📖 Docs & full guide: How to track AI referral traffic in GA4 Built by Nadia Mohamed — SEO Engineer · Technical SEO + GEO for SaaS. Try the free companion tools at nadiamohamed.me/ai-tools.

Referrals from AI assistants are the fastest-growing traffic source almost nobody measures cleanly. In GA4 they're scattered across dozens of referral rows — chatgpt.com, perplexity.ai, gemini.google.com — with no built-in "AI search" channel to group them.

ai-referral-tracker does that grouping. Point it at a GA4 CSV export or a raw access log and it tells you exactly how much of your traffic came from AI search, broken down by engine.

Zero dependencies. The engine list lives in a single, editable JSON file so you can keep it current as new assistants ship.

Install

npx @nadiamohamed/ai-referral-tracker ga4-export.csv
# or
npm install -g @nadiamohamed/ai-referral-tracker

Usage

ai-referral <file|-> [options]

| Option | Description | | --- | --- | | -f, --format <fmt> | auto · ga4 · log (default auto) | | --json | Emit JSON instead of a table |

Read from stdin with -:

cat ga4-export.csv | ai-referral -

Example

$ ai-referral examples/ga4-export.csv

AI-search referral traffic

  ChatGPT      612  ████████████████████████ 45.7%
  Perplexity   344  █████████████ 25.7%
  Gemini       151  ██████ 11.3%
  Copilot       96  ████ 7.2%
  Claude        54  ██ 4.0%
  You.com       12  ▌ 0.9%

  1373 of 7657 sessions (17.9%) came from AI search.

Inputs it understands

GA4 CSV export — from Reports → Acquisition → Traffic acquisition, use the export button. The tool skips GA4's # comment preamble and auto-detects the Session source and Sessions columns (it also accepts Source / medium, First user source, etc.).

Server access logs — Common or Combined Log Format (nginx/Apache). The referrer field is read from each line and hits are grouped by referrer.

What counts as "AI search"

The full, versioned list is in data/ai-sources.json — ChatGPT, Perplexity, Gemini, Google AI Overviews, Copilot, Claude, You.com, Poe, Phind, Grok, DeepSeek, Mistral, Meta AI, and more. Each entry maps a friendly engine name to the hostname fragments that identify it. PRs welcome as new assistants launch — this file is meant to be a community-maintained reference.

Programmatic API

import { loadSources, classify, summarize } from '@nadiamohamed/ai-referral-tracker';

const engines = await loadSources();

classify('https://www.perplexity.ai/search', engines);
// -> { isAI: true, engine: 'Perplexity', vendor: 'Perplexity', matched: 'perplexity.ai' }

summarize(
  [ { source: 'chatgpt.com', sessions: 100 }, { source: 'google', sessions: 900 } ],
  engines
);
// -> { totalSessions: 1000, aiSessions: 100, aiShare: 0.1, engines: [...] }

Why not just a GA4 segment?

You can build one by hand — and then rebuild it every time a new assistant launches, in every property, forever. This keeps the source-of-truth list in version control, works on server logs GA4 never sees (bots that don't run JS), and drops straight into a reporting script or CI job.

Part of the GEO toolkit

Open-source tools I maintain for AI-search visibility:

• llms-txt-generator — spec-compliant /llms.txt from any sitemap
• geo-audit-cli — score any URL on AI-search readiness
• json-ld-schema-templates-seo — production JSON-LD templates optimized for AI citation
• awesome-generative-engine-optimization — curated GEO resources

Web-based versions (no install): AEO Analyzer, Keyword Clustering & more →

License

MIT © Nadia Mohamed — SEO Engineer, Technical SEO + GEO for SaaS & tech.