More and more sites see chat.openai.com in their traffic sources. Where this traffic comes from, how to track it properly, and how to get more.
Last week, a client sent us a screenshot of their Google Analytics. In the traffic sources, a referrer they'd never seen: chat.openai.com. 23 visits in March. Not huge, but they wanted to understand.
"What is this? How does someone arrive on my site from ChatGPT?"
The answer is simple: someone asked ChatGPT a question, ChatGPT cited your site in its response (with a link), and the user clicked. You just got your first AI referral visitor.
Since late 2025, ChatGPT integrates web search into its responses. When a user asks a factual or commercial question, ChatGPT searches the web in real time and cites its sources with clickable links. Perplexity does the same from inception. Gemini too, via its "sources" at the bottom of responses.
The mechanism is different from a Google click. On Google, the user chooses from 10 blue links. On ChatGPT, the user reads a constructed response, and if they want to go deeper, they click on the source. This visitor is already convinced your content is useful. The AI pre-qualified them for you.
That's why the conversion rate of this traffic is often 2 to 3 times higher than classic organic traffic. Less volume, much more quality.
By default, GA4 classifies traffic from ChatGPT under "Referral," which is correct. But it gets buried in the mass of referrals. To track it properly, create a custom segment.
In GA4, go to Explore, create a new segment with the condition: source contains "chat.openai" OR "perplexity" OR "gemini.google". Name it "AI Traffic." You'll immediately see how many visitors come from LLMs, which pages they visit, and whether they convert.
On our B2B clients' sites, this traffic represents between 2% and 8% of total traffic in March 2026. Low in absolute volume, but the trend is clear: it doubles every 3-4 months.
ChatGPT doesn't cite at random. It favors pages that directly answer the question asked, with verifiable information, a clear structure, and recognized authority on the subject.
Our client who got those 23 visits? It was their "Guide to electronic invoicing in Belgium" page that was cited. A page with precise regulatory information, regularly updated, with FAQPage structured data and cited official sources. Exactly the type of content that LLMs love.
In contrast, their commercial pages ("our services," "why choose us") have never been cited. Too vague, too self-promotional, not enough factual data for the AI to extract and attribute.
Create "answer-first" pages. For each question your prospects ask ChatGPT, create a page that answers directly, clearly, with data. Not a sales pitch. A genuine answer that the AI can cite as a source.
Add structured data everywhere. FAQPage schema on your FAQ pages, Article schema on your blog posts, Organization schema on your homepage. Structured data is the language that AI crawlers read natively.
Build third-party mentions. The more your brand is mentioned on Reddit, industry forums, press articles, and review platforms, the more likely ChatGPT is to cite you. It's not enough to have a great website — others need to talk about you too.
23 visits from ChatGPT may not seem like much. But the trajectory is clear: those who have 23 today will have 230 in a year. And those 230 visitors will convert at 3x the rate of your Google traffic.
This is early. The companies that set up AI referral tracking now and optimize for it will have a massive advantage in 18 months. The others will wonder where their competitors' leads are coming from.
Co-fondatrice et CEO d'AISOS. Expert GEO, elle accompagne les entreprises dans leur strategie de visibilite Google + IA.