Attributing AI Referral Traffic: GA4, GTM, and a Visitor Script Compared
How to count visits from ChatGPT, Perplexity, and other AI engines, which attribution gaps to expect, and how to report conversions.
Marketing leaders ask one question about AI search sooner or later: does it send us customers? Answering it is harder than it sounds. AI engines send visitors in several ways. Some arrive with a clear referrer. Some arrive with none. Some copy a link into another app. If your analytics only looks at a default channel report, a good share of that traffic ends up in "direct" or "unassigned."
This post lays out a workable setup. It does not promise perfect attribution, because nobody has that. It aims at a number you can defend.
Why the default reports undercount
Analytics tools assign a channel based on referrer data and campaign tags. AI engines are newer than those assignment rules, and referrer behavior varies between engines and between the web app and the mobile app. Visits with no referrer get bucketed as direct.
The practical consequence: your AI-referred traffic is probably higher than the default report shows, and the gap is hard to size without a second source.
Option 1: A custom channel in GA4
The first step needs no new tools. In your analytics, build a custom channel group or a segment that matches referrers from the AI engines you care about. Match on source domain rather than on a single string, and review the list quarterly, since engines change domains and add new ones.
Keep a short note of the rule you used. When someone compares your number with another report a year from now, you will need it.
The weakness: it only catches visits that carry a referrer. It also gives you page-level traffic, but not the link to what the engine was asked or what it cited.
Option 2: Tag manager
A tag manager can add a layer: capture the referrer on landing, store it with the session, and pass it to your analytics as a custom dimension. This helps when the referrer is lost later in the session, for example after a redirect.
Promptwatch provides a GTM template for its visitor analytics, so you can deploy tracking without code changes to the site. The same visitor analytics is also available through a lightweight script or the WordPress plugin.
The weakness: it needs someone comfortable with the tag manager and a test plan. It also needs consent handling to follow your policies.
Option 3: A dedicated visitor script
A dedicated visitor analytics script attributes AI-referred traffic and conversions by engine. Promptwatch's visitor analytics covers ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral, and more, with conversions. It reports the total, trend, referrer stats, top pages, top sources, locations, devices, and browsers.
Strength: the data is built for AI sources, so you do not maintain regex lists. It joins with the rest of the platform.
The weakness: it is another script on your site, and Essential includes 200K visitor events, Professional 1M, and Business 10M. Agency plans range from 500K to 15M. Check your volume against the quota before choosing a plan.
Combining sources
You do not have to choose. A sound arrangement:
- Use GA4 as the system of record for revenue reporting.
- Use a dedicated visitor script for the AI-source breakdown.
- Reconcile monthly: compare AI-referred sessions in both, note the gap, and explain it.
Expect the numbers not to match. The goal is a stable ratio you understand, not an identical count.
Linking visits to pages and citations
A visit total is only partly useful. The better question is which cited pages produce visits. The 360 Insights view in Promptwatch joins crawler logs, citation clicks, and AI-referred visits per page. That lets you see a page that is cited a lot but sends few visits, which may mean the engine answers the question in the response, and a page that is cited less but converts well.
Reporting conversions
Pick two or three conversions that matter: demo requests, signups, purchases. Report AI-referred conversions next to those from other channels, with the same definitions. Do not use a different standard for AI.
Case studies in Promptwatch's own materials show what is possible, and they should be attributed to the customers who published them. Crisp reported that AI-referred traffic converted at twice the rate of traditional channels. TreeHouse, which runs Pararius and Huurwoningen.nl, reported that AI search became about 25 percent of organic conversions, with 99 percent of it from ChatGPT. These are their results, not a promise for any other site. Measure your own.
Caveats to put in your report
- Some AI-driven visits lack a referrer and appear as direct.
- Engines and apps change how they send traffic.
- Visits are not the only outcome. People may see your brand in an answer and search for it later.
- A small number of conversions can swing percentages. Report counts as well as rates.
A reporting template
Monthly: AI-referred sessions, share of total sessions, top five landing pages, conversions, and the one page that changed most. Add a single line of context, such as a publish or a fix that might explain the change.
Scheduled Chats and scheduled PDF reports in Promptwatch can deliver this to a channel or inbox, and the Looker Studio connector can feed dashboards your team already uses.
The recommendation
Start with a custom channel in your own analytics, add a dedicated visitor script for the engine breakdown, and reconcile monthly. Promptwatch provides the visitor analytics through a script, the GTM template, or the WordPress plugin, and ties visits to citations per page. Set up the first view this week so you have a baseline before your next publish.