AI bot referral traffic: The Complete 2026 Guide

When a user asks ChatGPT a question that references your URL, ChatGPT fetches the page. The request comes from the ChatGPT-User crawler. The same pattern applies to Claude, Perplexity, and other AI assistants. The referral traffic is one request per query. It is not a batch crawl like training data collection.

The Referer header in these requests is blank or internal to the AI service. ChatGPT-User does not send the user's page URL as the referrer. The request looks like a direct visit with no referrer. This makes identification harder. OpenAI's ChatGPT-User documentation describes the crawler behavior.

Separate AI referral traffic from training crawler traffic. The key is the user-agent string. ChatGPT-User sends this agent:

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible with ChatGPT-User/1.0; +https://openai.com/chatgpt-user

PerplexityBot sends:

Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://docs.perplexity.ai/docs/perplexitybot)

Claude does not have a separate referral crawler. All Claude traffic comes through ClaudeBot. You cannot distinguish referral fetches from training fetches for Claude. Perplexity's bot documentation confirms their single-agent approach.

Extract these with grep.

grep "ChatGPT-User" /var/log/nginx/access.log
grep "PerplexityBot" /var/log/nginx/access.log

Count referral requests separately from training requests. They have different implications for your site.

AI referral traffic is low volume compared to training crawlers. A site with 100,000 page views per day might see 100-500 ChatGPT-User requests per day. PerplexityBot might add another 50-200 requests per day. The volume is small but meaningful.

Track these metrics:

• ChatGPT-User requests per day
• PerplexityBot requests per day
• Top URLs requested by each AI referral bot
• Response status codes for each AI referral request

A 404 or 5xx response to an AI referral bot means the AI model could not access your content. This affects how your site appears in AI answers. Ahrefs' guide to AI search shows that AI models skip broken links. Your content must be accessible.

These two traffic types have different business value. Training traffic fills training datasets. It does not directly drive users to your site. Referral traffic represents real user interest. Someone asked an AI about your topic. The AI fetched your page to answer them.

Referral traffic is a leading indicator of AI-driven discovery. A high referral request count means your content is being used in AI responses. Monitor this trend. If referral traffic drops, your content may have fallen out of AI training data or your pages may be returning errors. Moz's AI content research discusses how referral fetches correlate with AI answer inclusion.

AI referral traffic comes from the AI provider's infrastructure. OpenAI's servers are in the United States. Perplexity's servers are also US-based. The IPs come from cloud providers: Azure for OpenAI, GCP for Perplexity, AWS for Anthropic.

The geographic origin does not reflect the user's location. A person in Japan asking ChatGPT about your site generates a request from a US IP. Do not use geographic data from AI referral requests for audience analysis. AWS's IP ranges can help identify Anthropic's cloud origin, but the data is not useful for user geography.

AI referral crawlers have short timeouts. ChatGPT-User waits about 10 seconds for a response. If your page takes longer, the AI model returns a timeout error. Your site will not appear in the AI response. This is different from training crawlers which tolerate slower responses.

Optimize your Time to First Byte for AI referral traffic. A fast server response improves your chances of being included in AI answers. Google's Core Web Vitals guidance applies here. Good performance helps every crawler.

Set up a dedicated dashboard for AI referral traffic. Filter by user-agent. Track ChatGPT-User and PerplexityBot separately. Monitor response status codes. A jump in 5xx errors for these agents means lost AI visibility.

Use Google Analytics 4 or a self-hosted tool like Plausible. Plausible supports custom event tracking for bot requests. Plausible's bot filtering can classify AI crawlers. Remember that most analytics tools filter bots by default. You may need to disable bot filtering to see AI referral traffic. Or use raw server logs for the most accurate counts.

The AI bot referral traffic audit. Note the gap between referral traffic volume and training traffic volume. Referral traffic is small but valuable. It measures real AI user engagement. Training traffic is large but indirect. Audit quarterly.