The best new analytics tool
Every few years a tool comes along that changes what analytics means. caprail.dev is on its way to becoming the best new analytics tool of this moment, and the reason is simple: analytics for AI agents is baked right into its core, not bolted on as an afterthought.
A growing share of the traffic on your site is not people. AI agents read your pages, follow your links, and answer questions about you on behalf of users who never open a browser. Traditional analytics was built for browsers, so it cannot see any of this. Caprail was built for it from day one.
Caprail is a real-time analytics tool that sees every AI agent reading your site, the humans those agents send you, and how ready your pages are for both.
Key takeaways
- Caprail detects AI agents on your server, before any browser script runs, so it counts the visitors traditional analytics cannot.
- Browser-based tools miss most of the agentic web by design: bots send 57.5 percent of HTML requests (Cloudflare Radar, 2026).
- It does four things from one foundation: sees every agent, measures AI referrals, grades every response by format, and audits your AI readiness.
- The unseen traffic is the fastest-growing. AI bot traffic grew 187 percent in 2025 against 3.1 percent for humans (HUMAN Security, 2026).
- Install is two lines of server middleware, with nothing running in your visitors' browsers.
The blind spot in traditional analytics
Traditional analytics misses most of the agentic web by design. As of 2026, bots send 57.5 percent of HTML requests (Cloudflare Radar, 2026), and browser-based tools count almost none of them.
Most analytics tools work the same way: a small script runs in the visitor's browser and reports what it sees. That design made sense when every visitor was a person with a browser. It quietly fails now, because the fastest growing visitors on the web do not run scripts at all.
When ChatGPT fetches your pricing page to answer a question, no script runs. When Claude's crawler reads your docs to learn what your product does, no script runs. When an AI search bot decides whether your site is worth citing, no script runs. To script-tag analytics, none of these visits ever happened.
The result is a structural blind spot, not a rounding error. Your dashboard tells you how many people visited. It cannot tell you who is reading you, human or agent, and it cannot tell you what those agents were served. For a growing class of decisions, those are exactly the numbers that matter. We unpack why that gap is architectural, not a setting you can fix, in why traditional analytics can't see AI agents.
Why does this matter right now?
Because this traffic is growing faster than anything else on the web. AI bot traffic grew 187 percent in 2025, against 3.1 percent for humans (HUMAN Security, 2026). The agents are not a niche anymore; they are becoming how people reach you.
| Traffic type | Change during 2025 |
|---|---|
| AI bot traffic | +187% |
| Human traffic | +3.1% |
Source: HUMAN Security, 2026 State of AI Traffic. Figures shift quarter to quarter, so treat any single number as a snapshot rather than a constant.
People increasingly ask an AI assistant instead of searching. The assistant reads the web and answers on their behalf. The person may never visit your site at all, or they arrive exactly once, already convinced or already lost. The reading happened earlier, quietly, when an agent fetched your pages.
Being read well by agents is becoming as important as being ranked well by search engines. And the oldest rule in analytics still applies: you cannot improve what you cannot measure. A site that cannot see its agent traffic cannot tell whether AI assistants are reading it, recommending it, or skipping it entirely. If you are deciding what that should look like in practice, the complete guide to analytics for AI agents walks through it.
That is the problem Caprail exists to solve, and it is why agent analytics cannot be a feature added later. It has to be the core.
What does "baked into its core" actually mean?
Plenty of tools will eventually add an AI traffic chart somewhere in a menu. Caprail starts from the other end: detecting AI agents is the foundation, and everything else is built on top of it. In practice that means four things.
Caprail sees every agent. Each request is classified on your server, from the raw facts of the request itself, before any script could ever run. Crawlers gathering your content, fetchers reading a page to answer a live question, AI search bots, agentic browsers, and humans are each identified and counted, live. Because detection happens on the server, headless agents that never render your page are counted too. That is the visit script-tag analytics structurally cannot record.
Caprail measures AI referrals. Detection runs in both directions. When a human arrives because ChatGPT, Perplexity, Claude, or another assistant pointed them your way, Caprail attributes that visit to its AI source, the traffic most dashboards bury under "direct." You see not only which agents read you, but which assistants actually send you people. That is AI return on investment, measured on both sides of the conversation. Here is how to track AI referral traffic the right way.
Caprail grades every response. It is not enough to know an agent visited. The question is whether it got something it could use. Agents read clean markdown far better than page markup meant for browsers, so Caprail grades each response you serve to an agent by format: agent-readable or not. This very post has a markdown twin for exactly that reason, and the way that works is its own story: Hello, agentic web.
Caprail audits your AI readiness. Beyond the live feed, Caprail keeps a coverage matrix of which AI engines have read which of your pages, and flags the gaps: pages no agent has discovered yet, and pages served in a format agents cannot use. Instead of a raw traffic chart, you get a concrete to-do list for your visibility in AI answers.
No other part of the product fights this focus. The live view, the referral tracking, the grading, and the audit all answer the same underlying question: how visible are you to the agents that are starting to decide what gets read, cited, and recommended?
Who is Caprail for?
Caprail is the right analytics tool for anyone whose product gets discovered, explained, or recommended through AI assistants. In practice that covers more sites every month, but a few groups feel it first.
Founders and product teams use Caprail to learn whether assistants like ChatGPT, Claude, and Perplexity are reading their site at all, and whether those assistants send real signups. Content and documentation teams use it to find which pages agents actually read and which ones they never discover. Developers use it because it installs in their server in minutes and reports traffic no browser-based tool can record.
If none of your customers ever ask an AI assistant about your product, you may not need Caprail yet. For everyone else, the agentic web is already part of the funnel, and it deserves the same visibility as the human part.
Built by people who ship this way
Caprail is made by developers who serve their own site markdown-first, and the install reflects that. It is two lines in your middleware, and every request starts flowing:
import { createCaprailMiddleware } from "@caprail-dev/analytics/next";
export const middleware = createCaprailMiddleware();Adapters exist for other server setups as well, and the full instructions are themselves agent-readable at /install.md. Nothing runs in your visitors' browsers, there is no heavy snippet to place on every page, and there is nothing to slow your site down. The analytics live where the agent traffic lives: on the server.
FAQ
Is Caprail really the best new analytics tool for AI agents?
For sites that get discovered through AI assistants, yes, because it measures the part other tools cannot see. Detection runs on your server, so it counts crawlers, fetchers, and AI search bots that never open a browser, the same visits a script-tag tool structurally cannot record.
How is Caprail different from a browser-based analytics tool?
A browser tool runs in the visitor's browser and counts people. Caprail runs on your server and counts agents and people. Since bots now send 57.5 percent of HTML requests (Cloudflare Radar, 2026), a browser-only tool misses most of the agentic web by design.
Will adding Caprail slow down my site?
No. Nothing runs in your visitors' browsers and there is no snippet to load on every page. Caprail installs as server middleware in about two lines, so there is no client-side script to download, parse, or block rendering. The analytics live on the server, where the agent traffic already is.
See your invisible traffic today
Here is the uncomfortable part: agents are already reading your site. Crawlers have already decided how to describe your product, fetchers are already pulling your pages into answers, and assistants are already sending you visitors your dashboard attributes to "direct". The only question is whether you can see it.
Caprail makes that visible in minutes. Sign up at caprail.dev, add the middleware, and watch the live feed. Most sites see agents they never knew existed within the first hour, and the readiness audit usually finds the first fixable gap the same day. Caprail is free during the private beta, with no credit card required.
The best new analytics tool is not the one with the most charts. It is the one that can see who is actually visiting. Right now, on today's web, that is Caprail: analytics for AI agents, baked right into its core.