Not every visit from AI is a person clicking a link in a chatbot answer. Increasingly, the "visitor" is software acting for a person: an assistant fetching your page to answer a question, a cloud browser filling in a form, or an AI browser that looks exactly like Chrome. These agent visits are real, often useful and usually invisible in standard reports, or worse, mixed in with your human traffic.
This guide explains the three kinds of AI agent visit, how each shows up (or does not) in server logs and analytics, and how to segment or flag them so your conversion rates and audience reports stay trustworthy. If you want the strategic background first, read our explainer on what agentic search is.
Key Takeaways
- There are three kinds of AI visit to separate: user-initiated fetchers (no JavaScript), cloud browser agents (full browsers on the AI company's servers) and AI browsers on the user's own device.
- Fetchers such as ChatGPT-User, Claude-User and Perplexity-User appear in server logs but usually not in GA4.
- OpenAI's ChatGPT agent signs its requests with HTTP Message Signatures and a
Signature-Agentheader of"https://chatgpt.com", which you can log at your CDN or server. - AI browsers that run on the user's device can look like ordinary Chrome; you can usually only see their referrer, not the agent.
- Flag agent traffic at the edge (CDN or server) and pass the flag into analytics, rather than trying to guess from behaviour alone.
Three Kinds of AI Agent Visit
| Type | Examples | Runs JavaScript? | Where you see it | How to identify |
|---|---|---|---|---|
| User-initiated fetcher | ChatGPT-User, Claude-User, Perplexity-User | Generally no | Server/CDN logs | User-agent token; IP lists where published |
| Cloud browser agent | ChatGPT agent's cloud browser | Yes (full browser) | Logs, and possibly GA4 as a session | Signed request headers (Web Bot Auth); CDN bot labels |
| AI browser on user's device | ChatGPT Atlas and similar AI browsers | Yes | GA4 and logs like a normal browser | Often only the referrer (for example chatgpt.com) |
| Crawlers (not agents) | GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot | Varies | Server/CDN logs | User-agent token; verify IPs |
Crawlers are covered in detail in our posts on AI bots in server logs and whether to block or allow AI crawlers. This article focuses on the first three rows: visits triggered by a person's request.
1. User-Initiated Fetchers
When someone asks an assistant about a page, it may fetch that page in real time. OpenAI documents ChatGPT-User as the agent for user-initiated actions in ChatGPT and Custom GPTs. Anthropic describes Claude-User as supporting Claude users who ask questions, and Perplexity documents Perplexity-User for user-requested fetches. Each has a distinct user-agent token, and OpenAI and Perplexity publish IP lists (for example openai.com/chatgpt-user.json and perplexity.com/perplexity-user.json) so you can confirm a request is genuine. Anthropic does not publish IP ranges.
“Because these actions are initiated by a user, robots.txt rules may not apply.”
— OpenAI, Overview of OpenAI crawlers, Overview of OpenAI crawlers
Perplexity says something similar about Perplexity-User. In practice these fetchers rarely execute your analytics script, so they will not appear in GA4. Your server or CDN logs are the only place to count them. See how to verify AI crawlers for the IP-checking process.
2. Cloud Browser Agents
Agents such as ChatGPT agent run a real browser on the AI company's servers. They load pages, run JavaScript, click buttons and fill in forms. That means they can trigger your GA4 tag and show up as sessions, which can distort engagement and conversion data.
OpenAI's ChatGPT agent allowlisting page explains how to identify these requests. They are signed using HTTP Message Signatures (RFC 9421) and include Signature, Signature-Input and Signature-Agent headers, with the Signature-Agent value "https://chatgpt.com". Public keys are published at chatgpt.com/.well-known/http-message-signatures-directory. The same page lists how major bot-management providers label it, including a Cloudflare bot tag of chatgpt-agent.
“A Verified bot is a bot or agent that Cloudflare has confirmed is transparent about who it is and what it does.”
— Cloudflare Docs, Verified bots and signed agents
Cloudflare's documentation also notes that since 1 July 2026, "signed agents" are part of its Verified category, with a metadata field that marks a bot as Intermediary (an agentic service many users operate) or Direct. If you use Cloudflare, that field is a ready-made way to segment agent traffic; our post on Cloudflare AI Crawl Control covers Cloudflare's AI bot settings.
3. AI Browsers on the User's Device
AI browsers are the hardest to separate. A test published by Analytics Playbook in October 2025 found ChatGPT Atlas used a user agent identical to Chrome on a Mac, appeared under "Chrome" in GA4, and attributed search clicks to chatgpt.com / referral, the same as ChatGPT on the web. Browser behaviour can change, so re-test, but plan on the assumption that you cannot isolate these visits. Our post on AI browsers and SEO covers what this means for content.
Step-by-Step: Segment Agent Traffic
- Turn on full request logging. Make sure your server or CDN logs user agent, IP, and (if possible) the
Signature-Agentheader. Our log file analysis guide covers the basics. - Build a user-agent lookup. List ChatGPT-User, Claude-User, Perplexity-User and the crawler tokens, and tag each log line by type.
- Verify, do not trust. Match IPs to published lists where available. User-agent strings alone can be spoofed.
- Flag signed agents at the edge. Use your CDN's bot labels or check the Signature-Agent header, and add a flag such as
agent_type=chatgpt_agent. - Pass the flag into analytics. With server-side tagging or a data-layer variable set by your edge worker, send the flag as a GA4 custom dimension. You can then build a segment, comparison or exclusion.
- Keep humans and agents in separate views. Use the flag to exclude agent sessions from conversion-rate reports, and report agent activity on its own, much as you would exclude internal traffic in GA4.
- Review monthly. New agents appear often. Check unknown user agents in logs and update your lookup.
Where to Flag Agent Traffic
| Layer | Method | Strengths | Weaknesses |
|---|---|---|---|
| CDN / bot management | Vendor bot labels (for example Cloudflare's verified bot categories) | Verification handled for you; works before the page loads | Depends on the vendor's coverage and plan |
| Web server logs | User-agent lookup plus IP verification; log Signature-Agent header | Complete record, including non-JavaScript fetchers | Needs log access and regular processing |
| Server-side tagging | Send an agent flag as a GA4 custom dimension | Lets you segment GA4 reports | Setup effort; only covers visits that reach your tag |
| GA4 alone | Referrer (AI Assistant channel) and behaviour | No extra setup | Cannot see headers; cannot tell agents from people |
Signals That Suggest an Agent (Use With Care)
- Requests from cloud data-centre IP ranges with a mainstream browser user agent.
- Very fast, linear journeys: landing page, form, submit, with no scrolling.
- Bursts of identical journeys at odd hours.
- Form submissions with structured, complete answers but unusual contact patterns.
Behavioural signals are hints, not proof. Use them to investigate, and rely on signatures and verified lists to flag.
Common Mistakes
- Assuming GA4 shows all AI visits. Fetchers usually do not run JavaScript, and AI browsers look like Chrome.
- Confusing the AI Assistant channel with agents. GA4's AI Assistant channel groups visits by referrer (people clicking links from assistants), not by whether software made the visit.
- Blocking all agents by reflex. An agent may be booking, buying or researching for a real customer. Decide per use case.
- Trusting user agents without verification. Scrapers impersonate well-known bots.
- Counting agent form fills as human leads. Flag them so sales knows who, or what, submitted.
If you want your site to work well for agents rather than just detect them, see our guides to preparing your website for AI agents and why AI traffic shows up as direct.
Related Guides
- AI Search for Local Businesses: A Practical Starter Plan
- AI Search for Restaurants: Getting Recommended for "Where Should We Eat?"
- AI Search for Hotels: Winning Direct Bookings in AI Trip Planning
Frequently Asked Questions
Does ChatGPT agent show up in Google Analytics?
It can. ChatGPT agent uses a full cloud browser that runs JavaScript, so it may trigger your GA4 tag. Identify it at your server or CDN using its signed headers and pass a flag to GA4 if you want to segment it.
What is the difference between ChatGPT-User and GPTBot?
OpenAI documents GPTBot as a crawler for content that may be used to train its models, while ChatGPT-User handles user-initiated actions such as visiting a page to answer a question. They have different user agents and IP lists.
Can I tell if a visitor used an AI browser like Atlas?
Usually not directly. A third-party test found Atlas used a standard Chrome user agent, so you mainly see the referrer, such as chatgpt.com, when a visit came from a ChatGPT link.
Should I block AI agents from my site?
Not by default. Agents often act for real customers, for example researching or booking. Block abusive traffic, protect sensitive actions and monitor the rest.
What is Web Bot Auth?
It is an approach where bots and agents cryptographically sign their HTTP requests so websites and CDNs can verify who is making them. OpenAI's ChatGPT agent uses HTTP Message Signatures for this.
Know Who, and What, Is Visiting
AI agents are becoming a normal part of web traffic. Identify them at the edge, verify them against official lists and signatures, and keep them in their own segment so your human metrics stay clean. If you would like help auditing your logs and analytics setup, our technical SEO team can help, or you can request a quote.
References
- Overview of OpenAI crawlers – OpenAI
- ChatGPT agent allowlisting – OpenAI Help Center
- Verified bots and signed agents – Cloudflare Docs
- Perplexity crawlers – Perplexity Docs
- Does Anthropic crawl data from the web, and how can site owners block the crawler? – Claude Help Center
- Google Analytics and ChatGPT Atlas – Analytics Playbook (October 2025)
- [GA4] Default channel group – Google Analytics Help
