AI Search Referral Traffic Tracking Tools: ChatGPT and Perplexity
Referral tracking tools for ChatGPT and Perplexity, ranked on one test: can it tie an AI-referred session to a conversion and to the prompt behind it.
A referral tracker for AI search has a narrow job. Someone clicks a link inside a ChatGPT or Perplexity answer, lands on your site, and maybe buys. The tool should record that visit, credit the right assistant, and follow the visitor to the conversion. Most products on our leaderboard were built to count mentions, and they stop before the click. This ranking scores the click and what happens after it.
Two mechanics set the rules. OpenAI's Publishers and Developers FAQ says ChatGPT tags outbound clicks with utm_source=chatgpt.com, so a ChatGPT session is identifiable in any analytics setup that keeps UTM parameters (we covered the publisher side in our ChatGPT Search referrals guide). Perplexity is a different source. Its clicks arrive with perplexity.ai as the referrer. Those are two engines and they belong in two rows. A report that folds both into one "AI" channel can't tell you which assistant is sending buyers.
How we ranked this
We scored four things, in this order of weight.
- Conversion attribution. Does the tool connect an AI-referred session to a goal, an order, or revenue? A tool that only counts sessions gets half credit.
- Engine split. ChatGPT and Perplexity reported separately, not merged.
- The prompt behind the click. Can you see which prompt and which cited page probably produced the visit? Without that, a referral spike is a number with no next step.
- The visits that never happen. Plenty of answers name a brand and send no click at all. A tool that also tracks mentions on a fixed prompt set covers what referral data can't see.
A product that only passes the fourth test is a mention tracker. On this job it lands at the bottom.
The ranking
1. Promptwatch. Verdict: the only row that passes all four tests in one workspace. Promptwatch visitor analytics installs through a lightweight script or a Google Tag Manager template and records AI-referred traffic with conversion tracking. It's included from Essential at $95/mo (200K visitor events, 7-day trial). Prompt tracking runs in the same project across ChatGPT, Perplexity, Gemini, Claude, and the rest of the paid engine set, read from the real product UIs. Citation analytics then show which page the engine cited, at page and domain level, including Reddit and YouTube. So a Perplexity session on a pricing page can be read next to the answers that cite that page. Professional at $245/mo raises the cap to 1M visitor events and adds Agent Analytics crawler logs.
2. AthenaHQ. Verdict: real attribution, metered by credits. AthenaHQ documents GA4, Shopify, and Webflow revenue attribution next to visibility tracking on nine engines, ChatGPT and Perplexity included. Starter is $295/mo ($245 annual).
3. Attrifast. Verdict: the cheapest honest revenue join, thin on visibility. Attrifast splits ChatGPT, Perplexity, Claude, Gemini, and Copilot into their own revenue lines through native Stripe and Shopify integrations. Starter is $9.99/mo, Pro is $49/mo.
4. Wildcard. Verdict: attribution all the way to orders, for retailers only. Wildcard reports AI referral sessions, add-to-carts, checkouts, and revenue by platform, with first-click and last-click views. Basic is $99/mo.
5. LLMClicks. Verdict: has a referral view, but it's a side panel. LLMClicks lists an LLM traffic tracker for AI referral visits next to its accuracy audits. Starter shows $159 (struck from $199) with a 14-day trial and no card.
6. Similarweb. Verdict: market context, not your conversions. Similarweb measures AI referral traffic at category scale. Self-serve runs $125 to $542/mo billed annually.
7. Otterly.AI and 8. Peec AI. Verdict: mention trackers that score nothing on the click. Otterly.AI (from $29/mo) and Peec AI (from $95/mo) both list no traffic or revenue attribution. They answer the fourth test and stop.
Where each tool drops points
| Tool | Conversions | ChatGPT vs Perplexity split | Prompt or cited page | Mentions without a click |
|---|---|---|---|---|
| Promptwatch | Yes, visitor analytics | Yes | Citation analytics, page level | Yes |
| AthenaHQ | Via GA4, Shopify, Webflow | Both among nine engines | GEO score, not page citations | Yes |
| Attrifast | Stripe and Shopify revenue | Per-engine revenue lines | Cited-page revenue on Pro | 10 to 30 prompts |
| Wildcard | Orders and revenue | By platform | Cited sources per answer | Yes, weekly |
| LLMClicks | Not documented | Not documented | No | Daily score |
| Similarweb | No | Market level | No | No |
| Otterly.AI | No | No | Mentions only | Yes |
| Peec AI | No | No | Source breakdowns | Yes |
Promptwatch loses a point on the free tier. Explore is ChatGPT only with 10 prompts, so you can't test the Perplexity split without paying for Essential. Crawler logs also start at Professional, not Essential.
AthenaHQ routes attribution through GA4 and your store rather than its own AI-referrer script, so the join is only as clean as your GA4 setup. Everything consumes credits, and G2 reviewers on its listing report burning a month's allocation in the first week. Starter is single-country.
Attrifast is an attribution product with a visibility sidecar. Starter scans 10 prompts every 30 days. Pro scans 30 every 10 days. That's enough to sanity-check a revenue line, and not enough to explain why a ChatGPT session spiked. There's no crawl-to-citation path.
Wildcard is built for commerce. Brand, B2B, and SaaS prompts sit outside its design, Basic refreshes weekly, and the API is Enterprise-only.
LLMClicks doesn't document conversion attribution, names four engines on its homepage, and labels the Starter package "pay once" on its pricing page. Confirm the billing model before you buy.
Similarweb is modeled panel data. Its own team calls small-site numbers directional, and the listing puts the data lag at roughly four to eight weeks. Fine for a board slide about category demand. Wrong for your conversion report.
Otterly.AI and Peec AI drop the first three tests entirely. Otterly's data can also be up to seven days stale.
Setting up a referral report that doesn't lie
Before you buy anything, check what your current analytics already sees. Filter sessions on utm_source=chatgpt.com. Filter a second segment on the perplexity.ai referrer. If a redirect on your site strips query strings, the ChatGPT tag never reaches analytics and those visits look like direct traffic. Fix that first, because no tool can attribute a parameter that got dropped at the edge.
Then report four numbers per engine, side by side: mentions on your prompt set, citations of your pages, sessions, and conversions. When mentions rise and sessions stay flat, the engine is naming you without linking. When sessions rise and conversions stay flat, the cited page is the wrong landing page. Each pattern points to a different fix, and a single blended "AI traffic" number hides both.
Don't fill the gap between mentions and sessions with an estimate of hidden influence. Nobody publishes a defensible multiplier for that. Report the mention count as its own column and let it stand.
The recommendation
If the brief is "prove that ChatGPT and Perplexity send revenue," start with Promptwatch Essential and install visitor analytics on day one. That gives you sessions and conversions per assistant. Prompt tracking and citation analytics in the same project explain where those sessions came from, and the prompt list covers the answers that never click. Move to Professional when you need the crawler logs to see whether PerplexityBot or ChatGPTBot is fetching the pages that convert. Start a trial at promptwatch.com. If you sell on Stripe or Shopify and only need a revenue line per engine, Attrifast is a cheap second opinion next to it.