When a user types a question into Google and clicks your result, your analytics tool sees a referral from google.com. The attribution is clean. When a user asks the same question to ChatGPT and gets an answer synthesized from several sources including your site, your analytics often sees nothing at all.
This is not a technical failure. It is a structural feature of how AI search works. Understanding the gap is the first step to building measurement strategies that actually capture what is happening.
The zero-click problem
Google's AI Overviews present a direct answer at the top of results for many queries. A user who finds what they need in the Overview has no reason to click through to a source site. Google Search Console may show that your page has impressions for that query, but without clicks, there is no session in your analytics to count.
This trend predates AI Overviews: featured snippets already demonstrated that being cited as a source does not automatically mean receiving a visitor. AI Overviews make the pattern more common and the answers more elaborate.
ChatGPT and Perplexity referral behavior
ChatGPT Search was launched in late 2024 and expanded broadly. When ChatGPT includes a citation, it creates a clickable link in its response. Some users click those links; others do not. The referral that does arrive in your analytics may be labeled as coming from chatgpt.com, but the volume is often much lower than the number of times your content was consulted to build the answer.
Perplexity surfaces citations more prominently than ChatGPT does, and its design encourages clicking sources. Anecdotally, Perplexity tends to drive more referral traffic per mention than ChatGPT, but the absolute volume depends heavily on the query and your position in the citation set.
What shows up in your analytics and what does not
- Direct traffic increase: users who found your brand through AI and then navigated directly appear as direct traffic.
- Chatgpt.com or perplexity.ai referral: visible only when a user clicks a cited link.
- AI Overview impressions: visible in Google Search Console but only as impressions, not always clicks.
- AI crawler traffic: visible in server logs as GPTBot, PerplexityBot, ClaudeBot, but these are crawlers, not visitors.
- Brand search volume: if AI exposure is driving brand awareness, branded searches may increase.
Server logs: the underused signal
Your server access logs record every request to your site, including requests from AI crawlers. GPTBot is OpenAI's crawler, PerplexityBot is Perplexity's, and ClaudeBot is Anthropic's. Seeing these bots crawl your pages tells you which content is being indexed for AI retrieval, even if no users are visiting directly.
- Enable server-level logging if it is not already on.
- Parse logs to separate AI crawler traffic from human visitor traffic.
- Track which pages AI bots crawl most frequently.
- A spike in AI crawler visits before a corresponding referral traffic increase is a leading indicator.
llms.txt: a proposed solution
llms.txt is a proposed convention for websites to declare which content is available for AI training and retrieval, similar to how robots.txt works for search crawlers. The specification is still community-driven and not universally adopted, but several publishers are experimenting with it to communicate content permissions to AI systems.
Practical measurement approaches
- Monitor direct traffic trends alongside branded search volume for correlation.
- Set up UTM parameters on all outbound links in AI-cited content to distinguish clicks from AI referrers.
- Check Search Console for AI Overview impressions by page.
- Review server logs weekly for AI bot activity patterns.
- Track which queries trigger AI Overviews that include your domain.
- Use brand mention monitoring tools to track citations across AI outputs.
If AI reads your content to answer a question, you may have influenced the answer without ever seeing a visitor. Measurement needs to catch up to that reality.
Dera's view
AI search visibility is becoming a distinct operational metric from traditional organic traffic. The content most likely to be cited is content that is clear, structured, trustworthy, and answers a specific question well. Building for that standard is both a GEO strategy and a measurement hedge: content that gets cited eventually drives traffic through brand awareness and direct search even when the click does not happen immediately.