Google Analytics 4 replaced Universal Analytics in 2023, and the migration forced a lot of teams to reckon with how much they relied on their analytics tool for measurement decisions they probably shouldn't have been making there in the first place.
GA4 is a genuinely better tool than UA for some things. The event-based data model is more flexible. BigQuery export (now free) is a significant upgrade for teams with the technical resources to use it. Cross-device measurement has improved, at least on paper. Session-based thinking has given way to user-based thinking, which is more accurate.
But GA4 also has meaningful limitations that are easy to overlook, especially if you're using it as your primary measurement system. Understanding them isn't about criticizing Google. It's about knowing what the tool can and can't do.
GA4 tracks on-site behavior accurately but misses the causal drivers of revenue
Data thresholds and sampling
GA4 applies thresholds to protect user privacy, if a dimension combination has too few users, the data gets withheld from standard reports. This is more aggressive than UA's sampling, and it shows up in places that are easy to miss.
Explore reports (the custom analysis tool in GA4) have different thresholds than standard reports, and those thresholds change based on data volume and the dimensions you're slicing. If you're building a report that segments by several dimensions at once, you may be seeing heavily thresholded data without a clear indicator of how much is missing.
The practical effect: segments of your audience that matter a lot for decision-making, specific geographic markets, users who came through specific channels, high-LTV customer cohorts, may show data that's been quietly withheld or aggregated in ways that distort the picture.
Modeled conversions (and the lack of transparency)
When GA4 can't observe a conversion directly, because of cookie consent refusals, ITP, or other tracking limitations, it models the likely conversion based on users it can observe. This is called "behavioral modeling."
Modeled conversions are not labeled differently from observed conversions in standard reports. They're blended in. The share of modeled versus observed conversions varies by market, traffic source, and time period, but there's no built-in way to see the split in most reporting interfaces.
In markets with high consent refusal rates (Germany, France, much of the EU), modeled conversions may make up a substantial fraction of what GA4 reports. In those markets especially, treating GA4 conversion numbers as observed counts rather than estimates is a mistake.
Attribution: last-click by default
GA4's default attribution model is still last-click. There's also a "data-driven attribution" model available, but it defaults to last-click for the channel reports most teams look at daily, and switching the model affects how data is retroactively reported in a way that's confusing.
More importantly, even GA4's data-driven attribution has the same fundamental constraints as all attribution models: it can only see digital touchpoints, it can't prove causation, and it can't see what happens outside the Google ecosystem.
If someone watched your YouTube ad on their TV, listened to your podcast sponsorship, and then three weeks later searched your brand and converted, GA4's data-driven model assigns credit to the last touchpoint it can observe. The YouTube ad and the podcast ad contributed nothing, in the model's accounting. This isn't a GA4 bug. It's a tracking reality.
No cross-channel view of paid social
GA4 shows you data from Google properties, search, display, YouTube, with reasonable accuracy. For paid social (Meta, TikTok, Pinterest, LinkedIn), it depends on click-through data from URL parameters. It doesn't see view-through conversions, which Meta in particular attributes significant value to.
This creates a systematically distorted picture of channel performance in any report that shows Google channels alongside paid social channels. Google channels will look better than they are relative to social channels, because their data is more complete.
What GA4 is actually good for
None of this means GA4 is a bad tool. It's the best free website analytics tool available and it does several things well.
For understanding on-site behavior, which pages people visit, where they drop off in your purchase flow, how different content performs across segments, GA4 is genuinely useful. The BigQuery export is excellent for teams who want to do custom analysis. The audience building and remarketing integration with Google Ads is powerful.
GA4 is also reasonable for tracking operational metrics: did the checkout flow work, are forms submitting correctly, is the site fast. These are questions that don't require causal measurement, they require observation.
What GA4 cannot do
GA4 cannot tell you which channels deserve more budget. It cannot prove that an advertising channel caused conversions. It cannot give you an accurate view of the full customer journey across all channels and devices. It cannot tell you what your conversion rates would be without advertising.
These aren't things GA4 is uniquely bad at. They're things that user-level analytics tools structurally cannot do, because they depend on complete tracking (which doesn't exist) and causal inference (which requires experimental design or statistical modeling, not pixel tracking).
The measurement decisions that matter most, budget allocation across channels, investment in awareness vs. performance advertising, incrementality of specific campaigns, require tools designed specifically for those questions: MMM, incrementality testing, and properly calibrated attribution.
GA4 is a useful operational tool. It's not a marketing measurement system.