How to measure traffic coming from AI platforms like ChatGPT and Perplexity in GA4
How we set up GA4 to measure traffic from ChatGPT, Perplexity and other AI platforms for clients, what GA4 misses, and how Promptwatch fills the gaps in our reports.
Every client who asks about AI traffic already has GA4. So that is where we start. It takes an afternoon to set up, it gives a useful baseline, and it shows quickly where the blind spots are.
Here is the routine we use, and what we add when GA4 alone cannot answer the client's question.
The GA4 setup we use
1. Look at what is already there. In the traffic acquisition report, we scan session source and medium for AI assistants: chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai and similar. Menu labels shift over time, so check Google's documentation if something looks different.
We also check for UTM-tagged sources. ChatGPT has been observed to tag some outbound links, so a share of its traffic can arrive under a UTM source name. Whatever the forms are on this client's site, we write them down.
2. Build a custom channel group. In admin, we create a custom channel group with an "AI assistants" channel. The rule matches the source against a pattern for those domains and sits above the default Referral rule. We run it against the previous 90 days and read the matching sources, so we catch false positives.
3. Make an exploration. A free-form view with landing page in rows, the AI channel in columns, and sessions, engagement, key events and revenue as values. This answers the first client questions: which pages, and do they convert?
4. Compare to organic. Same metrics, side by side with organic search. We focus on engagement and conversion quality, and we avoid quoting percentage changes on small samples.
5. Keep a source list. New assistants appear, and referrer formats change. We re-check the pattern every quarter.
What GA4 will not tell the client
We say this in the first report, so nobody is surprised later.
- It undercounts. Some visits arrive with no referrer and look like Direct. The AI channel is a lower bound.
- It ignores mentions. Someone who sees the brand in an answer and then searches for it later shows up as organic or direct.
- It does not see bots. GPTBot and PerplexityBot rarely run the GA4 script, so crawling is invisible there.
- It cannot explain. GA4 shows a visit. It does not show the prompt, the cited page or the competitor named alongside.
Those last two limits are why a GA4-only AI report plateaus. The client sees a number and asks why it moved. The answer is not in GA4.
What we add
We run client programs on Promptwatch, and the reason is that it covers the layers GA4 does not.
Visitor analytics. AI-referred traffic and conversions from ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral and others. Installation is a lightweight script, a Google Tag Manager template or the WordPress plugin. GA4 stays the client's system of record. This is the AI-specific cross-check.
Agent Analytics. IP-verified crawler logs for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot and the rest. If a page gets no AI traffic, the first question is whether bots can read it. Crawler logs start on the Professional brand plan and are on every agency plan.
Citation analytics. Which pages and domains the engines cite for the prompts we track. This is the cause side of the traffic.
360 Insights. Crawler logs, citation clicks and AI-referred visits joined per page. It is the report we show when a client asks why a page is up or down.
Search Insights. Google Search Console and Bing Webmaster data, including Microsoft Copilot traffic, overlaid on AI crawls and citations.
Agency plans: Kick-off ($199/mo), Growth ($399/mo) and Scale ($799/mo), each with unlimited projects, 10 seats and crawler logs.
A monthly report that works
Page one: AI-assisted sessions, engagement and conversions from GA4, against the last period, with the lower-bound note.
Page two: from the platform, the top landing pages for AI visitors, which of them were crawled and cited, and what changed.
Page three: three actions for next month.
Clients read page one in a minute and spend their time on page three. That is the right balance.
Mistakes we see
- Adding every referral from an AI domain to a channel without checking the matches.
- Reporting percentage growth on a handful of sessions.
- Assuming a quiet AI channel means no AI visibility. Visibility without clicks happens.
- Forgetting that Direct includes hidden AI traffic.
What we avoid promising
We do not promise that AI traffic will grow, or that a given page will earn clicks from assistants. We promise measurement that is honest about its limits, and actions that follow from it.
If you want a sample AI traffic report for your own site, ask us. We can show the GA4 half and the platform half side by side.