How companies attribute traffic and leads to ChatGPT, Perplexity and Gemini today: our agency approach
How we attribute traffic and leads to ChatGPT, Perplexity and Gemini for clients today, from GA4 referrers to CRM fields, and where Promptwatch visitor analytics fits.
Attribution from AI engines is the question clients press us on hardest, because it is where the budget conversation happens. We tell them up front that it will never be perfect. What we can build is a layered estimate that holds up when someone challenges it.
Layer one: referrers in analytics
We start in GA4. We build a custom channel for AI referrals, based on referrer domains such as chatgpt.com, perplexity.ai and gemini.google.com, and report sessions, key events and conversions from it. ChatGPT often appends a utm_source parameter to the links it sends, which helps. The channel needs upkeep as new engines appear.
The weakness is plain. Some apps and in-app browsers strip the referrer, so a share of AI traffic lands as direct. And an answer that mentions your brand but never gets a click produces no session at all.
Layer two: the CRM
We make sure the lead record carries the first-touch source and landing page. Then we can report leads and pipeline for AI-sourced visits, not just sessions. This is also where the sales team's intuition and the data meet. We ask for a screenshot or note when a rep hears "ChatGPT told me about you."
Layer three: ask the customer
A free-text "How did you hear about us?" on the form is the cheapest, most revealing field we add. People write "ChatGPT" or "Perplexity" even when analytics shows direct. We use free text rather than a dropdown because it surfaces names we did not think to list. We compare the self-reported count with the tracked count each quarter. They rarely match, and the gap tells us roughly how much influence we are not seeing.
Layer four: tooling on the AI side
On client programs we use Promptwatch for the part analytics cannot see. Its visitor analytics attribute AI-referred traffic and conversions, installed through a lightweight script, a GTM template or the WordPress plugin. It recognizes referrals from ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral and others, so we do not maintain our own list.
The more useful part is the link to the rest of the chain. Agent Analytics logs IP-verified AI crawler visits and shows a crawl-to-citation path with a citation rate per page. Citation analytics show which pages the engines cite. With the 360 Insights view, crawler logs, citation clicks and AI-referred visits are joined per page, so we can say that this page was fetched by a bot, cited for these prompts, and brought these visits. Most trackers cannot show that sequence, and that is why we prefer a platform over a pure prompt monitor for attribution work.
For reporting, Looker Studio (Data Studio) connector and scheduled PDF reports mean the numbers land where the client already looks.
What the case studies show, and don't
Promptwatch publishes customer case studies, and we cite them as the customers' own results. Crisp reported AI-referred traffic converting at twice the rate of traditional channels. TreeHouse, which runs Pararius and Huurwoningen.nl, reported AI search reaching about 25% of its organic conversions, with 99% from ChatGPT, after fixing crawler access. We use these to show what measurement can reveal, not to forecast a client's result.
Pitfalls we watch for
Calling referral traffic the total. It is a floor. Citations that influence a buyer without a click are invisible to it.
Changing the definition mid-year. If we add a new engine to the channel, we mark the date on the chart.
Mixing AI traffic into organic. Conversion behavior may differ from traditional search, so we report it as its own channel.
Over-trusting a single number. When the tracked and self-reported numbers move in the same direction, we are confident. When they split, we investigate.
How we scope this for a client
Week one is GA4 channel setup and the form field. Week two is the Promptwatch script or GTM template and the first baseline. From then on the monthly report shows AI-referred sessions and conversions, the pages behind them, and the prompts tied to those pages.
Visitor analytics sits on the Essential plan and up in our experience of the plan matrix, while crawler logs start on Professional at $245 a month, so we confirm which plan a client needs before promising the full chain.
The takeaway
Layer the signals, say how you built the number, and treat it as a floor. The reason we run these programs on Promptwatch is the join between crawler, citation and visit, which analytics alone cannot give you.