AI traffic · passively measured, never guessed

Your site's complete AI visibility: crawled, fetched, visited.

What AI bots crawl, what an AI fetches live to answer a request, and the visitors ChatGPT, Perplexity or Claude send you. All of it measured passively on your real traffic — no synthetic prompts, no extrapolated visibility.

Try it free for 14 days Run a free audit

tab · visibilite-ia Demo data

AI FUNNEL30 D

Crawled by AIs4 812

Fetched live317

Visited from an AI chat236

The funnel's three stages: Crawled = AI bot passes over your pages · Fetched live = pages retrieved by an AI while handling a request · Visited = humans arriving from an AI chat. Each stage stays separate, aggregated, with no analytics cookie.

It's AI observability applied to content and acquisition: observing what AIs actually do with your pages, rather than extrapolating visibility.

Real vs synthetic

Measured, not guessed.

Most AI visibility tools estimate your visibility by querying AIs with thousands of synthetic prompts. Snorklee reads your site's real AI traffic.

Typical AI visibility tools ℹ

Source: simulated prompts sent to AIs
Metric: estimated “share of voice”
Data: synthetic, outside your site

Snorklee

Source: real AI traffic on your site
Metric: observed crawls, live fetches and visits
Data: first-party, cookieless, hosted in Europe

Snorklee doesn't claim to know the exact question behind a fetch or a visit (that data isn't in the referrer) — it shows the AI traffic actually received.

Seen in your dashboard

AIs already read your site. Now you can see it.

Three dashboard cards, exactly as they appear: who crawls you, how many visits arrive from ChatGPT or Perplexity, and which pages AIs fetch to answer.

**AI activity detected** crawlers + visits Every assistant identified server-side: the bots indexing your site (ChatGPT, Claude, Perplexity, Gemini) on one side, visitors arriving from an AI chat on the other.

**AI presence** score + timeline Presence score, AI crawls, live fetches and AI chat visits — tracked over time.

**Fetched pages & breakdown** URLs + providers The pages AIs fetch and visit, and the breakdown by provider (OpenAI, Anthropic, Perplexity, Google).

Server-side recognised sources

The detection base behind the AI funnel.

Each AI crawler is detected server-side, then aggregated into the dashboard tiles: detected AI agents, AI activity over time and AI distribution. Human visits from AI chats are read separately from the referrer.

We keep this list current, and every night we re-download the IP ranges published by OpenAI, Perplexity and Anthropic. That is what lets us say a hit is verified rather than merely declared.

OpenAI

3 user-agents

GPTBot — associated with model training
ChatGPT-User — on-demand browsing
OAI-SearchBot — SearchGPT engine

Anthropic

3 user-agents

ClaudeBot — Claude indexing crawler
anthropic-ai — legacy Anthropic agent
Claude-Web — on-demand browsing

Google

1 user-agent

Google-Extended — associated with Gemini / Vertex AI

Perplexity

2 user-agents

PerplexityBot — indexing crawler
Perplexity-User — on-demand browsing

Mistral

1 family

MistralAI / Mistralbot — Mistral crawler (Le Chat)

ByteDance

1 user-agent

Bytespider — ByteDance AI crawler (Doubao)

xAI

2 user-agents

Grokbot — indexing crawler
Grok-User — on-demand browsing

Other LLMs & archives

9 user-agents

Applebot-Extended — Apple Intelligence
cohere-ai — associated with Cohere
CCBot — Common Crawl archive (base dataset for many LLMs)
YouBot — You.com engine
DuckAssistBot — DuckDuckGo Assist
Amazonbot — Amazon training (Alexa+)
PetalBot — Huawei AI engine
DeepSeekBot — DeepSeek crawler (China)
Diffbot — structured extraction (used by multiple LLMs)

Why it matters

AI traffic is no longer marginal.

In 2025-2026, ChatGPT, Perplexity, Gemini and their peers are becoming important access points to information. Users ask questions and often get answers built from web pages — sometimes with an outbound link, sometimes without a direct visit.

Your current tools can't see it. GA4 files 60 to 70% of traffic coming from AI assistants under “Direct”, and AI crawlers often represent 20 times the human traffic they send back — invisible to a classic JavaScript tracker, which never sees a bot go by.

Snorklee gives you a readable view: which bots come by, which pages they read, which get fetched live to answer — and the pages “crawled but never visited”, your best candidates for optimization. An editorial signal alongside search, social and direct.

Honest note: we measure AI visibility (who crawls, how often). We don't promise placement in ChatGPT or Perplexity answers — that depends on the LLM's own algorithm and nobody controls it.

Check my AI visibility — free →

The AI traffic tab

Everything the tab shows you.

Nine cards, all fed by your real traffic: the funnel and its curves, the standout moments, the sources, the crawlers, the pages — and an actionable AI summary. The probable stays separate from the certain.

Funnel Crawled → Fetched → Visited

3 stages + curves

The three counters with their change vs the previous period and a daily curve per stage — never added together: each stage keeps its own unit.

AI moments

timeline

First visit from an AI, first live fetch, weekly AI-share record, crawl spike: the standout moments, detected automatically and dated — with the AI's logo.

AI share of traffic

daily %

The share of your human visits coming from an AI assistant, day by day — to see whether the AI era already weighs on your acquisition.

Sources

assistant logos

ChatGPT, Perplexity, Claude, Gemini & co: who sends you humans, how many visits, and the trend — recognised server-side, never estimated.

AI crawlers

family + purpose

Every bot by family, with its observed purpose: training collection, search indexing, or live fetch while answering.

Pages as AIs see them

the lever

Page by page: crawled, fetched, visited. Pages “read, never visited” rise to the top — AIs consume your content without sending anything back: your first traffic reserve.

Pages visited from an AI answer

top visits

Where the humans sent by assistants land — ranked by real visits, with the change vs the previous period.

Snorklee Assistant summary

on click · 1 credit

A verdict, the numbers that matter, concrete actions and charts drawn from your real data. Nothing generates without your click — 30 to 120 generations included per month depending on your tier.

Invisible AI traffic (estimated)

separate tier

Text-fragment arrivals and the estimate of direct traffic closely following assistant fetches (temporal correlation, page by page). Always labelled “Estimated” — never blended with the measured.

All these cards are available as soon as your 14-day free trial starts.

How it works

Server-side detection, in two parts.

Human visits from AI chats: the snippet is enough. When a visitor clicks a link in ChatGPT, Claude or Perplexity, their browser loads your page with an identifiable referrer. The server classifies the visit as channel='ai' with the assistant name — zero friction here, nothing else to install.
AI bot crawls: server-side capture required. GPTBot, ClaudeBot or PerplexityBot read your HTML without running any JavaScript: no tracker can see them. To count them, enable the server beacon Snorklee provides (Express middleware, Cloudflare worker, Vercel/Netlify edge, PHP, WordPress plugin) — step-by-step guide in the dashboard Integration tab.
Aggregation in the AI traffic tab. No analytics cookies for either measurement. The AI traffic tab shows the Crawled → Fetched live → Visited funnel, the AI share of traffic, the moments, the sources, the crawlers by family and the pages involved.
Privacy documentation available. The stack is documented: Clever Cloud for the app, PostgreSQL and Cellar/S3; Scaleway Paris for the audit generator and domains; Brevo for email; DB-IP / Eris Networks for IP geolocation.

Référence

Who these bots actually are.

People often ask us whether GPTBot and ChatGPT-User are the same bot. They are not. The first collects pages to train a model; the second opens your page while an AI answers someone. Blocking both under one policy means cutting off traffic you meant to keep. So here, bot by bot, is what each one does — and which ones publish their IP addresses, which is what lets you check they really are who they claim to be.

Bot	Vendor	What it does	Identity verifiable
`GPTBot`	OpenAI	Collects to train a model	Yes — published IP ranges
`OAI-SearchBot`	OpenAI	Indexes for search	Yes — published IP ranges
`ChatGPT-User`	OpenAI	Reads the page live to answer	Yes — published IP ranges
`ClaudeBot`	Anthropic	Collects to train a model	Yes — published IP ranges
`Claude-User`	Anthropic	Reads the page live to answer	Yes — published IP ranges
`PerplexityBot`	Perplexity	Indexes for search	Yes — published IP ranges
`Perplexity-User`	Perplexity	Reads the page live to answer	Yes — published IP ranges
`Google-Extended`	Google	Collects to train a model	No — no ranges published
`Applebot-Extended`	Apple	Collects to train a model	No — no ranges published
`Meta-ExternalAgent`	Meta	Collects to train a model	No — no ranges published
`Bytespider`	ByteDance	Collects to train a model	No — no ranges published
`CCBot`	Common Crawl	Collects to train a model	No — no ranges published
`Amazonbot`	Amazon	Undetermined purpose	No — no ranges published
`mistralai-User`	Mistral	Reads the page live to answer	No — no ranges published
`DuckAssistBot`	DuckDuckGo	Reads the page live to answer	No — no ranges published
`PetalBot`	Huawei	Collects to train a model	No — no ranges published

A number is worthless if you do not know what it proves. Our three stages do not measure the same thing and do not count the same unit: we never add them up. Here is what each one establishes — and, more importantly, what it does not.

Stage	How it is captured	What it proves	What it does not prove
Crawled	Bot user-agent, read server-side	A bot fetched this page	Does not prove the page will be cited
Fetched live	Bot whose stated purpose is “assistant”	An AI opened the page during a conversation	Does not say what the AI did with it
Visited	Visit referrer (chat.openai.com, perplexity.ai…)	A human came from an AI answer	Misses arrivals with no referrer

FAQ

Frequently asked questions.

Does Snorklee guarantee placement in ChatGPT answers?

No. It measures visibility (who crawls, how often), not placement.

How are AI bots counted?

Server-side via a beacon; visits from chats via the referrer.

Which assistants are detected?

GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and others, on a maintained list.

Do you send prompts to the AIs?

No, never. Snorklee runs no synthetic prompts and extrapolates no visibility: it passively observes what AIs actually do on your site — bot crawls, live fetches and visitors sent from chats.

Do I need a cookie or a consent banner?

No. AI traffic is measured server-side for crawlers and via the referrer for visits from a chat, with no cookie and no banner. First-party data, cookieless, hosted in Europe.

How is this different from a typical AI visibility tool?

Typical AI visibility tools estimate your visibility by sending thousands of made-up prompts to AIs. Snorklee reads the AI traffic your site actually receives: bot crawls, live fetches and visits from chats. It's measured, not guessed.

How much does it cost?

One plan, Snorklee One: from €19/month ex. VAT, all included — full analytics, AI traffic, 30 to 120 Snorklee generations per month depending on the tier, up to 10 sites. 14-day free trial, cancel online anytime.

Start measuring your AI traffic today.

14-day free trial, cancel online anytime. Or run a free GDPR audit of your site (instant result, no signup).

Try it free for 14 days Contact