What we know about AI agent traffic on e-commerce in 2026 — and what your analytics aren't telling you
Corrected 24 September 2026. The first version of this post quoted traffic figures attributed to OpenAI, Perplexity, Cloudflare and Akamai that we could not trace to a published source, and described patterns from "Tactical's early-access classifier" that it could not have measured, such as agents comparing your prices on other sites. We have removed them. Everything below is either documented, with a link, or clearly our opinion.
What we know about AI agent traffic on e-commerce in 2026
Some of the visits to your store aren't people. They're AI systems: crawlers collecting pages to train models or build an AI search index, fetchers reading a product page because someone asked ChatGPT or Perplexity about it, and increasingly agent browsers that shop on a person's behalf.
Standard e-commerce analytics either filter that traffic out as "bots" or, when it runs in a real browser, count it as a person. That made sense when bots were mostly scrapers. It makes less sense when the visitor is doing a customer's research for them.
What's documented
- The AI operators publish their crawlers. OpenAI lists its crawlers and what each is for (OpenAI: bots), from GPTBot, which collects training data, to ChatGPT-User, which fetches a page because a user asked. Perplexity does the same (Perplexity: bots).
- Google Analytics excludes known bots, and you can't turn that off. GA4 filters traffic from known bots and spiders using Google's research and the IAB's list (Google Analytics Help).
- Most AI crawlers don't run JavaScript, so any analytics that works through a browser tag, GA4 included, never sees them at all. Only your server logs, your CDN or a server-side install can.
We are not going to quote a share-of-traffic number here. The published figures vary with who is measuring and how, and a number without a source is exactly what this post shouldn't add to.
What those agents do on a store
What an agent does depends on what it is:
- Crawlers fetch pages, often many in a row, and don't interact with them. They are how AI answers learn what you sell and at what price.
- Fetchers arrive one page at a time, because a person asked an assistant a question your product page answers.
- Agent browsers load pages in a real browser, can add to cart and, increasingly, check out.
What you can't see from your own store is what the agent did elsewhere: the prompt it was given, the other stores it visited, the prices it compared. Anyone who says they can show you that from your site alone is guessing.
What standard analytics miss
| Signal | What GA4 shows | What's actually happening |
|---|---|---|
| AI crawler fetches | Nothing: no JavaScript runs | A crawler read the page |
| Known AI bots | Excluded from reports | Bots you can't see or count |
| Agent browser sessions | An ordinary visitor | An agent shopping for someone |
| Visits from AI answers | Often "referral" or "direct" | A person who came from ChatGPT or Perplexity |
Practical takeaways
1. Stop treating "bot traffic" as one bucket. A training crawler, a user-triggered fetcher and a scraper want different things from you. Your server or CDN logs show which ones name themselves.
2. Watch your structured data. Product schema, JSON-LD prices and availability are what agents read. A product page that says $89 while its schema says $99 is telling agents two things.
3. Look for the agents your analytics count as people. Unusually regular sessions, deep product browsing with no scrolling or mouse movement, or a cart with no browsing before it are worth a second look.
What we're building
Tactical detects AI crawlers (with a server-side install), agent browsers and AI-referred visitors on your own store, scores agent sessions by intent, and shows which products agents read. The free Scout plan covers 100 agent sessions a week, with no credit card, on Shopify (App Store), WooCommerce (WordPress.org) or any custom storefront via one script tag.