How to track actual AI traffic to your website (not just AI rankings)

There are roughly a dozen tools selling "AI search visibility" right now. Most of them do the same thing: query ChatGPT, Gemini, Perplexity and Claude with prompts about your brand, score whether you appear, and call that an AI ranking.

It is useful data. It is also a projection.

A projection tells you what the model would say if asked the prompt you chose, at the moment you ran the query, in the country you targeted. It does not tell you whether a real human asked anything close to that prompt, or whether anything that happened in that AI conversation drove a single person to your site.

The traffic itself is the proof. And almost nobody is showing it.

This piece covers why, what the gap looks like, and how to track actual AI traffic to your website yourself - whether you use Cluo or not. The setup is free, takes about fifteen minutes, and works in Google Analytics 4 today.

The two kinds of AI search data

There are two completely different questions a business owner can ask about AI search, and they need different tools to answer them.

Question one: am I being mentioned in AI answers? This is AI rank tracking. You define the prompts you care about ("best plumber in Bondi", "compare childcare centres central Auckland"), the tool runs them against ChatGPT, Gemini, Perplexity, Copilot and Claude on a schedule, and reports whether your brand appears, where it appears, and what the answer said. This is what Otterly, Peec, Profound and the AI-tracking add-ons in Semrush and Ahrefs sell.

Question two: are AI tools actually sending people to my website? This is AI referral tracking. The AI engines render their answers as web pages. When a user clicks a citation link in a ChatGPT answer, or a source pill in a Perplexity response, or a footnote in a Gemini summary, their browser visits your site with a referrer header pointing back to chatgpt.com or perplexity.ai or copilot.microsoft.com. That referral lands in your analytics like any other inbound visit.

These two questions overlap, but they do not answer each other.

A high AI ranking with zero AI traffic means the model knows you exist but is not handing real users a clickable link to you. A solid stream of AI referral traffic with weak AI rankings on the prompts you track means the model is citing you for queries you have not measured. Both happen. Neither tool catches both.

Why most tools only show you the projection

It is much easier to build a ranking projection tool than a referral tracker.

To track AI rankings, you need an API key for each model, a prompt list, a country setting and a daily cron job. The tool runs the prompts, parses the answers for brand mentions and citations, and writes the results to a dashboard. It is self-contained. The customer never has to connect a single thing belonging to their own website.

To track AI referral traffic, you have to connect to the customer's actual analytics. That means a Google Analytics integration, an OAuth flow, permissions handling, a way to detect AI domains in the source/medium fields, and the ongoing work of keeping the AI source list current as new engines launch and existing engines change their referral behaviour. It is real engineering and it sits on top of the customer's own GA4 property.

Most AI tracking vendors have not done that work. They sell the projection side because it scales without per-customer onboarding and ships in a quarter, not a year.

The result: business owners get a number that says "you appeared in 1 of 6 AI answers this week" and have no idea whether that translated into anyone visiting their site. The data they actually need to make a budget decision is one connector away, and nobody is plugging it in for them.

What AI referral traffic looks like in GA4

Every AI search engine that links out to your site sends a referral that GA4 can detect. Some examples of the source/medium pairs you will start seeing once AI traffic picks up:

• chatgpt.com / referral
• chatgpt.com / (not set)
• perplexity.ai / referral
• copilot.microsoft.com / referral
• gemini.google.com / referral
• claude.ai / referral
• grok.com / referral
• meta.ai / referral
• you.com / referral
• phind.com / referral

The list grows. As new engines launch and existing ones change how they render citations, new domains turn up in the source/medium report. The discipline is to keep watching for them and not assume the list is fixed.

By default, GA4 buckets all of these under either "Referral" or "Unassigned" in the standard channel groupings. That is the problem. The default channel groupings predate generative AI and have not been updated to recognise these sources as a distinct traffic channel. Until you create a custom channel grouping, your AI traffic is hidden inside two of the noisiest buckets in GA4.

How to set up AI referral tracking in GA4 (free, fifteen minutes)

Anyone can do this in their own GA4 property. No vendor required.

Step 1: Open the Channel Group editor

In GA4, go to Admin (gear icon, bottom left), then Data display, then Channel groups. Click "Create new channel group". Name it "Channel grouping with AI".

Step 2: Add a rule for AI Search

At the top of the rule list, click "Add new channel". Name the channel "AI Search". The rule should match traffic where the Source contains any of the AI domains. Set the condition to: Source contains chatgpt.com OR perplexity.ai OR copilot.microsoft.com OR gemini.google.com OR claude.ai OR grok.com OR meta.ai OR you.com OR phind.com OR poe.com OR character.ai.

This rule has to sit above the default "Referral" rule, otherwise GA4 will catch the traffic in Referral first and your AI Search channel will stay empty. Order matters.

Step 3: Save and use the new grouping in reports

Save the channel group. Then in any report that supports a Channel Group dimension (Acquisition, User acquisition, Traffic acquisition), change the dimension from "Default channel group" to "Channel grouping with AI". The AI Search channel will appear with its own row.

Step 4: Set up a comparison

If you want to track AI Search separately from organic and paid, add it as a comparison in your standard acquisition reports. Now you can see week-on-week AI traffic growth in the same view as your other channels.

That is the entire setup. It is reversible, it does not cost anything, and it works on existing data the moment you save the rule.

What the numbers usually look like

For most small and mid-sized businesses, AI search traffic is currently a small share of total inbound but a fast-growing one. A few patterns we see across the businesses on Cluo Portal:

• AI Search rarely exceeds 5% of total sessions today, but is growing roughly 30-50% quarter on quarter for sites with strong topical authority.
• ChatGPT is the dominant source for almost every account, followed by Copilot (because it pulls from Bing's search index), then Perplexity. Gemini referrals are growing but still small.
• Conversion rates from AI referral traffic tend to be higher than from generic organic, because the user has already had a recommendation. They are not researching - they are shortlisting.
• Bounce rates are mixed. Some AI engines deep-link to interior pages that answer a specific question. The user lands, gets their answer, leaves. That is not a bad outcome - it is the new shape of SEO.

These are not universal numbers. A B2B consultancy will see very different patterns from a local tradie or an ecommerce store. The point of setting this up is to find out what your numbers look like, not to compare to a benchmark.

The trap of ranking-only data

We have watched accounts where the AI rank tracker says everything is great and the referral data says nobody is showing up.

The most common reason is that the prompts in the rank tracker are the prompts the business owner thinks customers ask. The prompts that actually drive traffic are different - they are longer, more specific, more local, and often include comparison language ("X vs Y in Sydney", "alternatives to X under $100"). A tracker tuned to short brand prompts will miss the long-tail queries that send the actual clicks.

The other common reason is that the model is citing the business in the answer text but not linking to it. ChatGPT and Gemini both do this - they reference a brand in plain text without surfacing a clickable source. The ranking tool counts it as a mention. The referral data shows nothing. The user reads the mention and then either Googles the brand separately or moves on.

If you only watch ranking data, you can be top of mind in AI answers and still get zero traffic. If you only watch referral data, you can be sending real visits and have no idea which prompts you are winning on. You need both.

How AI rankings and AI referrals work together

The right way to read AI search data is as a loop.

1. Rankings tell you where you appear. Which prompts are you cited on, which competitors appear alongside you, which engines list you and which do not.
2. Referrals tell you what is converting. Which AI domains are actually sending traffic, which landing pages they hit, what users do once they arrive.
3. The gap between them is your work list. A prompt where you rank well and get no traffic means your citation is not clickable - you may need to rework the answer or the landing page. A referral source you cannot tie back to any tracked prompt means there is a query you should be tracking. Add it.

This is the framework Cluo Portal runs on. Rankings sit on the left of the dashboard. Referrals sit on the right. Both update daily. The reconciliation between them is what makes the data useful instead of decorative.

You can run the same loop manually if you do not want to use Cluo. Pick a ranking tool, set up the GA4 channel group above, and once a week compare the two views side by side. The signal will be obvious within a month.

What to do this week

Whether you use a tool or not, three actions will pay back inside a month:

1. Set up the GA4 channel grouping described above. Fifteen minutes. Free. Gives you visibility into AI referrals from the moment you save it.
2. Pick five prompts your customers would ask and check what ChatGPT, Gemini and Perplexity return for each. Write down whether your brand appears, what it says, and whether the link is clickable. This is your starting baseline for AI rank tracking, with or without a paid tool.
3. Check whether your site is AI-readable. Crawlable for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Has structured FAQ blocks. Schema on every key page. The seventy percent of the work that wins in both Google and AI search is the same SEO foundations that have always worked. The thirty percent that is genuinely new is making sure the AI crawlers can actually read you.

If you do those three things and watch the data for thirty days, you will know more about your AI search performance than the average enterprise running a USD$699-a-month tool.

How Cluo handles this

Cluo Portal tracks both sides in one screen, for AU$50 per month.

The AI rankings side runs the prompts you care about across ChatGPT, Gemini, Perplexity, Copilot and Claude every day, scores how prominently you appear, and flags which competitors share the page.

The AI traffic side connects to your GA4 property and surfaces the actual referrals from the eleven AI domains we track today. As new engines launch and start sending referral traffic, we add them. The AI Search channel sits alongside Google Organic and Paid Search in the dashboard, so you can compare AI to your other channels in the same view.

There is no contract. You can cancel from inside the dashboard.

FAQ