LLM referral traffic, model by model
Visitors from ChatGPT, Gemini, Claude, Perplexity and 12+ large language models — targeted to your pages and countries, each with its own referrer in Google Analytics.
What LLM referral traffic is
Every AI assistant people use today — ChatGPT, Gemini, Claude, Perplexity, Copilot, Grok and the rest — runs on a large language model. When one of them cites and recommends your brand in an answer and the user clicks, that visit is an LLM referral — a real user with a precise need, far more engaged than a generic search visitor. It is the same thing as AI referral traffic; "LLM" is simply the name analysts and GA4 practitioners use.
The umbrella term hides a lot of variety. ChatGPT sends the bulk of the volume to a broad audience. Perplexity sends far fewer visits, from people who are verifying and researching. Claude's visitors are professional and technical. Gemini's are mobile and mainstream. Treating "LLM traffic" as one number throws that information away — the useful view is per model.
PerkFuel gives you that view on demand: visits from any of the 12supported models, to the pages and countries you choose, each one carrying its model's referrer so GA4 can tell them apart.
Why these visitors are worth more
A recommendation from an AI assistant is not a search result. It changes who arrives, and how they behave.
AI assistants cite and recommend your brand
Inside the answer itself: your page is named as the place to go for exactly what the user asked — a recommendation, not a listing.
A real user, with a precise need, clicks
They have already explained their situation in their own words. When they click, they arrive knowing why your page is the answer — the most qualified visitor a website can receive.
Far more engagement than a generic Google visit
No skimming ten results. They read, they explore, they act — which is why every study finds AI-referred visitors converting at a multiple of ordinary search traffic.
LLM-referred vs organic search conversion in one dataset (4.87% vs 4.6% in another) — the gap varies, the direction does not. See the data →
LLM traffic in numbers
Small, growing, and better qualified than its size suggests. What the public data says, with sources.
<2%
of total referral traffic comes from LLMs on the average site today. Small — which is exactly why most teams have never looked at it.
Source: LLM referral traffic statistics →+65%
year-to-date growth in LLM referrals to B2B sites — uneven month to month, but the direction over any quarter is up.
Source: LLM referral traffic statistics →4.87% vs 4.6%
LLM referral vs organic search conversion in one dataset; 20% vs 7% in another. The studies disagree on the size of the gap, not on its direction.
Source: LLM traffic vs organic search: the conversion data →Every LLM referrer, in one table
The session source each assistant produces in GA4, and what to know about it. Copy the domains into a regex filter or a custom channel group to build an 'LLM' channel.
| AI assistant | Session source in GA4 | What to know |
|---|---|---|
| ChatGPT | chatgpt.com | Largest source by far; a share of clicks arrive as 'direct' (referrer stripped by the apps). |
| Gemini | gemini.google.com | Google domain — filter on the session source or it blends into Google organic. |
| Claude | claude.ai | Fastest-growing; apps often strip the referrer, so the GA4 number is a floor. |
| Perplexity | perplexity.ai | Passes the referrer reliably; the cleanest LLM source to measure. |
| Copilot | copilot.microsoft.com | Microsoft's assistant; distinct from Bing organic. |
| Grok | grok.com | Often opened from X — check it isn't attributed to social. |
 Meta AI | meta.ai | Inside WhatsApp, Instagram and Messenger; referrer meta.ai when passed. |
 DeepSeek | deepseek.com | Growing fast in Asia; referrer deepseek.com. |
 Mistral | chat.mistral.ai | European assistant (Le Chat); referrer chat.mistral.ai. |
 Manus | manus.im | Agentic assistant; referrer manus.im. |
 Poe | poe.com | Multi-model app by Quora; referrer poe.com. |
 Kimi | kimi.com | Moonshot's assistant, popular in China; referrer kimi.com. |
Identify it in GA4 in two minutes
Open Reports → Acquisition → Traffic acquisition, switch the primary dimension to Session source, and add a filter: source matches regex
chatgpt\.com|gemini\.google\.com|claude\.ai|perplexity\.ai|copilot\.microsoft\.com|grok\.com|meta\.ai|deepseek\.com|chat\.mistral\.ai|manus\.im|poe\.com|kimi\.comFor a permanent view, create a custom channel group with the same rule and name it "LLM". Keep the per-source breakdown next to it — the model matters more than the total. Step-by-step guide →
Get LLM referral traffic in four steps
Add the pages LLMs should send people to
Answer-shaped pages: comparisons, pricing, how-to guides, documentation, product pages with specs. Not the home page.
Select the models
All 12 for a category-level test, or the two or three that match your audience — ChatGPT + Perplexity for B2B, Gemini for mobile-heavy consumer markets, Claude for developers.
Pick your countries and pace
Visits originate from the markets you choose, at the daily rhythm you set, with natural day-to-day variation.
Read it per model in GA4
Each visit carries its assistant's referrer. Group them with a regex on the session source and you have an 'LLM' channel, broken down by model.
Why work at the model level
One channel, twelve models
'LLM traffic' is the umbrella; underneath it, each assistant has its own audience, link behaviour and referrer. PerkFuel lets you work at either level.
Category test or single model
Run all twelve for the whole channel, or narrow to the models whose visitors engage best on your pages.
Built for GA4
Every visit keeps its referrer, so a single custom channel group — session source matches the LLM domain list — captures all of it, per model.
Compare models on equal terms
Same landing page, same countries, same pace, different model. The engagement gap between them is the most useful number you will get this month.
How people use it
Three setups that treat the channel as a set of models, not a single number.
Marketing team scoping the channel
- Goal
- Get the team's own LLM traffic numbers, model by model, rather than industry averages.
- Setup in PerkFuel
- All 12 models on, three key pages, core markets, 20 visits/day for one month.
- What they get
- An 'LLM' channel in GA4 with per-model engagement.
Agency building an AI-channel offer
- Goal
- Show clients per-model AI presence in their own analytics, with a repeatable setup across accounts.
- Setup in PerkFuel
- One project per client, models by audience, a GA4 custom channel group defined once and copied across properties.
- What they get
- A standard 'LLM referrals' line in every client report, broken down by assistant.
Product team comparing models
- Goal
- See how each model's visitors behave on the same page, side by side.
- Setup in PerkFuel
- ChatGPT, Gemini, Claude and Perplexity on the same comparison page, same countries, same pace.
- What they get
- Four engagement curves for one page, one per model, in GA4.
Traffic by AI source
Go deeper
LLM referral traffic — frequently asked questions
See every LLM referrer in your analytics today
10 free visits from the models you choose. Launch, open GA4 Realtime, watch them land — each with its own source.
No card required · Setup in minutes · Track every visit