AI Search Referral Traffic Tracking Tools: ChatGPT and Perplexity
Referral tracking is the verify step of an AI search program. Which tools count ChatGPT and Perplexity clicks, tie them to conversions, and explain the visits that never happen.
On this site we judge every tool by how much of one loop it closes: monitor what the engines say, diagnose why they say it, act on the page, then verify the change paid off. Referral traffic tracking lives almost entirely in that last step. It is the receipt. A tool that only counts ChatGPT and Perplexity sessions hands you a receipt with no idea what you bought, which is why the useful products here connect each click back to the answer that produced it.
What an AI click looks like when it lands
ChatGPT is the easy case. OpenAI's Publishers and Developers FAQ says clicks out of ChatGPT search carry utm_source=chatgpt.com. Filter on that parameter and you have a clean ChatGPT segment in any analytics product that reads UTMs.
Perplexity arrives differently. A click from a Perplexity answer shows up as a referral from perplexity.ai, so you catch it by referrer rather than by campaign tag. Keep those two rules apart. If you build one "AI" channel that blends a UTM rule and a referrer rule without labels, you lose the ability to say which engine moved when the total jumps, and that is the first question anyone asks.
Then there is everything that arrives untagged. A click that loses both its referrer and its UTM on the way in lands in direct traffic, and nothing can reassign it afterward with certainty. Be wary of any vendor that promises to recover a specific share of "dark" AI traffic. We have not seen a sourced figure for that, so we will not print one.
Why the referral count undercounts visibility
An AI answer can name your brand, recommend you over two competitors, and cite your pricing page without the reader clicking anything. That answer did real work for you. Your analytics records nothing.
So referral data has a ceiling. It measures the slice of AI visibility that turned into a visit, not the visibility itself. ChatGPT has 820M+ weekly active users and Perplexity 22M+ monthly, and plenty of those sessions end inside the chat window. If sessions are your only AI metric, a week where ChatGPT mentions you constantly but nobody clicks looks exactly like a week where you disappeared.
The practical answer is to report two lines next to each other. Referral sessions and conversions by engine on one. Prompt-level mentions and citations on the other. When sessions fall and mentions hold, suspect the landing page or the way the answer frames you. When both fall, you lost the answer, and the work moves upstream.
Scoring the tools on the loop
| Tool | Records AI referral visits | Ties visits to conversions | Counts mentions with no click | Crawl data for diagnosis |
|---|---|---|---|---|
| Promptwatch | Yes, script or GTM template | Yes | Yes, prompt tracking across its engine list | Agent Analytics on Professional, Business, and agency plans |
| AthenaHQ | Through GA4, Shopify, and Webflow attribution | Revenue attribution listed | Yes | Not published |
| Wildcard | Yes, AI referral sessions | Add-to-carts, checkouts, revenue | Yes, shopping surfaces | Not published |
| Attrifast | Yes | Stripe and Shopify revenue | 10 to 30 prompts | AI crawler tracking listed |
| LLM Pulse | AI traffic analytics listed | Not documented | Yes, 5 engines | Not published |
| LLMClicks | LLM traffic tracker listed | Not documented | Yes | Not published |
| Similarweb | AI referral traffic measurement | Not documented | No | No |
"Not documented" means our directory entry does not cover it. The feature might exist. We just have not sourced it.
The tools, ranked by how much of the loop they close
Promptwatch
Visitor analytics runs on a lightweight script or a Google Tag Manager template and attributes AI-referred visits to conversions. Essential ($95/mo) includes 200K visitor events, Professional ($245/mo) 1M, and Business ($579/mo) 10M.
The counter is not why it ranks first. The same workspace holds prompt tracking for the mentions that never click, citation analytics that show which of your URLs an engine used, and, from Professional up, Agent Analytics crawler logs with a crawl-to-citation path. When ChatGPT referrals to one URL climb, you can walk the chain backwards: which prompts changed in prompt trends, which page entered the citations, when ChatGPTBot last fetched it. That is monitor, diagnose, and verify in one place, and Content Agents cover the act step if you publish on Webflow or Framer.
AthenaHQ
AthenaHQ lists GA4, Shopify, and Webflow revenue attribution next to visibility tracking across ChatGPT, Perplexity, Claude, Gemini, DeepSeek, and AI Overviews. Its Action Center turns findings into assignable tasks, which is a real act step. Starter is $295/mo ($245 annual) with credit-based usage. It does not publish its own visitor script or crawler logs, so verification depends on your GA4 configuration being correct.
Wildcard
The ecommerce pick. Revenue attribution covers AI referral sessions, add-to-carts, checkouts, and revenue by platform, with first-click and last-click views, and actions can publish to Shopify. Basic is $99/mo on a weekly refresh. Strong for a Shopify catalog, not built for a B2B lead funnel.
Attrifast, LLM Pulse, LLMClicks, Similarweb
Attrifast is the cheapest way to join AI referrals to money. It detects traffic from ChatGPT, Perplexity, Claude, Gemini, and Copilot and connects it to Stripe and Shopify revenue. Starter costs $9.99/mo for one site and 50k events. Prompt coverage is thin (10 prompts per 30 days on Starter), so treat it as a verify tool that needs a partner for monitoring.
LLM Pulse lists AI traffic analytics inside a five-engine tracker from €49/mo annual. LLMClicks lists an LLM traffic tracker next to its accuracy audits, with Starter at $159. Similarweb stands apart: it measures AI referral traffic and benchmarks competitors from $125/mo annual. Use it to watch the market, not to attribute your own sign-ups.
A setup that closes the loop
- Create two channels in your analytics: one on
utm_source=chatgpt.com, one on the perplexity.ai referrer. Leave them separate. - Install a visitor script that ties AI referrals to the conversion events you already report.
- Load your commercial prompts into a tracker so mentions without clicks get counted too.
- Every week, put sessions and conversions by engine next to mentions and citations by prompt.
- When a page earns citations but no clicks, act on it. Rewrite the section the answer quotes, or fix the claim the engine gets wrong.
- Before you judge the result, confirm in crawler logs that the bot actually refetched the new version.
Step six is the one most teams skip. Without it, a flat referral line after a rewrite tells you nothing, because you cannot tell "the change didn't work" from "the engine hasn't seen it yet."
We covered the analytics side of this in more depth in our piece on real-time AI referral analytics. The short version holds here too: count the click, but never let the click be the only number.
Our recommendation
If you need one tool for this job, evaluate Promptwatch first. Its visitor analytics answers "what did ChatGPT and Perplexity send us, and did it convert," while prompt tracking and citation analytics answer the question referral data cannot: what the engines said when nobody clicked. On Professional, the crawler logs add the proof that your fix reached the bot. Start on Essential if you only need the visitor layer and 50 prompts, then move up when you want the crawl data. Details and the trial are on promptwatch.com, and the rest of the field is in our ranked directory.