Digital analytics platforms — like Google Analytics 4 (GA4) or Adobe Analytics — and real-user monitoring (RUM) tools aren’t capturing the wave of new AI-driven shopping traffic. This isn’t because these AI shopping bots are trying to evade tracking. It’s because analytics tools were designed to measure human activity in browsers, not AI agents that fetch content in fundamentally different ways.
Most reputable AI agents identify themselves in HTTP headers with clear user-agent strings (often ending in “-User,” such as Claude-User or Perplexity-User). That means they’re not trying to sneak in. But your analytics dashboards won’t show them today.
Why Analytics Tools Miss “-User” Bots
Analytics beacons and RUM libraries depend on JavaScript running in the browser. Bots like Claude-User, Perplexity-User, and ChatGPT-User typically fetch raw HTML from your server without executing JavaScript.
- Perplexity documents Perplexity-User with published IP lists.
- Anthropic describes Claude-User as user-initiated access.
- OpenAI’s ChatGPT-User is referenced but not documented as running site JavaScript.
- By contrast, Googlebot is transparent with its user agents and can render JavaScript if needed.
The key point: if JavaScript doesn’t run, your analytics beacon never fires.
How Server-Side Fetchers Work
When a shopper types a query into an AI assistant — like “Show me the warmest full-length women’s parkas” — the AI doesn’t open a browser window on their behalf. Instead, it dispatches a server-side fetcher:
- A program is running in the AI provider’s backend.
- It sends HTTP requests directly to the retailer’s site.
- It retrieves the HTML and parses it internally.
Some providers can run these fetchers in headless browsers (rendering JavaScript invisibly), and a few are adding full browser sessions for more complex interactions. But today, the vast majority of consumer AI shopping traffic is raw server-side fetching.
That’s why only companies with access to CDN/edge logs or server-side instrumentation can see this traffic at all.
The Illusion of “Growth”
You may have seen headlines claiming AI-driven traffic is exploding — like David Bell of Previsible reporting a 527% increase in August 2025. That number is real, but here’s the nuance:
- His team configured GA4 to catch traffic when users click a citation link from Perplexity or another AI system.
- But GA4 automatically filters out most known bots.
- So what’s measured is only referral traffic from human clicks on citations, not the full volume of server-side bot fetches happening upstream.
In other words: AI agents may be pulling your product pages constantly, but you only see the tiny fraction of users who click through.
What We Saw in the Data
At Yottaa, we looked across 700+ retail sites from August 7, 2025 to September 12, 2025. This included 525 million sessions. In the RUM data, 295 sessions (essentially 0%) presented a “-User” user agent string. That’s expected, because non-JS fetches don’t fire RUM beacons and these 295 sessions are likely just impersonators spoofing the bot.
So if you’re relying only on GA4, Adobe Analytics, or a RUM dashboard, you’re seeing — at best — the tip of the iceberg.
What’s Next
Analytics tools are browser-based by design. They’re great at telling you that “X% of traffic last week was on mobile Firefox”, but they’re blind to the new AI agentic shopping layer.
Working with Fastly, our Web Performance Services team is creating novel hybrid detection approaches that combine:
- Edge/CDN logs (for non-JS fetches, IP/ASN checks, bot behavior).
- Browser-based RUM (for human shoppers and Web Performance API insights).
Our goal is to shine a light on the invisible AI shopper traffic and help retailers understand how much of their digital storefront is already being mediated by bots.
If you’d like to dig into what this means for your storefront, reach out. We’d love to share what we’re seeing and how to prepare for the wave of AI shopping bots.
