How can we tell which of our new leads came from ChatGPT?

Your board wants to know how many leads came from AI. You open Google Analytics and find a handful of visits from ChatGPT last month, and no enquiries you can tie to them. Meanwhile your sales director says new prospects keep mentioning that "ChatGPT suggested you". Both are telling the truth. They are measuring different things.

The reason is the way buyers use AI assistants. A finance director asks ChatGPT, Gemini or Copilot for a shortlist, reads your name, then searches for you on Google. Or types your address the next morning. Or forwards the answer to a colleague, who finds you a week later on LinkedIn. In each case your analytics records the last step, and the AI answer that started it is invisible.

This matters because the habit is now common. Forrester reported in January 2026 that 94% of business buyers use AI in their buying process. Google Analytics 4 (GA4) has caught up only partly: since 13 May 2026 it has a default AI Assistant channel, but it leaves out Perplexity and Claude, and it cannot separate clicks from Google's own AI Overviews and AI Mode. This guide explains what GA4 now shows, what it misses, and the form, call and CRM questions that fill the gap, so you can give your board an honest figure.

#Why does your analytics undercount AI?

Analytics credits the last place a visitor came from before landing on your site. AI assistants are often the first place, not the last. There are four common journeys, and only one shows up clearly.

#The buyer clicks a link in the answer

ChatGPT, Perplexity, Copilot, Gemini and Claude often show links to their sources. A click from most of them now lands in GA4's AI Assistant channel, and a click from Perplexity or Claude lands in Referral. A click from a Google AI Overview or AI Mode is different: Google's own documentation says it counts as Organic Search, so it looks exactly like an ordinary Google click. This is the journey you can partly see, and for most firms it is the smallest.

#The buyer searches for your name

Many answers name firms without a link, or the buyer prefers to check you on Google. They search for your firm's name and click your listing. GA4 records organic search, and the visit looks like someone who already knew you.

#The buyer types your address, or the source is lost

Some buyers note your name and visit later by typing your address. Others click a link in the ChatGPT phone app, which opens it in the phone's browser; in that hand-off the referrer often does not survive, according to a July 2026 analysis by Nice Looking Data. Both show up as Direct.

#The buyer comes back through another channel

A buyer reads your name in an AI answer, does nothing for a fortnight, then sees your founder's post on LinkedIn and clicks through. GA4 credits LinkedIn.

These visitors often arrive knowing who you are and get to the point quickly, which we explore in what your website should do now visitors arrive pre-sold. The same shift explains why many firms see traffic falling while their rankings hold.

#What does GA4 show about AI traffic now?

Less than the channel name suggests. Here is where each kind of AI visit usually ends up, based on Google's channel definitions and the Nice Looking Data analysis.

The third row catches many firms out. ChatGPT adds its own name to some of the links it gives, as a tag that reads chatgpt or chatgpt.com depending on where in ChatGPT the link came from. Because the tag often carries a source but no medium, GA4 stops reading the referrer, the visit matches none of its channel rules and it falls into Unassigned. To check, open Reports, then Acquisition, then Traffic acquisition, find the Unassigned row and add Session source / medium as a secondary dimension.

#How to set up AI attribution

The set-up takes a few hours and no new software. Most of the value comes from the first three steps, because asking people catches the journeys that analytics cannot.

#Step 1: Add the question to every enquiry form

Add "How did you first hear about us?" as a required field, with options such as search engine, AI assistant (for example ChatGPT, Gemini, Copilot, Claude or Perplexity), LinkedIn, recommendation from someone, press or article, event, and other. Put a short free-text box beside it. People often write something more useful than the option they tick.

#Step 2: Ask on every first call

Train whoever takes first calls to ask an open question first: "How did you come across us?" Then, if AI has not come up, a gentle follow-up: "Did you use ChatGPT or any other AI tool while you were looking at firms like ours?" The order matters. Asking about AI first plants the idea and inflates the answer.

#Step 3: Record it in your CRM as a required field

Create two fields: where the buyer first heard of you, and whether they used an AI assistant at any point, with the assistant named if they remember. Make both required before an enquiry can move to the next stage, because optional fields get left blank in busy weeks.

In HubSpot, for example, that means:

• A contact property called "How did you first hear about us?", as a dropdown with the options from Step 1, added to your forms as a required field.
• A second dropdown, "Used an AI assistant in research?", with the values yes, no and not asked, plus a text field for which assistant. Sales fills these in after the first call.
• Matching deal properties, filled in when the deal is created (by a workflow if your plan includes them), and set as required when a deal moves to its first stage, so every proposal and win carries its source.

Other CRMs work the same way. The point is that the answer travels with the deal, so you can later count proposals and wins, not just enquiries.

#Step 4: Set up GA4 to catch every AI source

Create a custom channel group in GA4 (Admin, then Data display, then Channel groups), copying the default group so everything else stays the same. Add a channel called "AI assistants" with the condition Session source matches regex, using:

.*(chatgpt|openai|perplexity|claude|gemini|copilot).*

Move it above Referral in the list, so it catches Perplexity and Claude visits and the tagged ChatGPT visits that would otherwise land in Unassigned. Nice Looking Data reports that custom channel groups also reclassify past sessions, so you can see history as well.

To link those visits to enquiries, build an exploration: Explore, then Free-form, with Session source / medium as rows and Sessions and Key events as values, filtered to the key event your enquiry form fires (for example generate_lead). That shows which assistants sent visits that went on to enquire.

#Step 5: Watch searches for your name, and Google's AI impressions

In Google Search Console, filter the Performance report for queries containing your firm's name. A steady rise with no campaign, event or press coverage to explain it suggests people are hearing about you somewhere, and AI answers are among the likeliest places.

Search Console also has generative AI performance reports, live for all sites since 31 August 2026. They show impressions, pages, countries and devices for AI Overviews and AI Mode, though not clicks or the searches behind them. Use them to see which pages Google's AI features show, alongside the Organic Search figures in GA4.

#Step 6: Bring the signals together monthly

Once a month, put the form answers, call notes, AI channel visits, branded search trend and Search Console AI impressions side by side, with what the assistants currently say about you. Our guide to tracking whether AI assistants recommend your firm explains that last check.

#How to read the signals together

Read them the way a detective would. Each has a blind spot, but when they point the same way you can be reasonably confident.

For example, say a firm receives 40 enquiries in a month. Five people pick "AI assistant" on the form. First calls reveal three more who used ChatGPT or Perplexity during their research but ticked "search engine", because that is how they finally arrived. GA4 shows a dozen visits in the AI assistants channel, two of which became enquiries already counted. And searches for the firm's name rose, with nothing else to explain it.

A fair summary is that at least 8 of the 40 enquiries involved an AI assistant, and the true number is probably higher, because some buyers will not remember or mention it. That figure rests on evidence and gives you a trend to follow. Then track how many AI-involved enquiries become proposals and clients compared with the rest, rather than assuming they are better or worse.

#What share of leads should we expect from AI search?

There is no reliable benchmark for B2B services, so judge your own trend rather than someone else's average. The share depends on how many of your buyers use assistants to shortlist and whether you are named when they do. In G2's April 2026 survey, 51% of software buyers said they start research in an AI chatbot, but that is one sector.

Once a firm is named, the share can be meaningful. For Helm Club, a founder membership community we work with, "1 in 5 new members this year came through an AI recommendation." It is one organisation in one market, so treat it as a sign of what is possible, not a benchmark. The detail is in the Helm Club case study.

#How to report it to your board

Report AI-sourced leads as a floor, explain the method and show the trend. Boards are comfortable with estimates when the method is clear, and lose trust when a precise-looking number turns out to be a guess.

Using the example above, a monthly line might read: "At least 8 of 40 enquiries this month involved an AI assistant, based on form answers and first calls. Visits in our AI assistants channel rose again, and searches for our name are up on last quarter." Put next to it what the assistants say for your buyers' questions, so the board sees cause and effect side by side. Over time, add how many AI-involved enquiries became proposals and clients. Our guides to building an AI visibility dashboard and board report and measuring the return from your website show how to fit it into a wider report.

#Mistakes that distort the numbers

• Counting only GA4's AI Assistant channel. It misses Perplexity, Claude, Unassigned ChatGPT visits and every buyer who searched for your name.
• Asking leading questions. "Did ChatGPT recommend us?" invites a polite yes. Ask openly first.
• Leaving fields optional. Partial data can look like a trend when it is really a change in who bothered to fill in the form.
• Crediting all branded search to AI. Press, events, referrals and your own sales activity drive searches for your name too. Only the unexplained rise is a clue.
• Building a complicated model too early. Weighted models that split credit across every touchpoint rest on guesses. Get the basic questions answered consistently for a few months first.
• Changing the method every month. A rough measure used the same way for a year tells you more than a perfect one changed every quarter.

#What to do this month

1. Add "How did you first hear about us?" with an AI assistant option to every enquiry form, as a required field.
2. Brief whoever takes first calls on the open question and the gentle follow-up.
3. Create the source and AI fields in your CRM and make them travel with each deal.
4. Set up the GA4 custom channel group and check last month's Unassigned sessions for ChatGPT.
5. Put a monthly half hour in the diary to bring the signals together and write one line for the board.

#Frequently asked questions

#Can GA4 show us leads from ChatGPT?

Partly. With the AI Assistant channel, or a custom channel group, it can show visits that came from ChatGPT and which of those visitors went on to enquire. It cannot show buyers who read an AI answer and then searched for you or typed your address, which is why asking buyers matters so much.

#Why does ChatGPT traffic show as Direct in GA4?

Usually because the referrer was lost. Links opened from the ChatGPT phone app often reach your site with no referrer, and buyers who copy your name or address from an answer arrive by typing it. GA4 then has nothing to go on and records Direct. Asking on the form and first call is the only way to recover these.

#Does GA4's AI Assistant channel include Perplexity and Claude?

No. Google's definition, checked in September 2026, lists ChatGPT, Gemini, DeepSeek, Copilot and Grok, and excludes Google's own AI Overviews and AI Mode. Perplexity and Claude visits appear as Referral unless you add them to a custom channel group, as in Step 4.

#How do I see Google AI Mode traffic in Search Console?

You can see impressions, not clicks. The generative AI performance reports show how often your pages appeared in AI Overviews and AI Mode, which pages, and by country and device. Clicks from those features are counted in GA4 as Organic Search and cannot be split out.

#Do people remember that ChatGPT recommended us?

Many do, especially if you ask soon after they get in touch and in conversation rather than on a form. Some will not, and some think of it as research rather than a recommendation. That is why the follow-up asks whether they used an AI tool while looking, and why the answers are a minimum.

#Is a rise in searches for our name proof that AI is working?

No, it is a clue. Searches for your name also rise after press coverage, events, referrals and sales campaigns. If you can rule those out, and your tracking shows assistants naming you more often over the same period, the link is reasonable. Present it as supporting evidence, not proof.

#What if almost nobody mentions AI?

Then you have learned something. Either buyers in your market are not yet using assistants to shortlist, or they are and you are not being named. Check what ChatGPT, Gemini, Copilot, Claude and Perplexity say when asked your buyers' questions. If competitors are named and you are not, a low number is a warning rather than a reassurance.

If you would like a second pair of eyes on how you track where enquiries come from, book a free Growth Review: 30 minutes with David Adams, and three fixes in writing afterwards.

#Sources