GA4 vs Plausible vs PostHog vs smolanalytics: an honest comparison

GA4, Plausible, PostHog, and smolanalytics solve overlapping problems from very different directions. GA4 (Google Analytics 4) is free, event-based web and app analytics that ties tightly into the Google Ads loop, though its low-traffic data thresholding and exploration sampling are widely documented frustrations. Plausible is simple, privacy-first, cookieless web analytics you can use hosted or self-host as the AGPL Community Edition, and it deliberately stays light. PostHog is a deep product platform with session replay, feature flags, experiments, and surveys, powerful but a heavy self-host stack of Kubernetes, ClickHouse, and Kafka. smolanalytics is a single static Go binary that does web and product analytics from one snippet, stores about 7 bytes per event with no external database, and lets you ask questions in plain English where the AI provably cannot invent the number. All four are real, useful tools. The rest of this post is about which fits which job.

which is best for what?

there is no single winner, so here is the plain version by job.

if you need free analytics that feeds Google Ads remarketing and audience building, GA4 is the default, because that loop is the whole point of the product and nothing else matches it for free. if you want a clean, privacy-first pageview picture with no cookie banner and almost no weight, Plausible is the tidy pick. if you are a product team that lives on session replay, feature flags, and experiments in one place, PostHog is the most complete platform of the four. and if you want one small self-hosted binary that does both web and product analytics, where you can ask questions in plain English and the AI cannot make up the figure, that is the case for smolanalytics.

the honest framing: these are not four versions of the same thing. GA4 is an ads-attached measurement tool, Plausible is lightweight web stats, PostHog is a product suite, and smolanalytics is a trust-first single binary. pick the shape, not the leaderboard. for the direct head-to-heads see smolanalytics vs GA4, smolanalytics vs Plausible, and smolanalytics vs PostHog.

how do they compare side by side?

only widely-known, verifiable characteristics here. numbers are the ones each vendor documents publicly.

a few notes so the table is not read as a scoreboard. GA4's raw event data is available with no thresholds and no sampling through its BigQuery export, so the thresholding complaints are about the in-product reports, not the underlying data. Plausible's Community Edition has no pageview cap because you pay only for your own server. PostHog's free tier is genuinely generous (documented as around 1M events a month for analytics) before usage pricing starts. smolanalytics deliberately does not do session replay, feature flags, experiments, heatmaps, or surveys, so PostHog wins outright the moment you need any of those.

can the AI make up the number?

this is the one line in the table that is genuinely different, so it is worth being precise rather than salesy.

GA4, Plausible, and PostHog are all excellent at storing and querying data, but none of them is built around asking questions in plain english from your editor with a guarantee about the answer. smolanalytics is. you can ask from a dashboard bar, or point your own Cursor or Claude at its MCP endpoint (47 tools, 13 prompts) and ask from where you already work. the important part is that the answers are computed from the same deterministic reports the dashboard renders, never generated by the model, and a CI agreement test fails the build if the AI's answer ever drifts from the dashboard's. because you bring your own model, the AI is free, and your questions go to your model rather than through a middle layer. see the MCP page for how that connects.

how do they self-host, and what does it cost?

GA4 is hosted only, so self-hosting is not on the table; the tradeoff you accept is that Google runs it and monetizes it through ads. Plausible self-hosts as the AGPL Community Edition, which in practice means running Elixir with Postgres and ClickHouse, and it is community supported with no vendor guarantee if something breaks. PostHog self-hosts too, and it is fully open source with no license cost, but a production deployment realistically wants a DevOps team comfortable with Kubernetes, ClickHouse, and Kafka. smolanalytics is a single MIT-licensed static Go binary with no external database: docker run -p 8080:8080 ghcr.io/arjun0606/smolanalytics and it is up in about 30 seconds, storing roughly 7 bytes per event in plain files. hosted plans run Solo $9, Pro $29, Scale $149, Business $499 per month with $5 per extra million events, against roughly $50 at the big tools, and self-host is free forever with a 14-day trial and no card. for the trade-first version see the smolanalytics docs.

which one should i pick?

honest guidance, no false modesty and no invented flaws.

pick GA4 if you want free analytics and you live in the Google Ads ecosystem. the audience and remarketing loop is the real reason to use it, and nothing else on this list replaces that for free. accept the in-product thresholding on low-traffic reports and the exploration sampling, or lean on the BigQuery export for raw data.

pick Plausible if you want a clean, privacy-first, cookieless picture of your web traffic with no consent banner and almost no weight, and you do not need deep product analytics. hosted from $9/mo or the free Community Edition if you are happy running Elixir, Postgres, and ClickHouse yourself.

pick PostHog if you are a product team that needs session replay, feature flags, experiments, and surveys in one platform, and you either take the generous free tier and usage pricing or have the ops muscle to run Kubernetes, ClickHouse, and Kafka. it is the most complete product of the four, and if you need those features it wins.

pick smolanalytics if you want one small binary that does both web and product analytics (visitors, referrers, funnels, retention, paths, cohorts), that you self-host free under MIT, that leads with a daily verdict on what to fix and a morning brief, and where you can ask questions in plain english and the AI provably cannot invent the figure. it is deliberately not a replay or experiments tool. it is the focused choice for "just tell me the true number when i ask, cheap, and let me own it."

the honest close

this is one person, and it launched recently. there are no testimonials on the site because there are none yet, and none will be invented. GA4, Plausible, and PostHog are all good software built by serious teams, and the differences above are about shape and design choices, not quality gaps.

smolanalytics is open source under MIT, a single Go binary with no Kafka, ClickHouse, or Postgres beside it. cookieless mode means no consent banner, and importers cover PostHog, Umami, CSV, and JSONL if you are switching. web install is a <script src=".../sdk.js"> tag plus smolanalytics.init(key, {host}). self-host free forever, or hosted from $9/mo with a 14-day trial and no card at smolanalytics.com, and the code is at github.com/Arjun0606/smolanalytics. ready to try it? start free, and self-hosting stays free forever.