Why do different analytics tools show different numbers?

Install two analytics tools on the same site and check them the next morning: the numbers won't match. A 10–30% gap is routine, bigger ones aren't rare, and the usual reaction — one of these must be broken — is almost always wrong. Each tool is counting correctly by its own definitions; the definitions just differ in half a dozen places at once.

The gap matters because sooner or later a decision hangs on it: an investor asks for traffic, an ad platform reports conversions your dashboard doesn't show, a client's GA4 disagrees with your report. Knowing where the differences come from turns an awkward discrepancy into an explainable one.

Nearly every tool-to-tool gap decomposes into some mix of six causes:

• Bot filtering. Some tools filter crawlers aggressively, others barely; server-side counters filter nothing. Bots are a double-digit share of raw requests, so this alone moves numbers substantially.
• Adblockers. Blocklists target some scripts more than others — Google Analytics most of all. In a tech-savvy audience, a quarter of visitors can be invisible to one tool and visible to another.
• Consent banners. Cookie-based tools lose everyone who declines — commonly 30–50% of EU visitors. Cookieless tools keep counting, banner-free. This is often the single biggest gap between GA4 and a privacy-first tool.
• Definitions. Session timeouts vary, GA4 redefined bounce rate entirely, and 'unique visitor' depends on how a tool recognizes a returning browser. Same behavior, different labels — most of it lives in the three-number hierarchy.
• Sampling and thresholding. GA4 samples large queries and hides small report rows for privacy, so its reports can omit data its own raw counts contain.
• Single-page apps. Client-side navigation only counts as pageviews if the tool tracks route changes. On the same SPA, a tool that does can show five times the pageviews of one that doesn't.

There is no true count to be closest to — 'a visitor' is partly a definition — but the errors do rank. Server logs sit furthest from human reality: they overcount, inflated by every bot on the internet. Cookie-based scripts behind a consent banner undercount, missing the decliners and the adblocked. A cookieless client-side script sits closest for counting humans: it dodges the consent gap, still loses some adblocked visits, and filters most bots by requiring a real browser to execute.

So: hosting dashboard highest, cookieless counter in the middle and most honest, consent-gated GA4 lowest. If your tools bracket a number in that order, nothing is broken — that's the expected shape.

Pick one tool as the source of truth and let it settle internal questions; which one matters less than the consistency. Read trends, not levels — the level is an artifact of definitions, but a trend is trustworthy within one tool. And never compare across tools: this week in tool A against last month in tool B manufactures insights out of methodology.

When someone else's number disagrees with yours — an ad platform, a client's GA — reconcile the definitions before doubting the data: whose bots, whose consent gap, whose session timeout. A number from SimilarWeb or a similar checker belongs to a different category entirely: that's a model, with error bars of its own.

Tool disagreement is structural: bots, adblockers, consent, definitions, sampling and SPA handling each move the count, and no two tools make the same six choices. Expect a 10–30% spread, and pick one source of truth to read direction from. The reconciliation energy is better spent on questions a traffic count can't answer anyway — like which sources produce paying customers.