GEO & Measurement
GEO vs SEO: are ChatGPT, Claude and Perplexity citing your site? (and how to measure it)
Your Google traffic is slipping, your visitors now ask their questions to AI assistants — and you have no idea whether ChatGPT, Claude or Perplexity read your site, let alone whether they cite it. Google Analytics won't tell you. Here's the difference between SEO and GEO, the signals to watch, and how to actually measure them.
A quiet shift is under way. A growing share of searches no longer ends on a Google results page, but inside an AI-written answer — in ChatGPT, in Perplexity, in Gemini, or straight in Google's AI Overviews. For a site owner, the question is no longer just "where do I rank on Google?" but "are the AIs reading me, and are they citing me?"
The problem: the usual tools don't answer that question. And the "GEO" market is filling up with magic promises. Let's be honest — we've explained elsewhere why GEO is mostly sold as snake oil (see GEO is a myth). But one thing is very real, and perfectly measurable: who, on the AI side, reads and relays your content. That's what this article is about.
GEO vs SEO: two different questions
SEO (Search Engine Optimization) answers one question: "where does my site rank in Google's results?" You measure it in Search Console (impressions, clicks, position, queries).
GEO (Generative Engine Optimization) answers a different one: "does my site show up in AI-generated answers?" That's a visibility, an exposure — being read, being cited as a source — not a ranking. And crucially, it isn't measured in the same place.
The costliest confusion is assuming good SEO guarantees good AI presence. It isn't automatic: ranking first on Google helps, but doesn't guarantee a citation in an AI Overview or by ChatGPT.
Three AI signals you must never conflate
"AI traffic" doesn't exist as a single block. There are really three distinct signals, counting very different things:
- Training crawlers: bots that ingest your content to train future models (GPTBot, ClaudeBot, CCBot…). They read you, but send no one today.
- Live search bots: bots that fetch your page at the moment a user asks a question, to write and source the answer (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-User…). This is the "I'm being used live" signal.
- Referral clicks: a human who, from an AI answer, clicks a link and lands on your site (referrer
chatgpt.com,perplexity.ai…). This is the "AI is sending me real visitors" signal.
Adding these three signals together makes no sense: they're different units. Telling them apart is the first skill of serious GEO.
The main AI bots and their role
| Bot (user-agent) | Vendor | Role |
|---|---|---|
GPTBot | OpenAI | Model training |
OAI-SearchBot | OpenAI | ChatGPT Search index |
ChatGPT-User | OpenAI | Live fetch when a user asks a question |
ClaudeBot | Anthropic | Crawl / training |
Claude-User | Anthropic | Live fetch on a user's request |
PerplexityBot | Perplexity | Perplexity index |
Google-Extended | robots.txt token (Gemini / Vertex AI) | |
CCBot | Common Crawl | Corpus reused by many models |
Why Google Analytics is blind to AI traffic
Here's the point few marketing leads have internalised: Google Analytics 4 sees neither AI crawlers, nor — properly — the clicks coming from AI.
- Bots: GA4 is a client-side JavaScript tag. Crawlers generally don't run that tag, or are filtered as "bot traffic". So GPTBot or ClaudeBot can read your entire site without leaving a trace in GA4.
- Google AI Overview clicks: they arrive with a
google.comreferrer and blend into organic traffic. GA4 doesn't isolate them. - AI chat clicks: GA4's "AI Assistant" channel covers some assistants, misses others, and a large share of those visits arrive with no referrer and fall into "direct". Any figure you get is a floor.
As for Search Console, its "generative AI" report only shows impressions (exposure), no clicks or queries, and isn't available through its API. In other words: the official tools measure, at best, part of the exposure — never the real traffic on the AI side.
How to know if ChatGPT, Claude or Perplexity cite your site
There's no magic button, but several complementary methods — from most reliable to fastest:
1. Read your server logs (the one undeniable fact)
Every bot visit leaves a line in your logs: user-agent, page, timestamp. It's the only source that proves, beyond dispute, that ChatGPT-User read a given page at a given time. It's also what most consumer tools ignore.
2. Analyse your referrers
A human coming from an assistant often arrives with an identifiable referrer (chatgpt.com, perplexity.ai…). Properly categorised, these referrers reveal which AIs actually send you people — and to which pages.
3. Test by hand, in neutral conditions
Type your strategic queries into ChatGPT, Perplexity or Gemini and see if you're cited. But beware: log out of your account, neutralise location, vary the wording and repeat — model answers aren't deterministic, a single test isn't a measurement. (We cover five ways to check in Am I cited by ChatGPT?.)
4. GEO tracking platforms
Semrush, Ahrefs and others have added AI-visibility modules. Useful, but they're sampling estimates (they replay prompts, often via API), not your real traffic — cross-check them, never take them at face value.
Making your site readable by AIs (without buying magic)
There's no "GEO hack". What makes a site citable by AIs looks a lot like a good website, plain and simple:
- Be crawlable: don't accidentally block the bots you want to be cited by. The distinction between training bot and search bot is crucial — we explain it in GPTBot, ClaudeBot, PerplexityBot: should you block them?.
- Structure clearly: explicit headings, direct answers, structured data (FAQPage markup helps machines read your Q&A).
- Say verifiable things: models favour factual, dated, sourced content.
And the llms.txt file? Worth knowing, but let's be honest: it's not a standard and the major answer engines don't use it (we detailed it in llms.txt: what it's actually for). Don't build your strategy on it.
Measure, don't guess: AI traffic in Snorklee
This is exactly the gap Snorklee fills. Where GA4 is blind, Snorklee measures passively, server-side, the three signals separately:
- AI crawlers reading your pages, by family (OpenAI, Anthropic, Perplexity…) and by purpose (training, search).
- Human visits from AI chats, with the real source and landing pages.
- The "Crawled → Fetched live → Visited" funnel: from the page a bot reads to the human visitor it eventually sends you.
All in real time, cookieless, banner-free, hosted in the EU. You stop guessing your AI presence: you see it.
The bottom line
SEO measures Google. GEO measures your presence in AI answers — a real exposure, invisible to classic tools. Don't pay for GEO magic; invest in measurement: read your logs, categorise your referrers, test in neutral conditions, and use a tool that sees what GA4 can't.
Measure your AI presence. Stop guessing it.
How do I know if ChatGPT cites my site?
Three complementary ways: read your server logs (user-agents like ChatGPT-User or OAI-SearchBot appear there), analyse chatgpt.com referrers in your analytics, and test your queries by hand in ChatGPT under neutral conditions (logged out, repeated). A tool like Snorklee automates the first two.
Does Google Analytics see AI traffic?
No, not properly. GA4 doesn't detect AI crawlers (they don't run its tag), AI Overview clicks blend into organic, and many AI chat clicks arrive with no referrer and fall into "direct". Any GA4 figure on AI is a floor — or a complete blind spot for bots.
What's the difference between GEO and SEO?
SEO optimises and measures your rank in Google's results. GEO is about your presence in AI-generated answers (ChatGPT, Perplexity, Gemini, AI Overviews). The first is a ranking, the second an exposure — and they aren't measured in the same place.
What are OAI-SearchBot and ChatGPT-User?
Two OpenAI bots: OAI-SearchBot feeds the ChatGPT Search index, while ChatGPT-User fetches a page live when a user asks a question. Seeing them in your logs means ChatGPT relies on your content to answer.
Should you block GPTBot and ClaudeBot?
It depends on your goal. Blocking training bots (GPTBot, ClaudeBot) prevents your content from training models, but blocking search bots (OAI-SearchBot, PerplexityBot) stops you from being cited in answers. The distinction is essential.
Published July 2026. Bot names (user-agents) match the public declarations of OpenAI, Anthropic, Perplexity, Google and Common Crawl as of that date; they change, always check your logs. This article distinguishes exposure (being read/cited) from traffic (visits received): they are two different measurements.
See who reads you on the AI side: Snorklee's AI Traffic tab passively measures AI crawlers and visits from assistants — what Google Analytics can't see.