The State of AI Agent Traffic
Salespeak analyzes first-party website activity to understand how AI agents use B2B websites. This research separates automated agent access from human visits referred by AI products and documents the methodology behind every finding.
Agent traffic versus AI referrals
These are two different observations. Reports that combine them overstate what either one shows.
Agent traffic
Automated requests to your website from identified AI agents and crawlers, such as ChatGPT-User, ClaudeBot or PerplexityBot. No person is loading the page in a browser.
What Salespeak records: the requested URL, the classified agent and the time, captured at the website edge through a supported integration.
AI referrals
People who arrive on your website from an AI product, for example by clicking a link in a ChatGPT answer.
What Salespeak records: human page views whose referrer or campaign parameter identifies an AI product.
What this research does not show
- The buyer’s question. A request shows which page an agent accessed. Topics of interest are inferred from pages, not observed prompts.
- Citations. Access does not prove that an AI answer used or cited a page.
- People or deals. An individual agent request cannot be connected to a specific human visitor or deal.
- Control of AI answers. Better source information does not guarantee how an independent AI system describes or recommends a company.
How Salespeak observes agent requests
Salespeak records requests at the website edge through a supported integration on each participating site. Requests are classified as AI agent traffic from the user agent they declare. Since August 2026, Salespeak also verifies agent requests against provider IP ranges. Earlier data was classified by user agent only.
In this research, a site means one Salespeak customer account. Every finding in the Index states its population, period and limitations.
Research editions
Each study keeps its original sample, period and labels.
| Study | Period covered | Published |
|---|---|---|
| January to February 2026 study | January 8 to February 7, 2026 | February 22, 2026 |
| Optimization and agent visit study | August 2025 to September 2026 | September 1, 2026 |
An interactive explorer presents the January to February 2026 study.
Use this research on your own site
- Measure agent requests and AI referrals separately: Agent Analytics.
- Improve the company information agents can access: Agent Optimizer.
- Provide governed discovery and answers for external agents: Agent Interaction Platform.