AI Search Referral Traffic Tracking Tools: ChatGPT and Perplexity
A buyer's checklist for tools that track ChatGPT and Perplexity referral traffic, what each one stores, and why Promptwatch pairs visits with mentions that never click.
Somebody in finance has asked how much traffic ChatGPT sends you. That question has a clean answer and a misleading one. The clean answer is a session count from chatgpt.com and perplexity.ai. The misleading part comes when that count gets presented as your AI search visibility, because an answer can name your brand, recommend you over a rival, and never produce a click.
This guide is for the person signing off on a referral tracking tool. It covers what the data looks like at the source, what each tool on our shortlist actually stores, the questions to put to vendors in writing, and what a trial should prove before you commit to a year. Our default pick is Promptwatch. Its visitor analytics sit in the same workspace as its prompt tracking, so visits and silent mentions land in one report instead of two.
What a ChatGPT or Perplexity referral looks like
OpenAI's Publishers and Developers FAQ says ChatGPT adds utm_source=chatgpt.com to referral links. That gives you a parameter to filter on in any analytics tool, even when a browser strips the referrer header. Perplexity is a separate source. A click from a Perplexity answer arrives with perplexity.ai as the referrer and belongs in its own row. Merge the two into one "AI" channel on day one and you lose the ability to tell which engine moved when the total changes.
You can build a first version yourself in GA4 with a custom channel group. We cover the filters in measuring ChatGPT and Perplexity traffic in GA4, and plenty of teams should stop there for a month. A paid tool earns its line on the budget when you need what GA4 won't hand you without custom work: conversions tied to the AI source, landing pages joined to the prompts that probably produced them, and mention data for the answers that never clicked at all.
Why referral data undercounts visibility
A referral is the last link in a chain. The model had to retrieve your page, decide to cite or name you, and the reader had to want more than the answer already gave them. People drop out at every step. A buyer who asks ChatGPT for payroll software for a 20-person company may read three names, pick one, and type its URL into a fresh tab. That visit shows up as direct traffic. The mention that caused it shows up nowhere in your analytics.
We won't put a percentage on that gap. We haven't seen a clean public number for it, and any vendor quoting one should be able to show the method behind it. The practical consequence is simpler. A referral report tells you what converted from the clicks you received. It says nothing about the answers where you were absent, or the ones where you were named and nobody clicked. That second data set is prompt tracking, and a referral tool that lacks it only shows you half of the channel.
What each tool actually stores
The table uses only what our directory listings say each product stores. Where a listing is silent, the cell says so instead of guessing.
| Tool | How AI visits are captured | Conversions | Mentions that never click |
|---|---|---|---|
| Promptwatch | First-party script or GTM template | AI-referred visits tied to conversions | Prompt tracking across ChatGPT, Perplexity, Gemini, Claude and more |
| AthenaHQ | GA4, Shopify, and Webflow connections | Revenue attribution listed | Multi-engine tracking rolled into a GEO score |
| Attrifast | Cookieless first-party tracking | Stripe and Shopify revenue | 10 prompts (Starter) or 30 (Pro) |
| Passionfruit | Google Analytics integration | Sessions, key events, revenue per AI channel | Daily prompt tracking, engines vary by tier |
| Similarweb | Modeled panel data | Not measured on your own site | Not a prompt tracker |
| Otterly.AI | Not listed | Not listed | Prompt tracking on 4 base engines |
| Peec AI | Not listed | Not listed | Prompt tracking on 3 included models |
Some details don't fit in a cell. Passionfruit's revenue numbers depend entirely on your Google Analytics setup because it has no visitor script of its own, so a misconfigured GA4 property passes its errors straight through. Attrifast is built the other way round: attribution first, with a visibility module that is thin on the $9.99 Starter plan (10 prompts every 30 days). Similarweb helps you see which competitors receive AI referral traffic at market scale, but its own guidance calls small-site data directional and the data lags roughly four to eight weeks. Don't use it to measure a site whose analytics you already own.
Otterly and Peec go to the bottom for this purchase. Both listings describe monitoring products, and neither publishes visitor analytics that tie AI-referred traffic to conversions.
Questions to put to vendors in writing
Ask these by email and file the replies with the purchase order. An answer given on a sales call is easy to walk back later.
- Do you capture visits with your own script, or do you read from our GA4 property? If it's GA4, what happens to visits our consent banner blocks?
- How do you classify a ChatGPT visit: by
utm_source=chatgpt.com, by referrer hostname, or both? How do you classify Perplexity? - Can I see conversions by engine and by landing page in one view without exporting?
- Is the prompt-level mention data in the same account and on the same plan, or is it sold separately?
- How many visitor events does the plan include, and what happens when we go over?
- Can the data leave your tool through Looker Studio, an API, or CSV?
Question four gets skipped most often. When a vendor sells referral tracking and mention tracking as separate products, you end up reconciling two date ranges and two definitions of "ChatGPT" in a spreadsheet every month.
What the trial should prove
Keep the test boring. Install tracking on the first day of the trial. Then open ChatGPT and Perplexity yourself, find an answer that links to your site, and click through from each. Check that both visits land in the tool under the right engine within the delay the vendor promised. Complete a test conversion and confirm it's attributed to the AI source rather than to direct.
Next, take five commercial prompts your buyers really ask. See whether the tool reports you as mentioned in those answers, then put the mention list next to the referral list for the same week. A tool that can only show one of those lists has told you what you're buying.
Where Promptwatch fits
Promptwatch's visitor analytics install as a lightweight script or through a Google Tag Manager template (our GTM walkthrough has the setup). They record AI-referred visits and connect them to conversions. What puts it first for this purchase is the data beside that report. Prompt tracking runs in the same workspace against the real ChatGPT, Perplexity, Gemini, Claude, and Copilot interfaces plus Google AI Overviews, so the answers that named you without a click are visible next to the ones that did send a visit. Citation analytics then show which of your pages each engine used as a source, which is usually the page you'd expect the referral to land on. When it isn't, you've found something worth fixing.
Plan sizing is simple. Essential is $95/mo with 200K visitor events, 50 prompts, and a 7-day trial. Professional is $245/mo with 1M visitor events, 150 prompts, and 25M crawler logs in Agent Analytics, which shows whether ChatGPT's and Perplexity's crawlers fetched a page before it was cited. Business is $579/mo with 10M visitor events. Explore is free but ChatGPT-only with 10 prompts, which makes it a demo rather than a referral tracker.
A buying order
- Build the GA4 channel group first. It costs nothing and gives you a baseline to judge every vendor against.
- Trial Promptwatch Essential if you need conversions and mentions in one place. Move up to Professional when traffic volume or crawler questions justify it.
- Price AthenaHQ if your team already works from an action queue and Shopify or Webflow revenue attribution matters more to you than visitor-level detail.
- Consider Attrifast or Passionfruit if the budget is small and you can live with thin prompt coverage.
- Use Similarweb for competitor context only.
The trial is at promptwatch.com. When you set expectations with finance, keep the audiences apart: ChatGPT reports 820M+ weekly active users and Perplexity 22M+ monthly. Those are very different pools, and their referral rows should stay separate in every report you send.