ga4 says 912 users, plausible says 1,430 visitors, my database says 274 signups - which number do i plan with

Instrument your own funnel and stop caring about the top number. Write a row when someone hits the landing page from a cheap server-side event, a row when they start signup, a row when they finish, a row on first meaningful action. Now every step is measured by the same instrument, so the ratios between steps are trustworthy even when the absolute numbers are wrong.

Analytics tools are for 'where did they come from'. Your database is for 'what happened next'. Making one do both is where the frustration comes from.

Going to argue the conversion rate you're worrying about isn't a useful number at your size anyway. 274 signups from somewhere between 900 and 1,400 visitors is 20-30%, which for a landing page is either fantastic or means most of that traffic had already decided before arriving.

Either way the interesting question isn't the rate, it's which sources produce signups that are still there in 30 days. Split by source, look at week four retention. That number is worth arguing about. The top-of-funnel one isn't.

They're measuring different things and neither is wrong. One counts distinct people over a period after applying its own identity logic; the other counts visits, and a visit is not a person. Before reconciling anything, make sure you're comparing users to users and not users to sessions - that alone explains a big chunk of gaps this size.

The rest is blockers, consent banners, and one script loading before the other. You will never close it fully. Pick one and never switch, because your decisions depend on the trend, not the level.

Pick the pessimistic one. Plan with it. If you're wrong, you're pleasantly wrong.

One practical warning: whatever you pick, write down the date you switched. In six months you'll look at a chart with a step in it and spend an evening investigating a change you made yourself.

The thing that finally made this click for me was writing the actual definitions in a text file next to each number. What counts as a user, over what window, whether bots are filtered, whether it deduplicates across devices. An hour of reading docs.

Two of my three discrepancies turned out to be me comparing a 28 day window against a calendar month.