Agentic commerce · May 20, 2026 · 3 min read
AI Shopping Agents 101: Who's Actually Shopping Your Store
ChatGPT, Perplexity, Gemini, Copilot, and the browsing agents behind them — a field guide to the AI systems visiting D2C storefronts, what each one reads, and how their traffic behaves.
Somewhere in your analytics right now is a visit that wasn't a person: an AI assistant reading your PDP to answer a question you'll never see, for a shopper you haven't met yet. Multiply it by every assistant, every day.
This is the field guide: the kinds of agents touching commerce sites in 2026, what each reads, and what their traffic means. Not futurism — just what's observable in server logs and referral reports today.
The four kinds of agent traffic
- Assistant search — ChatGPT (OAI-SearchBot), Perplexity, Gemini, Copilot answering shopper questions with live retrieval. Reads PDPs, policies, reviews; produces cited answers and referral clicks.
- Training and index crawlers — GPTBot, ClaudeBot, CCBot and peers building the corpora that shape what models 'know' about your brand between crawls. No immediate traffic; long-term narrative.
- Task agents — operator-style systems that navigate your actual site: search, add to cart, sometimes check out. Still a sliver of volume, growing, and brutal at exposing stores that only work with a mouse.
- Commerce-native surfaces — Shopify's MCP endpoint and Catalog, ACP feeds, shopping integrations: agents querying your data directly, no page render involved.
How the traffic behaves
The signature pattern, consistent across our store network and industry reporting: assistant-referred visitors are few and ferocious. They arrive deep in consideration — the assistant already did the comparing — so they skip discovery, land on specific PDPs, and convert at multiples of cold traffic. Treating channel value as sessions × average CVR will make you underinvest here every time; the intent mix is different in kind.
The second pattern is invisibility. Most of the influence never registers as a visit: the shopper asked, the assistant answered from crawl and feeds, and the eventual purchase shows up as 'direct' three days later. What you can see — labeled referrals, agent user-agents in logs — is the shadow of a much larger effect.
What to instrument this quarter
Minimum viable visibility: segment known assistant referrers and agent user-agents in analytics; keep robots.txt welcoming to the crawlers you want (blocking them doesn't stop the answers — it just removes your side of the story); and run a monthly probe panel of your buyers' real questions across the major assistants, graded for accuracy.
Then close the loop on-site. An assistant-referred shopper is mid-conversation when they land; a storefront that can keep answering — grounded in your real catalog and policies — is the natural landing experience. That's the product Kinect runs: agent-legible store data on the outside, an AI sales rep on the inside, and measurement honest enough to publish. Shoppers who engage convert at 6–21% — engaged-cohort behavior, stated as such.
Frequently asked questions
How do I see agent traffic in my analytics today?
Segment referrers (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com) and check server logs for agent user-agents like GPTBot, OAI-SearchBot, PerplexityBot, and ClaudeBot. Expect the visible slice to undercount the influence.
Should I block AI crawlers to protect my content?
For a commerce brand, blocking is usually self-harm: assistants keep answering questions about you from feeds, retailers, and forums — you've only removed the primary source. Media companies have a paywall case; merchants rarely do.
Are AI agents actually completing purchases?
Yes, at small but real volume — through ChatGPT's Instant Checkout via ACP and through task agents driving normal checkouts. The near-term revenue story is still assistant-shaped research changing which store the human buys from.
Related reading
See Kinect on your store
An AI sales rep on your storefront and an agent-ready store behind it — trained on your catalog and policies, live the same day, measured against your own baseline.