How to get traffic from AI agents (ChatGPT, Perplexity, Claude, Gemini)
Last updated: April 18, 2026
Treat each AI agent as its own channel. For each one: (1) allow its
crawler in robots.txt, (2) expose a machine-friendly
surface so it can read you, and (3) instrument referrals with the
agent's referrer host and a per-agent utm_source on links
you place yourself. Then build a dashboard that segments by referrer
host so you can actually see what's working.
How AI agents send you traffic
There are two distinct flows, and most teams accidentally measure only one.
- Citation clicks (humans). A user reads an answer in ChatGPT/Perplexity/Claude/Gemini/Copilot, sees your link in the citation list or footnotes, and clicks. Your analytics see the referrer host.
- Agentic fetches (bots). The agent itself fetches your
page on a user's behalf — to read it before answering, to follow a
citation, or to perform an action. These show up in your access logs as
named user agents like
ChatGPT-UserorClaude-User, not in your normal analytics.
Both matter. Citation clicks are revenue-adjacent traffic today. Agentic
fetches are leading indicators — a rising count of ChatGPT-User
hits to a URL means the model is actively pulling that page into answers.
Per-agent reference table
The five agents that matter for most sites in 2026, with the strings you need to filter on. Crawler UAs change occasionally — verify against each vendor's bot directory before hard-coding alerts.
| Agent | Citation referrer host | Indexing crawler UA | On-demand fetcher UA | Recommended utm_source |
|---|---|---|---|---|
| ChatGPT | chatgpt.com, chat.openai.com |
OAI-SearchBot |
ChatGPT-User |
chatgpt |
| Perplexity | perplexity.ai, www.perplexity.ai |
PerplexityBot |
Perplexity-User |
perplexity |
| Claude | claude.ai |
ClaudeBot |
Claude-User |
claude |
| Gemini / Google AI | gemini.google.com |
Google-Extended |
(uses Googlebot) | gemini |
| Copilot (Microsoft) | copilot.microsoft.com, m365.cloud.microsoft |
Bingbot (Bing index) |
(via Bing) | copilot |
Note: training-data crawlers (GPTBot,
Applebot-Extended, Bytespider, etc.) are a
separate category. They affect whether a future model knows about you, not
today's traffic. See the full
allowlist.
What each agent fetches before citing you
- ChatGPT — fetches HTML, reads
robots.txt,sitemap.xml, and (when present)llms.txt. Will follow same-origin links shallowly. Does not execute most JS. - Perplexity — most aggressive on-demand fetcher. Will
re-fetch a page near-realtime when a user asks a related question.
Sensitive to
Last-Modified. - Claude — uses ClaudeBot to build its index;
Claude-User fetches when a user explicitly asks Claude to read a URL.
Less reliant on
llms.txtthan ChatGPT and Perplexity. - Gemini / Google AI Overviews — uses the same Google
index Search uses, gated by the
Google-Extendedopt-in for generative use. If you allow Googlebot but disallow Google-Extended, you rank but you don't appear in Overviews. - Copilot — uses Bing's index. If you're not in Bing Webmaster Tools, you're invisible to Copilot.
Setting up a clean referral dashboard
You can build the whole thing in your existing analytics tool in an hour. The trick is doing it once, with named segments, instead of scrolling a referrer report every week.
- Create a custom segment "AI citation traffic" with
referrer host matching any of:
chatgpt.com,chat.openai.com,perplexity.ai,claude.ai,gemini.google.com,copilot.microsoft.com. This becomes your top-line AI number. - Create one segment per agent with the host(s) for that agent. This is what you'll trend week over week.
- Add a server-log dashboard counting hits per AI
user-agent string (
OAI-SearchBot,ChatGPT-User,PerplexityBot,Perplexity-User,ClaudeBot,Claude-User,Google-Extended). Group by URL — rising counts on a URL mean that page is being actively cited. - Tag every link you control that lives inside an agent
context (an embed widget, your own agent, a Slack bot) with the matching
utm_sourcefrom the table above. This disambiguates from organic citation clicks.
What converts: agent traffic vs. Google traffic
Anecdotally, citation traffic from AI agents converts at a higher rate per session than generic Google organic — the user has already read an answer that mentioned you and is clicking to verify or go deeper. Three things consistently fall out of the data we and others have looked at:
- Higher pages per session. Agent visitors arrive on a deeper page than organic and tend to read 1–2 adjacent pages.
- Higher signup-to-visit ratio for B2B. The model has done some pre-qualification; the user arrives knowing what you do.
- Lower volume than peak Google organic — for most sites, AI citation traffic in 2026 is 1–10% of Google organic by raw sessions, but a much larger share of net new pipeline.
The honest summary: it's a small, growing channel that punches above its weight on quality. Track it; don't expect it to replace Google overnight.
Next steps
- How to show up in ChatGPT results
- llms.txt and ai.txt: a copy-pasteable guide
- Why your Google referrals are dropping