Webclat / AI Visibility
Webclat / AI Visibility  /  agentic traffic

You searched "agent analytics" because of Profound. Here's what it's actually measuring.

Profound's marketing calls it Agent Analytics. What it measures is the AI-visibility sense of a three-way-ambiguous term - and it's the same question this site's own agentic-traffic cluster has been answering all along, engine-neutral.

In short

Profound's Agent Analytics feature reads CDN and server logs - not a client-side script - to detect, verify, and classify the AI agents and crawlers visiting your site, then ties that activity back to citations in AI answers. That is the visibility sense of "agent analytics," distinct from the product sense Pendo and Amplitude ship (measuring users talking to an AI assistant inside your own product). The engineering underneath the visibility sense - composite classification, server-side capture, spoofing checks - is vendor-neutral and documented on this site.

Landscape as of September 2026

Three things people mean by "agent analytics"

The term routes to at least two live products and two unrelated noise topics, which is why it's worth a page of its own before anything else. The product sense - what Pendo and Amplitude both ship - measures how your own users interact with an AI assistant built into your product: intent clustering, where the assistant fails, whether it moves conversion. That's a different discipline, a different buyer, and a different part of this estate; we don't cover it here. The visibility sense - what Profound calls Agent Analytics, and what this page is about - measures the AI agents and crawlers visiting your site from the outside: which bots, how often, doing what. There are also two noise senses worth ruling out explicitly: "agent" meaning a human call-center agent, and "agent" meaning infrastructure-monitoring software (an Azure or cloud logging agent). Neither has anything to do with AI. The full three-way disambiguation, including both noise senses, lives at webclat.com's "What is agent analytics?" hub - start there if you're not sure which sense brought you here.

What Profound's Agent Analytics feature actually measures, verified

Per Profound's own documentation and product pages [fact, verified via web search against tryprofound.com and its documentation, 2026-09-02]: Agent Analytics does not rely on a client-side script the way a conventional analytics tag does. It reads CDN and server-layer logs instead - Profound names Cloudflare, Akamai, Fastly, AWS CloudFront, Google Cloud CDN, and Netlify as supported connections, with WordPress support added since launch - to see every HTTP request from every recognized AI system, including the ones that never render a page or execute JavaScript. It verifies claimed AI user-agents against the vendors' official IP ranges to filter out spoofed traffic (a real problem this site documents separately - see spoofed agent user-agents), and it classifies each verified visit by apparent intent: citation (the bot is pulling content to answer a live question), indexing (discovering content for future retrieval), or training (reading content to improve a model). The stated goal is closing the loop from an AI citation through to on-site engagement.

That's a genuinely useful capability, and we're not here to diminish it: server-side, spoof-checked, intent-classified visibility into AI bot traffic is exactly the right shape of answer to "is anything from ChatGPT or Perplexity actually reading my site." [hypothesis, not verified] Anything about Profound's specific pricing tiers, rollout dates for individual CDN connectors, or engine coverage per plan should be checked against Profound directly before you rely on it - that detail changes faster than this page updates.

Where this overlaps our own agentic-traffic cluster, and where it doesn't

The mechanism Profound built - detect an AI system's request, verify it isn't spoofed, classify what it's doing, capture it without depending on a client-side script - is a general measurement-engineering problem, not a capability that belongs to one vendor. It's the exact problem this site's agentic-traffic cluster works through directly, engine-neutral and buildable into your own stack rather than bundled inside someone else's answer-engine-tracking subscription: how to detect AI agent traffic walks the composite signals - no single one proves an agent visit on its own; the AI agent user-agent directory lists which agents identify themselves and which present as an ordinary browser; why AI/agent traffic is undercounted 3-4x explains the structural reason client-side tools miss most of it in the first place; and is agent traffic real, and worth measuring? makes the case for why any of this is worth the engineering effort before you build it. On the capture side, capturing agent traffic server-side is the implementation pattern behind the "no client-side script" half of what Profound's feature does.

Why this is engineering, not an AI-visibility retainer

This property's own position is in its tagline: we are not an SEO agency, and this page is not a pitch to buy AI-visibility or GEO services from us. Whether you use Profound, build the detection and classification logic yourself, or some mix of both, the underlying work - instrumenting requests correctly, verifying identity instead of trusting a header, capturing events server-side when a client-side script won't fire - is measurement engineering, the same discipline as tag management. That's the only angle this site takes on any of it.

Common questions

What does Profound's Agent Analytics feature actually measure?

Per Profound's own documentation and product pages, Agent Analytics reads CDN and server-layer logs (Cloudflare, Akamai, Fastly, AWS CloudFront, Google Cloud CDN, Netlify, with WordPress support added) rather than relying on a client-side script, verifies claimed AI user-agents against official IP ranges to filter spoofed traffic, and classifies each verified bot visit by intent - citation, indexing, or training. It is a real, useful capability: server-side visibility into which AI systems are reading your site, and why.

Is "agent analytics" always about AI agents visiting a website?

No - the phrase covers at least two distinct, unrelated products. The visibility sense (this page, Profound) measures AI agents and crawlers visiting YOUR site. The product sense (Pendo, Amplitude) measures how your own users interact with an AI assistant INSIDE your product - intent clustering, failure points, conversion. Neither is wrong; they answer different questions for different buyers. The full three-way map, including two noise senses (call-center agents, infrastructure agents), lives at webclat.com's disambiguation hub.

Is Profound the only way to measure AI agents visiting a site?

No. Detecting and classifying agent visits without a client-side script is a general measurement-engineering problem, not a capability proprietary to one vendor - the same composite-signal classification and server-side capture pattern this site documents in its own agentic-traffic cluster. Profound packages it as part of a broader AI-visibility and answer-engine tracking platform; the underlying mechanism can be built directly into your own analytics stack.

Do I need a vendor platform to know if AI agents are visiting my site?

Not necessarily. A vendor platform bundles agent-visit measurement with its own answer-engine tracking and reporting layer, which is convenient if you already buy AI-visibility monitoring. If you'd rather own the instrumentation - server-side capture, composite classification, no dependency on a third party's log pipeline - that's a build, not a subscription, and it's the engineering this site's agentic-traffic cluster covers directly.

Same question Profound answers. Built into your own stack, engine-neutral.

We build the composite classification and server-side capture layer that measures AI agents visiting your site - on your own infrastructure, not inside a third-party subscription. Measurement engineering, not an AI-visibility retainer.

Talk to Us About the Model