Notes from the maker
Analytics, MCP, and what's being built, also in RSS.
Jul 26, 2026
how to track Claude referral traffic
claude.ai sends smaller but unusually qualified traffic, and almost none of it looks like a referral. here's why Claude clicks land in direct, how to tell Claude's crawlers from real people, and the steps to get a number you can act on.
Jul 26, 2026
how to track ChatGPT referral traffic
ChatGPT sends real traffic, but most of it never shows up as chatgpt.com. here's why it gets stripped of its referrer and bucketed as direct, and the steps that actually get you a number you can trust.
Jul 26, 2026
how to analyse AI agent conversations without paying per message
server-side conversation analysis bills you for every message your agent ever sends. here's the other way: your own model reads a sample over MCP, writes labels back, and the counts stay computed. the arithmetic, the loop, and the honest limits.
Jul 26, 2026
conversation intelligence: which numbers are computed and which are a model's opinion
counting conversations is arithmetic. deciding one was 'frustrated' is a judgement. analytics tools mislead people by printing both in the same font. here's how to tell them apart and how to keep them apart.
Jul 26, 2026
AI referral traffic: what to actually measure
once ChatGPT, Claude and Perplexity start sending you people, session count is the least useful number you have. here's what to measure instead: conversion by channel, which pages get cited, and how to tell AI crawlers from AI-referred humans.
Jul 26, 2026
agent observability: what to actually measure
traces are not metrics and token counts are not health. the six numbers that tell you whether an AI agent is working in production: tool-call latency percentiles, error taxonomy, re-ask rate, abandon rate, time to first token, and resolution rate, with a clear line between what is computed and what is a model's guess.
Jul 21, 2026
which commit moved the metric: deploy-aware analytics
every analytics tool shows you that a number dropped. none of them tell you which deploy dropped it, because they don't have your code. here's how to tie every metric change to the ship behind it, and ask your editor 'did my last deploy move signups?'
Jul 16, 2026
how to see how much traffic ChatGPT, Claude, and Perplexity send you
AI answer engines now refer real traffic, but most analytics tools bucket it as 'direct' or bury it in referrers. here's why AI referral traffic is easy to miss, and how to track chatgpt.com, claude.ai, and perplexity.ai as their own channel.
Jul 14, 2026
why product analytics gets expensive, and how to cap the bill
PostHog, Mixpanel, and Amplitude all price on volume, so the bill climbs as you grow, often into a nasty second-year surprise. here's how event-based pricing actually works, why it spikes, and how to make the cost predictable.
Jul 10, 2026
the best mixpanel alternatives in 2026, an honest roundup
if you're leaving Mixpanel over the bill, the free-tier cliff, or the learning curve, here's an honest look at the real alternatives (PostHog, Amplitude, smolanalytics, Plausible) and which one fits which problem.
Jul 9, 2026
instrument your app without writing tracking code
automatic event instrumentation for an app built with a coding agent: autocapture handles the baseline, and cursor or claude code writes the custom track() calls over MCP. no tracking code from you.
Jul 9, 2026
define product events from clicks you already captured
retroactive events let you name a business event like checkout from clicks smolanalytics already captured, no code, no re-deploy, retroactive to install.
Jul 9, 2026
analytics for apps built with cursor or claude code (over mcp)
the analyst for apps built with a coding agent. run one command and cursor or claude code reads your event log back to you in the editor. keep the analytics you already have; nothing gets replaced.
Jul 9, 2026
the analytics that autocaptures rage clicks, dead clicks, and js errors
autocapture rage clicks, dead click tracking, and javascript error analytics from one snippet, zero code. the frustration signals that tell you what to fix, cookieless.
Jul 9, 2026
the analytics ai that cannot make up your numbers
can AI analytics be trusted? most tools let the model write sql and hallucinate. smolanalytics never does. a CI test asserts the number is real, byte for byte, or the build fails.
Jul 8, 2026
how to see where users drop off
users drop off at the step where the biggest share of them leaves and never comes back. a funnel finds that step, and smolanalytics names the leak for you when you ask.
Jul 8, 2026
GA4 vs Plausible vs PostHog vs smolanalytics: an honest comparison
an honest, side-by-side look at GA4, Plausible, PostHog, and smolanalytics: what each is for, how they self-host and price, and how to pick.
Jul 8, 2026
the best analytics MCP servers, and the one honest question to ask of each
several analytics tools now ship an MCP server so your AI can query them. here's an honest look at the landscape, the one thing that decides whether you can trust the number, and the one that reads a log it does not own.
Jul 7, 2026
you can now just ask smolanalytics a question, right on the homepage
written july 2026, when the ask box lived on the homepage. the box is gone and the homepage is the test runner now, but the rule it was built on still holds: the ai calls the same deterministic reports the api serves, so it returns the real computed figure or nothing.
Jul 5, 2026
self-hosted analytics in one binary: the architecture, and whether you should run it yourself
most self-hosted analytics needs a database beside it, sometimes a whole cluster. here's the one-binary architecture (21 bytes per event on disk, measured, no ClickHouse, no Postgres), what self-hosting actually costs you, and how to get the control without the server.
Jul 5, 2026
how do i know what to fix in my product? (analytics that gives a verdict)
your ai assistant admits it hallucinates your numbers. mine can't. it's a CI test. here's the verdict concept, asking in your editor, and the trust story that lets you act on the answer.
Jul 5, 2026
the ga4 alternative for developers who just want the answer
GA4 confuses you, samples your data, and gets stripped by ad-blockers. here's an honest look at the alternatives for developers: where Plausible fits, where smolanalytics fits, and how to pick.
Jul 5, 2026
analytics for apps built with ai (lovable, bolt, v0, replit)
you built the app with Lovable, Bolt, v0 or Replit and never opened a terminal. here's how to add analytics the same way: paste one prompt, and then it reads your numbers back to you once a week instead of waiting to be asked. keep whatever you already have.
Jul 4, 2026
why i built smolanalytics
GA4 confused me into quitting analytics entirely. so i built the tool i wanted: a verdict that says what to fix, exact answers in my editor, one isolated instance per project.