Skip to main content
AI Traffic shows what AI systems and their users do with your site. It leads with evidence of what AI systems crawl, retrieve, and cite, then shows the visible referral and conversion tail from platforms like ChatGPT, Claude, Perplexity, and Gemini. Open it from Insights › AI Traffic in the project sidebar. It has four tabs:
  • Overview - AI traffic signals from training crawls and citations through answer fetches to referral sessions and conversions, then your top pages
  • Crawls & fetches - training-crawler and answer-fetcher activity by platform, with recent answer fetches
  • Referrals - referral sessions, conversion events and direct traffic, broken down by AI engine and landing page
  • 404 Demand - missing pages AI systems keep requesting
To compare Google search clicks with AI citations and review page-level opportunities, open Insights › Audit and select the Pages tab.

What AI Traffic Tracks

AI Traffic answers two questions:
  1. What are AI systems retrieving and citing? - Detects crawlers and answer fetchers such as GPTBot, ClaudeBot, and PerplexityBot, alongside citation evidence from Visibility.
  2. What visible human activity follows? - Tracks attributable referral sessions and conversions without treating them as the full measure of AI-driven demand.
AI referrals are systematically undercounted because most AI-driven visitors arrive as direct traffic. Referral and conversion metrics are the visible tail of AI traffic, not a complete linear funnel. AI Traffic is fed by sensors you connect once: Cloudflare, edge middleware or a Vercel log drain for crawler and answer-fetcher requests (the drain also records AI-tagged visits as at-least counts), GA4 or PostHog for referral sessions and conversions, the tracking snippet, or edge middleware version 0.13.0 or later, for first-party referral page views, and Search Console for classic-search comparison. DevTune classifies each request server-side and presents the data with trends, platform breakdowns, and top pages.

Why AI Traffic Matters

AI Platforms Are a Growing Traffic Source

As more users rely on AI assistants to find information and products, the traffic they send to websites is becoming a meaningful channel. Understanding this traffic helps you:
  • Measure retrieval and visibility - See which content AI systems fetch and cite before looking at downstream activity
  • Track AI bot crawling - Know which AI platforms are indexing your content, how frequently, and which pages they visit most
  • Interpret the visible tail - Pair AI Traffic with Visibility while treating attributed referrals and conversions as an undercounted fraction

Beyond Traditional Analytics

Standard analytics tools like Google Analytics do not break down AI bot traffic or AI referral traffic in useful ways. AI Traffic is purpose-built to:
  • Distinguish AI bots from regular crawlers
  • Identify referral traffic specifically from AI conversation platforms
  • Detect AI-sourced visits via utm_source parameters
  • Show trends and breakdowns by AI platform

Plan Requirements

What AI Traffic includes depends on your plans: On an Agent Picks plan alone, AI Traffic opens on Crawls & fetches with the coding agents’ fetches, and 404 Demand lists the missing pages coding agents asked for. Overview, Referrals, crawl volume, Search Console, GA4 and PostHog need an AI Search plan. DevTune stores only coding-agent requests for these accounts, so buying AI Search later starts the rest of AI Traffic from that day. Coding agents are recognised by edge middleware, the Vercel log drain and the tracking snippet. The Cloudflare log pull can’t tell coding agents apart, so on an Agent Picks plan alone it adds nothing; connect one of the others. The monthly event allowance and how long raw events are retained scale by plan; see Working with Projects for the AI Search figures. An Agent Picks plan alone allows 100,000 events a month, kept for 90 days. The allowance counts AI crawler, assistant and answer-fetch requests, plus visits from AI answers. Other human visits are not counted. See event allowance for referral visits two sensors both record.

What Gets Detected

Bot Crawlers

DevTune detects AI bot crawlers by their user-agent strings. The list includes: DevTune adds newly recognised AI bots to its list centrally, so you don’t need to change your setup to detect them.

Coding agents

Claude Code, Gemini CLI, OpenCode, Qwen Code, Crush and Aider fetch pages while developers work with them. DevTune counts these requests as answer fetches, like any other answer fetcher, and also reports how many of the answer fetches came from each coding agent. The coding-agent count is a subset of answer fetches: it never adds to or changes the answer-fetch total. Coding agents are attributed from edge middleware and the tracking snippet, which record each request’s own user agent. Qwen Code, Crush and Aider are recognised only through edge middleware and the snippet: Cloudflare pulls don’t include their requests, so on a hostname that uses the Cloudflare log pull, their fetches don’t appear in AI Traffic. Some answer fetches can’t be attributed: Cloudflare pulls, for example, arrive in buckets that can combine requests from a hosted fetcher and a coding agent under one user agent. When your window includes answer fetches that couldn’t be attributed, DevTune gives the coding-agent count on its own instead of a share of all answer fetches. The API and MCP tool also report how many answer fetches couldn’t be attributed (unattributedAnswerFetches). DevTune started counting coding agents when this feature was released. Earlier fetches stay in your answer-fetch totals but are not attributed to a coding agent, and the share is measured from the counting start date: both the coding-agent count and the answer-fetch total it is compared with begin there. The AI Traffic dashboard, the API traffic/summary endpoint (codingAgents) and the devtune_get_traffic_summary MCP tool report the same counts, with that date. The API traffic/coding-agents endpoint and the devtune_get_traffic_coding_agents MCP tool return the same coding-agent counts and coverage without the answer-fetch totals. With 404 Demand and sensor status (traffic/sensors, devtune_get_traffic_sensors), they are the reads an Agent Picks plan alone can use.

AI Referral Traffic

DevTune detects visitors arriving from AI platforms by checking the referrer URL:

UTM Source Detection

DevTune also checks the utm_source query parameter for AI platform identifiers (e.g., chatgpt, claude, perplexity, gemini, copilot). This catches traffic from AI platforms that include UTM parameters in their outbound links. Clicks on ads you run inside an AI platform, such as ChatGPT ads, are not counted as AI referrals. DevTune treats a visit as paid when its utm_medium is a paid medium (cpc, ppc, paid, paid_ai, cpm and similar), even if it arrives from chatgpt.com. The tracking snippet also treats a landing URL carrying OpenAI’s ad click ID (oppref) as paid; GA4 and PostHog don’t pass that ID to DevTune, so for those sources only the utm_medium counts. Tag your AI platform ads with a paid utm_medium to keep them out of AI referrals and Results.

Traffic Classification

Every visit is classified into one of three types:
  • Bot - An AI crawler indexing your content
  • Referral - A human visitor arriving from an AI platform
  • Other - Regular traffic (not AI-sourced). It does not count toward your monthly event allowance
Classification happens entirely server-side. The tracking snippet sends raw signals (user agent, referrer, page URL) and DevTune handles all detection logic. This means detection rules are updated centrally without requiring snippet changes on your website.

How It Works

  1. Your sensors report requests and visits: a crawler sensor (Cloudflare, edge middleware or a Vercel log drain) for AI crawls and answer fetches, GA4 or PostHog for referral sessions and conversions, and the optional tracking snippet, or edge middleware version 0.13.0 or later, for referral page views; a Vercel log drain also records AI-tagged visits as at-least counts
  2. DevTune classifies each request by traffic type and AI platform
  3. Data is summarized into hourly and daily views
  4. Insights › AI Traffic displays trends, platform breakdowns, and top pages
The tracking snippet is a single <script> tag of under 1 KB. It uses navigator.sendBeacon() for minimal performance impact, and generates an anonymous session ID (stored in sessionStorage) to count unique sessions without tracking individual users.

Privacy

  • The snippet does not use cookies
  • Session IDs are anonymous and stored only in sessionStorage (cleared when the tab closes)
  • The snippet collects no personally identifiable information
  • All data is associated with your project, not with individual visitors
The crawler sensors handle one more piece of data. To check that a request claiming to be an AI crawler really comes from that operator, edge middleware sends the address the request arrived from, and the Cloudflare log pull uses it when Cloudflare’s analytics include it. An address is sent only for a request whose user agent matches a known AI bot, so ordinary browser traffic never carries one. DevTune uses the address to reach a verdict, then discards it. It keeps the verdict and the reason for it, never the address:
  • Verified — the address, or a valid Web Bot Auth signature, matches the operator’s published networks or keys.
  • Identity mismatch — the address contradicts the claimed operator.
  • Unknown — the check could not be made: the sensor sent no address, the operator publishes no way to check, or the check was temporarily unavailable. Unknown requests recorded before DevTune kept the reason appear as before verification.
Anyone can send a crawler’s user agent, so a person whose client does that would have their address sent and discarded in the same way. Set trustedProxy: 'none' in the middleware to send no address at all; that traffic is stored as unverified. See crawler identity checks for the operators DevTune can check.

Getting Started

  1. Choose Your Sensors - Work out which sensors you need
  2. Set Up Sensors - Connect the sensors that feed AI Traffic
  3. Read AI Traffic - Explore your demand data

Next Steps