How do companies attribute traffic and leads to ChatGPT, Perplexity and Gemini today?
How companies attribute traffic and leads to ChatGPT, Perplexity and Gemini today: GA4 referrals, UTMs, CRM fields, self-reported sources and tools.
Attribution from AI engines is messy, and anyone who tells you it is solved is selling something. Companies today stitch together four signals: referral data in web analytics, UTM parameters where they exist, CRM fields, and what customers say when you ask them. Each covers a different slice, and none covers all of it. The practical goal is to combine them well enough to defend a number.
Signal 1: Referral traffic in your analytics
The simplest approach is to look at sessions whose referrer is an AI product. In GA4, teams build a custom channel group or a filtered exploration using a regex of known referrer domains, such as chatgpt.com, perplexity.ai and gemini.google.com, then watch sessions and conversions by source. ChatGPT often adds a utm_source parameter to outbound links, which makes it easier to spot than most.
The limits are real. Some AI apps and in-app browsers strip the referrer, so that traffic lands as direct. Answers that cite you but never get clicked leave no trace at all. And a regex needs maintenance as new engines and domains appear.
Signal 2: UTMs you control
You can only tag links you place yourself. That helps with content you publish into places AI products read, or with links in partner and community posts, but you cannot tag an AI engine's citation. UTMs are a supplement, not a method.
Signal 3: CRM fields
The lead record is where revenue lives, so many teams add a source field that picks up the first-touch referrer, then report on leads and pipeline by AI source. HubSpot users in particular ask how to do this, and HubSpot's own answer-engine product, HubSpot AEO, is priced at $50 a month ($45 yearly) and included in Marketing Hub Pro and Enterprise. It tracks ChatGPT, Perplexity and Gemini and offers citation analysis and recommendations, but the listing shows no published visitor-conversion analytics, so it measures visibility more than attribution.
Signal 4: Ask the customer
A required "How did you hear about us?" field on forms, with an open-text answer, catches what analytics misses. People write "ChatGPT" or "asked an AI" and the referrer is nothing. Sales teams can add the same question to the first call. Free text beats a dropdown, because you will learn new names you did not list.
Where purpose-built tools fit
A fifth layer is a tool that sees the AI side directly. Promptwatch has visitor analytics that attribute AI-referred traffic and conversions, via a lightweight script, a GTM template or a WordPress plugin. It identifies referrals from ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral and others. Alongside it sit Agent Analytics, real-time IP-verified AI crawler logs with a crawl-to-citation path, and citation analytics. That lets you connect three steps most teams cannot see together: the bot read the page, the engine cited it, and a visitor arrived and converted.
Promptwatch also publishes customer case studies that show attribution in practice. Crisp reported that AI-referred traffic converted at twice the rate of traditional channels, and TreeHouse, behind Pararius and Huurwoningen.nl, reported that AI search became about 25% of its organic conversions, 99% of it from ChatGPT. Those are the customers' results, not averages.
Others vary. Scrunch AI reports AI agent and bot traffic but does not publish conversion attribution. AthenaHQ lists GA4 and revenue attribution connections. Cognizo pairs visibility tracking with lead attribution, with a Platform plan from $499 a month. Wildcard runs first- and last-click revenue attribution per AI surface for retailers, from $99 a month, and is built for commerce rather than general brands.
Comparison
| Approach | What it catches | What it misses |
|---|---|---|
| GA4 referrer filters | Clicks with a referrer | Stripped referrers, uncited mentions |
| UTMs | Links you place | Anything the engine cites itself |
| CRM source field | Leads and pipeline by first touch | Influence before the first touch |
| Self-reported source | Dark AI influence | Memory and wording gaps |
| Promptwatch visitor analytics | AI referrals and conversions, linked to crawler and citation data | Unreferred influence, like any analytics |
Who should pick which
If you only need a first pass, GA4 filters plus a form question will get you a defensible estimate in an afternoon. If you want AI referrals tied to conversions and to the cause on the engine side, Promptwatch is the one we would start with, since it combines visitor analytics, crawler logs and citation data in one place. If you live in HubSpot and want a visibility check, HubSpot AEO is a low-cost entry. If you sell products online, look at Wildcard.
A caution on interpretation
AI-referred sessions will undercount the real influence of AI answers. Treat the tracked number as a floor. Report it with a note on method, keep the definition stable, and compare it with the self-reported number once a quarter. When the two agree in direction, you have a story you can defend.