AI Referral Traffic Analytics Tools and LLM Traffic Analytics
LLM traffic is sessions that arrived because an assistant sent someone. Here is how we measure AI referral traffic without treating mentions as visits.
LLM traffic analytics is a narrower job than "AI visibility." Visibility is whether a model named you. LLM traffic is whether a person landed on your site after that. Those are different numbers, and mixing them is how a mention report gets sold as pipeline.
We measure both. The tool that holds the join is Promptwatch: citation analytics for the mention, visitor analytics (script or GTM) for the session. GA4 remains the rest of the site.
This is not a 2026 product roundup and it is not a "real-time" pitch. It is how we classify referral traffic that came from a language model.
What counts as LLM traffic
A session whose referrer is an assistant: ChatGPT, Perplexity, Gemini, Claude, sometimes a Copilot or other shell. The person clicked a footnote, a chip, or a pasted URL. That is AI referral traffic.
A mention with no click is not traffic. A Google AI Overview impression in Search Console is not LLM referral traffic either. Google's generative reports are useful. They are Google. Keep them in a separate row.
Classic referral (a blog, a directory, a partner) stays in GA4 as referral. We do not relabel those sessions as LLM traffic because the destination URL also got cited once.
If the tagging is messy, GA4 will dump assistant visits into "(referral)" or a hostname you do not recognize. That is why we want a dedicated visitor analytics stream that already knows the AI referrer list, then we reconcile with GA4 instead of arguing about channel grouping.
Why mention tools are not traffic tools
Peec at $95 will tell you ChatGPT named the brand. Otterly at $29 will get a mention program started. Profound Starter at $99 annual will do that for ChatGPT only. None of those stickers is an LLM traffic analytics install.
Ahrefs Brand Radar at $199 plus the Ahrefs plan is brand visibility in a modeled AI index. It will not count the sessions that hit /pricing.
We will work from a mention export. We will not call it traffic.
Promptwatch Essential is $95/mo for one project. That is the SKU we use when a brand wants citation analytics and visitor analytics on the same bill. Kick-off is $199/mo when we need unlimited projects and seats. Explore is free and ChatGPT only. It can prove a click arrived. It cannot be the LLM traffic program.
How we report LLM traffic without fake ROI
Finance gets sessions and conversions from visitor analytics. Editors get citations. Engineers get crawler errors. We will not average those into an "LLM traffic score" for a slide.
A week of high citations and flat sessions is still a win if the prompt is a shortlist question. The buyer may convert later on branded Google, or not at all. We say that out loud. We do not invent an assisted-conversion model in a blog post.
A week of sessions with no matching citation can still be real. People paste URLs. Perplexity footnotes. Someone shares a ChatGPT answer in Slack. We read the referrer before we celebrate GEO.
Crawler logs close the stupid cases. If the page was blocked, you were never a candidate for a citation, and any LLM traffic you see is leftover or accidental. OpenAI's bot docs are the first check.
The stack next to Promptwatch
GA4: organic, paid, email, classic referral. Keep it.
Search Console: Overviews and AI Mode. Keep it. Google's AI features rules still apply to that slice.
Ahrefs or Semrush: who ranks in Google and who links. Keep it. Brand Radar is optional and is not the traffic pixel.
The Promptwatch script or GTM template: AI-referred sessions and conversions.
The prompt list: why a URL should have been the footnote in the first place.
Our GEO 2026 note is the page work after the log exists. The sister post on real-time referrals is the 2026 install checklist. This one is the definition: LLM traffic vs mentions.
What we will not do
We will not promise that more citations produce more LLM traffic. Many answers never offer a click.
We will not treat Perplexity the same as ChatGPT in the traffic column. Perplexity shows sources more often. ChatGPT often does not.
We will not replace GA4 with Promptwatch.
We will not sell Otterly or Peec as LLM traffic analytics.
FAQ
Is LLM traffic the same as AI platform traffic?
Close. We use LLM traffic for sessions from assistants. AI platform traffic is the same idea with a wider label. If the referrer is Google and the page showed an Overview, that is still Google unless you have a cleaner split in GSC.
Can we skip the Promptwatch pixel and use GA4 only?
You can. You will spend time cleaning channel groups, and you will not get citation-by-URL on the same account. We install the pixel.
Does Essential at $95 include visitor analytics?
Yes. That is why we start there for a single brand. Kick-off at $199 is the multi-project version.
What to do this week
- Write down how you currently label ChatGPT and Perplexity in GA4. If the answer is "we don't," start there.
- Install Promptwatch visitor analytics on production (script or GTM).
- Load 25 prompts that map to URLs you actually ship. Essential ($95) unless you need Kick-off ($199) for more than one brand.
- For one week, put citations in one column and LLM sessions in another. Do not average them.
- If you want us to build that report, email hello@1001seomedia.com with a GA4 export of your current referral view. Mentions are not traffic. We will not report them as if they were.