel, sometimes they don’t. So don’t rely on this one report to show your complete AI picture — it shows you the engines it recognizes, not all of them.

Here’s how to get the most out of it:

Use the channel as a starting point. Look at the AI channel in GA4 to get a first sense of scale for ChatGPT, Claude and Gemini clicks.
Tie it to conversions. Connect the channel to your goals (form, call, purchase). This is where it gets interesting, because AI traffic often converts above average.
Stay aware of the gaps. Treat the channel as a lower bound, not the total. Whatever Perplexity and Copilot bring in, you have to collect elsewhere (see referrer segmentation below).

Which specialist tools measure AI visibility?

Specialist tools tackle a different point than GA4: they don’t measure the click to your site, but whether your brand even shows up in the AI engines’ answers. That’s the part analytics fundamentally can’t see. Three tools are relevant in practice:

Tool	Strength	Best for
Peec AI	Covers multiple engines, works in near real-time	Ongoing monitoring of brand visibility across many prompts
OtterlyAI	Affordable entry point, multiple engines	Smaller budgets, a first structured overview
HubSpot AEO Grader	Quick visibility check	An initial baseline check without committing to a subscription right away

What these tools do at their core: they automatically send a series of prompts to the AI engines and log whether your brand gets mentioned, in what position, with what sentiment, and which competitors show up alongside you. That spares you the tedious manual querying and makes trends over time visible.

Honestly placed: these tools are very useful, but they’re not a replacement for the other methods. They tell you whether you get cited — not how much traffic and revenue comes out of it. And because AI answers fluctuate, they too need many data points over time to separate noise from a real trend. You don’t strictly need them to get started; a disciplined spreadsheet gets you surprisingly far. But once you want to measure systematically and across many prompts, they save you a lot of manual work.

How does manual prompt baseline testing work?

Prompt baseline testing means: you set up a fixed pool of real questions and run them monthly in ChatGPT, Perplexity and Claude to document whether and how your brand shows up. It’s the most honest, cheapest and surprisingly revealing method — and it works without any tool at all. It’s the foundation the more expensive tools merely build on.

Here’s how to do it:

Define a prompt pool. Collect 20 to 50 real questions from your industry — with a focus on purchase intent. So not just “What is Answer Engine Optimization?” but “Which agency in Munich does AEO for mid-sized companies?” or “Alternatives to provider X”. The uncomfortable prompts are precisely the valuable ones.
Query across multiple engines. Run each prompt in ChatGPT (with web search), Perplexity and Claude. The engines answer differently, and those differences are exactly where the insight lies.
Document in a structured way. For each prompt, note: Are you mentioned? In what position (first, in the list, on the margins)? Are you represented correctly? Which competitors get cited — and why perhaps more prominently than you?
Repeat monthly. Because the answers fluctuate and shift after model updates, only repetition is meaningful. A fixed monthly rhythm turns individual observations into a trend.

The effort is real but manageable — and the result is worth its weight in gold, because you don’t just see your own position, but also which competitors the engines prefer to cite and what content sits behind that. How to turn this into concrete action — writing in a citable way and sharpening your brand as an entity — is something we cover in detail in How to get your brand into the answers of ChatGPT and Perplexity.

How do I segment AI referrers in analytics?

With referrer segmentation, you capture exactly the AI clicks the GA4 AI channel misses — above all Perplexity. The idea is simple: you build a segment or filter in your analytics tool that specifically matches the hostnames of the AI services.

Relevant source hosts to filter on:

chatgpt.com and chat.openai.com (ChatGPT)
perplexity.ai (Perplexity — important, because GA4 doesn’t attribute it cleanly natively)
gemini.google.com (Gemini)
claude.ai (Claude)
copilot.microsoft.com (Copilot)

In practice, you create a dedicated “AI referrers” segment that groups these hosts in the source/referral field, then compare it with the native GA4 channel. The difference shows you what the channel misses — typically Perplexity and Copilot above all. Watch out for two pitfalls here: first, some AI clicks arrive without a referrer and land in “Direct”, so you won’t see them this way. Second, the providers occasionally change their domains and redirects, so review your filter list regularly. This method isn’t a silver bullet either, but it closes a specific gap in the GA4 channel.

What do server logs tell you about AI crawlers?

Your server logs show you the technical prerequisite for any AI citation: whether the AI services’ crawlers are fetching your pages at all. That’s the other side of the coin — not the human click, but the bot visit that lays the groundwork for you to appear in an answer in the first place.

Filter your logs for these user agents:

GPTBot — collects content for OpenAI’s model training.
OAI-SearchBot — feeds the web search in ChatGPT; especially relevant for visibility.
PerplexityBot — builds Perplexity’s search index.
ClaudeBot — Anthropic’s crawler for Claude.

What you learn from this: Which of these bots visit you at all, how often, and which pages do they fetch? If OAI-SearchBot never crawls your most important pages, it’s no surprise you’re missing from ChatGPT search — in that case the problem isn’t the content, it’s access. A common cause is overly aggressive bot protection or firewall rules that accidentally block legitimate AI crawlers. Here, server logs are your direct, unvarnished feedback on whether the technical foundation is in place — even before you start thinking about content.

Is AI traffic even worth it — given the small volume?

Yes, in most cases — because quality makes up for quantity. In 2026, AI traffic is significantly smaller in volume than classic organic clicks from Google. That’s the honest reality, and nobody should pretend AI search already replaces the regular search engine today. But volume is only half the story.

The decisive point: AI traffic tends to convert higher than classic organic clicks. That makes sense when you look at how it gets there. Someone who asks a specific question in ChatGPT or Perplexity, gets an answer featuring your brand and then deliberately clicks your link has already done much of their research. This person isn’t passing by at random — they arrive pre-qualified and with clear intent. Especially in the B2B mid-market, where decision-makers increasingly pre-research providers via chatbots, a click like that is often a very warm first touch.

The catch remains attribution. Because some AI clicks arrive without a clean referrer and disappear into “Direct”, your analytics almost always underestimates the actual AI impact. You see the conversion, but not always the correct source. So the rule is: evaluate AI traffic not by raw volume, but by conversion rate and the quality of the inquiries — and accept that part of the effect stays in the dark from a measurement standpoint.

For most companies, we at Rocket-Monkeys recommend a tiered mix that starts with little effort and only grows as needed. No single tool and no single channel is complete on its own — the meaning emerges only from the combination.

Here’s a sensible order to follow:

Activate and watch the GA4 AI channel. It costs nothing extra and gives you a first sense of scale for ChatGPT, Claude and Gemini.
Set up an “AI” referrer segment. This closes the gap on Perplexity and Copilot that the native channel misses.
Introduce monthly prompt baseline testing. The most honest method for understanding your visibility and competitive position — perfectly fine with a spreadsheet at first.
Filter your server logs for AI crawlers. This makes sure the technical foundation is actually in place.
Add a specialist tool when needed. Peec AI or OtterlyAI, once you want to automate prompt monitoring and scale it across many questions.

We’re honest about the limits: even this complete mix won’t deliver an “AI traffic figure” that’s exact to the decimal point. What it does deliver is a reliable, multiply-validated picture of where you stand in AI search, where it’s heading, and where the biggest lever lies. In 2026, that’s as far as you can responsibly go — and anyone promising the opposite is selling a false sense of precision.

If you want to know whether you even show up in ChatGPT, Perplexity and the rest, and how to make that measurable, let’s look at your GA4 channel, your referrer data, your logs and your prompt pool together. Write to us at info@rocket-monkeys.com for a no-obligation first conversation — we’ll tell you honestly which method gives you the most and where you can save yourself the effort.