All posts
3 min readannouncement

Welcome to PerchLens

Why we built an analytics tool for the AI era — and what's different when ChatGPT, Claude, and Perplexity become real traffic sources.

We started PerchLens after a confused afternoon staring at our own analytics.

The "Direct" bucket had grown to 31% of sessions in a single quarter. That's the bucket where everything uncategorizable ends up — bookmarks, typed URLs, links from messaging apps, mystery traffic. It had been around 12% the year before. Something had clearly changed.

So we dug in. Most of it wasn't direct at all. People were finding us through ChatGPT, Perplexity citations, Claude answers, Bing-with-Gemini summary boxes. Browsers were stripping referrers. AI tools that did pass referrers used custom headers no analytics pipeline recognized. Every tool we tried — Plausible, Fathom, GA4 — was silently dumping all of it into Direct.

We wanted to know who was actually sending us visitors. The only way to find out was to write the thing ourselves.

The web changed. Analytics didn't.

Two years ago "search" meant ten blue links. Today it also means a chat window. People still type queries into Google, but they also ask Claude or ChatGPT or Perplexity, and a meaningful share of qualified traffic now arrives through the second path.

Only the first path shows up clean in your analytics.

The second has been quietly absorbing into Direct for about eighteen months. If you've been watching your direct bucket grow and assumed people were typing your URL from memory — they probably aren't. They're being recommended by an AI, and the AI is stripping the breadcrumb trail on the way through.

What PerchLens does about it

We identify traffic from AI assistants — ChatGPT, Claude, Perplexity, Gemini, Copilot, Phind — and surface each one as its own source. They sit alongside Google and Direct in your dashboard, with their own landing pages, conversion rates, and trend lines. Same mental model you already use; one new column you didn't have before.

The rest is the kind of thing you'd hope for in a 2026 analytics tool:

The AI-traffic part is the wedge. The rest is table stakes done carefully.

Why we think it matters

Not every analytics-tool launch deserves a blog post, so here's the honest version.

The flow of attention on the web is in the middle of a real change. AI assistants now sit between users and websites in a way that didn't exist three years ago. When ChatGPT recommends a tool, that recommendation is worth roughly what a front-page Hacker News mention is worth — and the founder watching their dashboard has no way to see it happen.

Concretely, that changes what content you decide to write more of next quarter, how you read your conversion rates, and where you spend distribution effort. It changes whether you're trying to rank in a Google result that's about to be summarized into an AI snippet, or trying to be the snippet.

We don't think AI is going to replace traditional search. We think both will coexist for a long time. But you can't decide what to invest in if you can only see half of what's happening.

What this blog is for

Roughly monthly. Two kinds of posts.

Data. Aggregate, anonymized findings from PerchLens-tracked sites. Which AI tools are sending the most traffic. What kinds of content get cited and what those pages have in common. How AI traffic converts compared to organic. Numbers, not vibes.

Craft. The smaller stuff. How we built the AI-traffic detector and where it currently gets things wrong. Why the tracking script is small and what we cut to keep it that way. How we think about privacy in a post-cookie world. The behind-the-scenes we'd want to read if someone else were writing it.

The subscribe link is at the bottom of the page if you want a heads-up when something new goes up. Otherwise check back when you remember.

Glad you're here.

— The PerchLens team