Agentic SEO Tools
All posts
By Agentic SEO Tools TeamGA4ChatGPT trafficAI search ROIAI referral attribution

GA4 shows traffic from chatgpt.com, but which pages or prompts drive it? How to prove AI search ROI

GA4 shows chatgpt.com visits but not the prompts behind them. Here is how to find the pages, count the traffic properly and prove AI search ROI with real tools.

You open GA4, see chatgpt.com in the source list, and feel a brief surge of hope. Then the questions start. Which pages is it landing on? What did people ask before they clicked? Is any of it converting? And how do you put a number in front of a finance lead when half of it may be hiding in "Direct"? This is a measurement gap, not a mystery, and most of it can be closed.

What GA4 can and cannot tell you

It can tell you the landing pages. In Reports, Acquisition, Traffic acquisition, filter the session source for chatgpt, then add Landing page as a secondary dimension. That gives you which pages ChatGPT sends people to.

It cannot tell you the prompt. The visit arrives with a referrer of chatgpt.com, and the browser sends the origin only, not the conversation. Nothing in GA4 contains what the person asked. Anyone promising prompt-level attribution from GA4 alone is guessing.

It undercounts. ChatGPT search links carry utm_source=chatgpt.com, according to OpenAI's publisher guidance as quoted by analytics practitioners, but no utm_medium. Under GA4's default grouping that can land visits in Unassigned or show a medium of (not set) rather than Referral. And some AI visits, such as from mobile apps or without a referrer, end up in Direct. Treat the GA4 number as a floor.

It will not show Google's AI features separately. Clicks from AI Overviews and AI Mode are counted in Search Console under the Web search type, per Google's documentation, so they sit inside organic search.

Quick GA4 fixes

  1. Create a custom channel group with a rule on session source matching a regex such as chatgpt.com|chat.openai.com|perplexity.ai|claude.ai|gemini.google.com, and place it above Referral so it catches the traffic first.
  2. Check whether your property already groups recognized AI referrers into an AI Assistant channel, and compare its numbers with yours.
  3. Make sure redirects keep the query string, so the UTM survives.
  4. Store the first-touch source in a hidden form field or your CRM, so a lead is tagged even if the visitor returns later.

How to get from "visits" to "prompts"

You cannot read the prompt from the click. You can, however, work backward from the page. The method:

  • Find which prompts cite each of your pages in AI answers.
  • Count AI-referred visits and conversions per page.
  • Join the two. A page that is cited for five buying prompts and receives most of your AI visits is a good bet for the prompts that drive it.

That is inference, not proof, and you should present it as such. But it is far better than nothing, and it points you at the pages worth investing in.

Tools that close the gap

Promptwatch

Promptwatch is built for the join. Visitor analytics attributes AI-referred traffic and conversions across ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral and more, using a lightweight script, a GTM template or the WordPress plugin. Citation analytics shows which pages are cited for which tracked prompts. 360 Insights joins crawler logs, citation clicks and AI-referred visits per page, so one row tells you whether a page is crawled, cited and visited. Agent Analytics logs IP-verified AI crawler visits. Search Insights overlays Google Search Console and Bing data. Dashboards and PDF reports package the story for stakeholders.

Two published customer case studies show what the measurement can support, and should be read as those customers' results only. Crisp reported that AI-referred traffic converted at twice the rate of traditional channels. TreeHouse, the company behind Pararius and Huurwoningen.nl, reported AI search growing to about 25% of organic conversions, 99% of it from ChatGPT, after fixing crawler access and automating weekly reporting.

Pricing: free Explore (10 prompts, ChatGPT only), Essential $95 a month (includes 200K visitor events), Professional $245 (1M visitor events, crawler logs), Business $579 (10M visitor events). Agency plans run $199 to $799.

Alternatives

AthenaHQ lists GA4, Shopify and Webflow revenue attribution from $295 a month. LLM Pulse lists AI traffic analytics from €49 a month, with weekly cadence by default. Ahrefs Brand Radar and Semrush AI Toolkit report AI visibility beside SEO data but do not publish AI-referred conversion analytics. Scrunch AI lists AI agent and bot traffic reporting rather than conversion attribution.

Building the ROI case

  1. Define the conversion. Demo request, signup, purchase. Mark it in GA4 and in your AI analytics.
  2. Segment AI traffic. Use your custom channel group and your AI analytics script together.
  3. Compare conversion rate to organic. Report your own measured ratio, not an industry average.
  4. Report the floor. Say the figure undercounts, and why.
  5. Show the trend by page. Cited pages, their visits and conversions, month over month.
  6. Add visibility. Share of voice rising on the prompts behind those pages.

What not to claim

Do not state a prompt-level cause you cannot see. Do not use industry averages for AI conversion rates unless you can name the source. Do not promise that more visibility will produce a conversion on a given timeline. Present measured traffic, measured conversions and measured citations, and let the pattern make the case.