ChatGPT Just Sent You a Visitor — Here's How to Actually Measure It in 2026

Ask most UK small business owners where last month's website traffic came from and you'll get a familiar list: Google organic, a bit of Meta, maybe the newsletter. What rarely makes the list is the growing trickle of visitors arriving from ChatGPT, Perplexity, Copilot and Google's AI Overviews. Those visitors are real, often unusually well-qualified, and right now plenty of analytics accounts are quietly filing them under 'direct' or 'referral' and moving on.

That is a measurement gap worth closing in 2026 — not because AI referrals are huge (for most SMEs they're still a modest slice), but because they're growing quickly and because people who arrive that way tend to be further along in the buying journey than a casual browser.

Why AI traffic is so easy to miss

Three things conspire against you. First, referrer data gets stripped — plenty of AI apps and in-app browsers simply don't pass a referrer, so the session lands as direct traffic and dies in the 'unknown' bucket. Second, Google AI Overviews don't generate a distinct referrer at all: a click from an AI-written summary still arrives looking like plain old google / organic. Third, the volumes are small enough that nobody spots them in a monthly report until someone asks why 'direct' has crept up three months running.

Accept up front that you'll never get perfect attribution here. The goal is directionally right, consistently measured, and honest about its limitations.

Sort out your UTM convention first

For any link you control — a placement inside an AI tool, a sponsored mention, a tracked link in a partner's chatbot — keep it boring and consistent:

• Source: the platform, lowercased and hyphenated (chatgpt, perplexity, copilot, gemini)
• Medium: always ai-assistant, so you can group them later without guessing
• Campaign: the actual reason for the link (spring-launch, product-faq, partner-test)
• Content: the variant, if you're testing placements

Never put an email address, order number or anything personally identifying in a UTM. It ends up in your analytics, in your exports, and eventually in someone's spreadsheet — which is a UK GDPR headache you don't need.

For organic AI referrals you can't control, you'll lean on referrer data plus one low-tech trick that still outperforms most software: a simple 'How did you hear about us?' field at checkout or on the enquiry form. A surprising number of people will happily type 'I asked ChatGPT'.

Build an AI channel group in GA4

Head to Admin → Data display → Channel groups → Create new channel group. Add a channel called 'AI Assistants' with a session source condition matching a regex like:

chatgpt|openai|perplexity|copilot|claude|gemini|poe\.com|you\.com

Order is everything. Drag AI Assistants above Organic Search and Referral, or Google's default rules will claim those sessions before your custom channel gets a look in. Because GA4 evaluates channel groups at report time rather than at collection, your historical sessions get reclassified too — a rare free win.

AI Overviews remain the awkward one

Google's AI summaries sit inside the same results page as ordinary links, so there's no clean referrer to catch. You can only reason around it. Watch Search Console for a pattern of stable or rising impressions with a falling click-through rate on informational queries — that gap is content being answered in the results page rather than clicked. Pair it with a look at which landing pages are suddenly getting shorter sessions from organic, and with manual checks on your ten most commercially important queries to see whether an AI Overview appears.

Treat this as a hypothesis, not a fact. Impressions up, clicks down can also mean a ranking shuffle or a seasonal dip.

Now the only question that matters: does it convert?

Create a simple exploration comparing conversion rate, engagement rate and average session duration across three channel groups: AI Assistants, Organic Search and Paid Search. Use a 90-day window rather than 30 — AI referral volumes are still small enough that a fortnight of data tells you nothing except noise.

What we typically see for UK clients: fewer sessions, longer dwell time, more question-shaped behaviour, and conversion rates that either beat organic comfortably or sit at zero because the landing page doesn't answer the question that was asked. If it's the latter, the problem is your page, not the channel.

The UK compliance bits nobody enjoys

Analytics storage still needs consent under PECR, and any personal data you capture needs a lawful basis under UK GDPR. Referral parameters in a URL are fine; persisting them in a cookie or pushing them into a CRM profile is where you should be checking your consent banner actually does what the label claims.

Server-side tracking — sending events through a cloud container or the Measurement Protocol rather than the browser — is genuinely useful in 2026. It survives ad blockers and intelligent tracking prevention, and it gives you cleaner event data. What it does not do is remove the need for consent, and it does change where data flows. If your vendor processes events outside the UK, get the transfer paperwork in order before someone from legal asks.

A 30-minute setup for today

1. Write your UTM convention down and share it with anyone who builds links.
2. Create the AI Assistants channel group and put it top of the list.
3. Add a 'how did you hear about us?' field to your checkout or form.
4. Build one exploration comparing AI, organic and paid conversion rates.
5. Diarise a quarterly review — this channel is moving too fast for an annual check.

The bottom line

AI search won't rescue a weak offer, and for most UK small businesses it's still a supporting act rather than the headline. But it is now a real source of traffic that your reports should be able to see, name and judge — and that starts with ten minutes of naming discipline, not another subscription.