If your GA4 report shows a healthy number of enquiries from "Referral" traffic and you have no idea where they came from, there's a decent chance some of them were sent by ChatGPT, Perplexity or Gemini. Your analytics isn't broken. It's just filing AI visitors in the wrong drawer, and that leads to some expensive decisions.
This is a quiet but real tracking mistake. Business owners look at the wrong numbers, conclude that something isn't working, and then cut the very thing that is. Here's what's going on and how to fix it in about twenty minutes.
What "AI referrer confusion" actually means
Every time someone clicks a link to your site, their browser usually tells your analytics where they came from. That label is called the referrer. GA4 reads it and sorts the visit into a channel: Organic Search, Direct, Referral, Social, Paid Search and so on.
The trouble is that AI assistants send referrers that GA4's default rules were never designed for. A visit from chatgpt.com can land in the generic "Referral" bucket, alongside a link from a local directory or a supplier's website. Some AI visits arrive with no referrer at all and fall into "Direct", where they look like someone who typed your web address from memory.
The result is that a new, fast-growing source of customers is blended into channels that mean something quite different. You can't judge what you can't see.
Why it's a conversion tracking problem, not just a reporting nuisance
It would be easy to treat this as a tidy-up job for the analytics geeks. It's more serious than that, because conversion decisions follow the data.
Say a plumber in Brighton gets twelve quote requests a month from "Referral" and fewer than half that from Organic Search. Without separating AI traffic, they might assume a mystery directory is their best partner and pay for a listing there. In reality, several of those enquiries may have come from someone asking an AI assistant for a recommended emergency plumber.
Put numbers on it. Suppose that directory listing costs £40 a month, so £480 a year. If five of those twelve enquiries were actually AI-driven and only seven came from the directory and other genuine referrals, the listing is doing far less than it appears. Meanwhile the service page the AI keeps citing might be the real earner, and it's the page nobody has touched in two years because "Referral" never pointed to it.
Here's how the mix-up usually hurts:
- You misjudge your SEO. Visitors who found you through AI answers aren't counted as search, so your SEO looks weaker than it is.
- You misjudge your content. The pages AI tools cite are often your clearest explainer or service pages. If you can't see that, you may rewrite or remove them.
- You misjudge your conversion rate. AI visitors often arrive well informed and ready to act. Mixed into a general bucket, their behaviour is averaged away and you miss that this audience converts differently.
- You waste ad budget. Credit for a sale goes to the wrong channel, so money moves to the wrong place.
The first fix: build an AI channel in GA4
The most practical solution is a custom channel group that pulls AI visits into their own labelled channel. Several analytics guides, including this AI referral tracking guide, recommend the same approach: use regex rules that match the known AI referrer domains. The current core set is chatgpt.com, claude.ai, perplexity.ai, gemini.google.com and copilot.microsoft.com.
The set-up is straightforward:
- In GA4, go to Admin and open Channel Groups under data display.
- Create a new channel group and give it a clear name such as "AI / LLM Traffic".
- Add a new channel called "AI Search" with a condition on the Source dimension, using a regex match.
- Enter a pattern such as chatgpt|perplexity|claude|anthropic|gemini|copilot.
- Save, then reorder the channels so your AI channel sits above the defaults.
That regex pattern is the kind of thing a free GA4 channel group approach suggests, and it works without paying for any extra tool. Treat the domain list as a living document, because these tools change their addresses and new ones appear.
Edge cases that catch people out
Order matters more than you'd think
GA4 checks channels from top to bottom and stops at the first match. If your AI channel sits below the default "Referral" rule, the visit gets claimed first and your new channel stays empty. This is the most common reason people build the channel, wait a week, and see a zero. Drag it to the top and check again.
Loose patterns catch the wrong visitors
A short pattern like "gemini" will also match gemini.com, a crypto exchange, and "copilot" can catch GitHub Copilot's own pages. For a local business that's unlikely to matter, but if you see odd numbers, tighten the rule to full domains, for example ^(chatgpt\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com)$, and add new ones as they appear.
The Direct bucket will never be fully clean
Some AI apps, particularly on mobile, strip the referrer, so those clicks still show as Direct. You can't recover every one, but you can spot the pattern. Look at your Direct traffic by landing page. If deep service pages or blog posts are getting plenty of "Direct" visits, nobody is typing those long addresses from memory. Many AI tools also add a utm_source tag to the links they share, so it's worth checking your Source report for those too.
Check what happens after the visit
Splitting out the channel only helps if conversions are being recorded. Make sure your enquiry form submissions, phone-number taps and booking clicks are set up as key events in GA4. Then compare the AI channel against Organic Search on three things: engagement rate, key events per session, and which landing pages bring people in.
Go in with realistic expectations. For most small UK sites, AI visits are still a small slice, perhaps a few dozen sessions a month. Don't read too much into two enquiries. Give it a full quarter before you change budgets, and treat it as a trend to watch rather than a verdict.
Your history may look different afterwards
Custom channel groups in GA4 generally apply to your existing data in reports, so you can look back across previous months once the channel is built. Reports that rely on the default grouping, and any figures you've already exported or shared, won't change, so note the date you made the change when you compare periods.
