How can I tell which leads came from ChatGPT or other AI assistants? Our attribution setup
Our attribution setup for telling which leads came from ChatGPT or other AI assistants: analytics channels, a CRM source field, and visitor analytics tied to citations.
Clients usually ask this after a good month: "leads are up and we think it is the AI work, but how would we know?" Our honest answer is that you can get close, never exact, and that the method is three layers deep.
Layer one: reading the referrers
First we set up the analytics. In GA4 or the client's tool we create a channel group for AI assistants, built from referrers such as chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com. Then we watch where those sessions land and what they do.
This layer has a known flaw. Some AI visits come with no referrer and fall into direct traffic. And plenty of influenced buyers never click: they see the brand in an answer, then search for it by name later. That shows up as branded organic or direct. So referrer data alone undercounts, and we say so up front.
Layer two: asking the lead
The cheapest fix is a question. Every demo and contact form gets an open "How did you hear about us?" field with an AI assistant option and a text box. The answer flows into a CRM field, and we ask sales to repeat the question on the first call and log it.
We normalize the answer into a source value (AI assistant, organic search, referral and so on) and keep the free text as well. The free text is gold: it often says "I asked ChatGPT for tools that do X", which tells us the exact kind of prompt that is working.
In most accounts, self-reported AI leads exceed what the referrer data shows. We report both numbers and treat the true figure as somewhere above the larger of them.
Layer three: connecting visits to prompts and pages
Referrer and form data tell you that AI sent someone. They do not tell you which prompt, which cited page or whether the crawler even reached the page. For that we install Promptwatch visitor analytics on the client's site. It tracks AI-referred traffic and conversions from ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral and others, through a lightweight script, a Google Tag Manager template or the WordPress plugin.
The real value comes from joining it with the rest. Agent Analytics shows which AI crawlers fetched which pages, and the crawl-to-citation path shows which pages were cited. 360 Insights joins crawler logs, citation clicks and AI-referred visits per page. So we can say something concrete: this guide is cited often, its AI-referred visitors reach the pricing page, and some of them submit the form. That is a story a CFO can follow.
What the CRM report looks like
We build a simple view with three groups of leads: referrer says AI assistant, lead says AI assistant, and either. We compare each against organic search on lead quality and pipeline, with a note on sample size. If there are 12 leads in a group, we say the numbers are directional.
We do not state industry averages. Whether AI-sourced leads are better or worse than organic ones depends on the business, and only the client's own data can say. For context, Promptwatch's published case studies include Crisp, which reported that AI-referred traffic converted at twice the rate of traditional channels, and TreeHouse, which reported AI search became about 25 percent of its organic conversions after crawler fixes. Those are those customers' results.
Common mistakes we fix
Relying on last click only. AI-influenced buyers often return later through another channel, so assisted views matter.
Free-text only source fields. Without a normalized value, nobody can report on it.
Forgetting calls and chat. Many leads come by phone or live chat, and nobody asks where they came from.
Treating visibility as attribution. A rising Visibility Score says nothing about leads until visits and conversions are connected.
When we say we cannot tell
Sometimes the honest finding is that the volume is too small to say anything. We would rather report that than invent a conclusion. The setup still pays off, because the data will be ready when the volume grows.
If you want to try this yourself, Promptwatch has a free Explore plan for visibility on ChatGPT, and visitor analytics sits on paid plans starting at $95 a month.