Choosing a web analytics tool in 2026

What a web analytics tool needs to do in 2026

The core job hasn't changed: count visitors, show where they came from, show what they did, and tell you whether the things you care about (signups, purchases, contact forms) are happening. Traffic, sources, top pages, goals. Any tool on the market does this.

What has changed is the environment around that job:

• AI-referral traffic is real now. Visitors arrive from ChatGPT, Claude, Perplexity, and Gemini citations — for content sites, often a double-digit share and growing. Most analytics tools bury this under generic "referral," so you can't see your fastest-growing channel. A 2026-era tool should classify AI assistants as a source category, like organic search or social.
• Page weight counts against you. The GA4 stack (gtag + Tag Manager + consent platform) adds ~100KB+ of JavaScript to every page. Modern trackers are ~1KB. On Core Web Vitals and mobile conversion, that difference is measurable.
• Privacy is a design decision, not a checkbox. If the tool sets cookies or fingerprints devices, you owe EU visitors a consent banner — and consent declines silently delete 30–60% of your EU data. Cookieless tools skip the banner and count everyone.
• One site is the exception. Operators, agencies, and indie builders run fleets. Per-site accounts and per-site pricing punish exactly the people shipping the most.

How to choose a web analytics tool

Six questions that separate the field. Ask them in this order:

• Script weight. Under 2KB or over 50KB — there's not much in between. Check the actual network tab, not the marketing page.
• Cookieless or consent-banner? If it stores anything on the visitor's device, you need a banner for EU traffic and you lose the visitors who decline. If it's cookieless (daily-rotating salted hash, no device storage), no banner — but no cross-day visitor identity either. Know which trade you're making.
• Data ownership. Can you export everything? Is your data used to train ad products (GA4's business model) or is the vendor paid by you directly? Self-hosting is the maximal answer, if you'll actually maintain the server.
• Multi-site support. How many sites per account, what does site #10 cost, and is there a view that compares them? If you run one site forever, skip this question.
• API and AI access. Can a script — or an AI agent — read your numbers? A REST API is table stakes; an MCP server means Claude, Cursor, or Codex can install and query the tool without you writing integration code.
• Price ceiling. Free tiers are marketing; what matters is the bill at 10x your current traffic. Check the per-event or per-pageview pricing curve and whether extra sites multiply it.

Web analytics tool comparison: the field, honestly

Every tool below is good at its actual job. The mistake is picking one whose job isn't yours.

• Google Analytics 4 — free at nearly any volume, and unbeatable for paid-ads attribution: Google Ads integration, audience modeling, BigQuery export. The costs: a ~100KB script stack, a consent banner in the EU, sampling and delays at scale, and a reporting UI most people need training to use. Right for: marketing teams spending real money on Google Ads.
• Plausible and Fathom — the simple, privacy-first hosted tools that proved the category. Tiny scripts, cookieless, one clean dashboard, EU-friendly by design. No free tier — Plausible starts at $9/month, Fathom at $15 — and API access is read-only reporting. Right for: a business with one or a few sites that wants a no-drama GA replacement and is happy to pay for it.
• Umami and Matomo — the self-hosted route. Free software, your server, your database, complete data ownership. Matomo is the full-featured (heavier) one; Umami is the lightweight one. You take on hosting, updates, backups, and scaling. Right for: teams with real data-sovereignty requirements or an ops person who enjoys this.
• Gizmo Analytics — cookieless like Plausible/Fathom (~1KB script, daily-rotating salted hash, no banner), plus the parts the others treat as extras: AI assistants as a tracked source category, unlimited sites with a fleet view, and an MCP server so AI coding agents can install it and query it — "add analytics to this site" is one prompt in Cursor or Claude. Free forever for 10k events/month. We don't do ads attribution or audience modeling, and we're not self-hostable. Right for: operators shipping multiple sites with AI tools who want simple numbers without banners or maintenance.

Simple beats complete

The dirty secret of analytics: most operators use about six reports. Visitors over time, sources, top pages, countries, devices, and goal conversions. That's the whole daily loop. Everything else — attribution model comparisons, audience segments, custom dimensions — is capability you pay for in complexity whether or not you ever open it.

The tool you actually check every morning beats the one you dread opening. If "look at analytics" means clicking one bookmark and seeing everything on one screen — or asking your AI assistant "how did the launch do?" — you'll stay close to your numbers. If it means navigating Explorations, you'll check monthly, then quarterly, then never. Pick for the habit, not the feature list.

FAQ

What is the best free web analytics tool?

Depends on what 'free' needs to include. GA4 is free at almost any volume but costs you a consent banner, a heavy script, and hours of learning Explorations. Umami and Matomo are free if you self-host — you pay in server bills and update maintenance instead. Cloudflare Web Analytics is free and simple but has no custom events or goals. Gizmo is free forever for 10k events/month with unlimited sites, custom events, funnels, and MCP access — the same product as paid, just capped. For most small-to-medium sites that want zero maintenance and no banner, a free hosted cookieless tier beats both GA4 and self-hosting.

Do I need Google Analytics, or is a simpler tool enough?

Ask one question: do you run paid ads at a scale where multi-touch attribution decides budget? If yes, keep GA4 — its Google Ads integration and attribution modeling are genuinely best in class, and no lightweight tool replicates them. If no — you run content, organic, product-led, or client sites — a simpler tool covers everything you actually look at: traffic, sources, top pages, goals, funnels. Many teams run both: a simple cookieless tool as the daily dashboard, GA4 quietly in the background for ads.

What's the lightest web analytics tool?

The cookieless generation all land around 1–2KB: Gizmo (~1KB), Fathom, Plausible (<1KB), and Umami are all in the same class — effectively invisible in Lighthouse. GA4 via gtag.js pulls roughly 100KB of JavaScript before you add Tag Manager and a consent management platform on top. If page weight is a ranking or conversion concern, any tool in the ~1KB class is fine; the meaningful line is between that class and the GA4 stack, not within it.

Can a web analytics tool track visitors coming from ChatGPT and other AI assistants?

Only if it's built to. AI assistants send referrer headers (chat.openai.com, claude.ai, perplexity.ai, and others), but most analytics tools lump these under generic 'referral' traffic — so you can't see whether that spike came from a Perplexity citation or a random blog link. Gizmo classifies AI assistants as a first-class source category, next to organic search and social. If you publish content in 2026, a growing share of your traffic arrives this way, and a tool that can't attribute it is leaving your fastest-growing channel unlabeled.

Cookieless or consent banner — which should I pick?

Cookieless, unless you specifically need cross-session user identity (remarketing, multi-week attribution). A cookieless tool derives visitor IDs server-side from a daily-rotating salted hash of IP + User-Agent — nothing is stored on the device, so no GDPR/ePrivacy consent is required and no banner is needed. Banner-based tools lose 30–60% of EU traffic to consent declines, meaning your 'complete' data is actually missing a third of your visitors. Cookieless data is slightly fuzzier across days but includes everyone.

Do I need a separate analytics account for each website?

With GA4, effectively yes — one property per domain, each with its own setup, consent config, and reports. If you run more than two or three sites, look for a tool with real multi-site support: unlimited sites in one workspace, a fleet view that compares them side by side, and one script pattern for all of them. Gizmo, Plausible, and Umami all handle multiple sites in one account; Gizmo doesn't charge per site and adds a cross-site fleet view plus site tagging.

What does 'MCP-first' analytics mean, and does it matter?

MCP (Model Context Protocol) is the standard AI assistants like Claude, Cursor, and Codex use to operate tools. An MCP-first analytics tool exposes its whole surface — install, queries, goals, funnels, anomaly checks — as MCP tools, so your AI agent can say 'add analytics to this site' and actually do it, or answer 'which of my sites grew last week?' without you opening a dashboard or writing API glue. If you don't code with AI assistants, it doesn't matter yet. If you do, it collapses analytics setup and reporting into one prompt.

Keep reading

• Google Analytics alternativeWhy operators are switching, and what the switch costs you.
• Plausible alternativeGizmo vs Plausible, feature by feature.
• Website analytics dashboardOne dashboard for every site you run.
• Free Google Analytics alternativeThe free tiers compared — what you get without paying.