Your GA4 Setup Is Quietly Lying to You — A Teardown

Almost every enterprise GA4 property I've audited reports numbers that are wrong in predictable, structural ways. Not catastrophically wrong. Quietly wrong. The kind of wrong that turns into the wrong budget decision six months later.

This essay is the argument. The GA4 Audit Hub → is the operating reference — the 8-point reality check, the FAQ, and the path to a measurement layer your finance partners will actually believe.

The problem isn't GA4 itself. The product is a serious, capable measurement platform. The problem is that the defaults are designed to make GA4 look full of data the moment you turn it on, and the data is full of modeling, attribution heuristics, and consent-mode estimation that almost nobody on the marketing team has actually inspected. Then the data flows into dashboards, the dashboards inform budget meetings, and budget meetings allocate millions of dollars based on a model nobody audited.

This is the teardown. Five places GA4 is structurally biased by default, what direction the bias runs, and the audit checklist to find out where you actually stand.

1. Data-driven attribution rewards what it can already see

GA4's default attribution is now data-driven, which sounds rigorous and is in fact much better than last-click. But "data-driven" doesn't mean unbiased. It means a model trained on the conversion paths GA4 can observe — which means the platform's view of attribution is shaped by which touchpoints fire trackable events.

Channels with strong, clean tracking — paid search, your own email, branded display — show up in the conversion path consistently and accumulate model weight. Channels that are harder to measure — connected TV, podcasts, organic social, anything view-through — show up sporadically or not at all, and the model learns they don't matter much. Not because they don't drive incremental outcomes, but because the model can't see them.

The bias direction is consistent: data-driven attribution over-credits the channels you can measure cleanly and under-credits the channels you can't. That's the opposite of what most marketing leaders assume the model is doing. It's also the structural reason Meridian MMM sits next to GA4 rather than getting replaced by it — and the reason that, as Google's reporting surface goes opaque in Q3 2026, owning an evidence layer outside the platform stops being optional.

2. Modeled conversions are estimates, not measurements

When GA4 can't observe a conversion path — because of consent rejection, cross-device sessions, ITP, or other tracking gaps — it fills in the gap with modeled conversions. These get reported alongside observed conversions in the same number, in the same dashboard, with no visual distinction. Most teams don't know which percentage of their reported conversions are observed versus modeled.

For some properties that ratio is fine. For others it's 30% or higher modeled, and the modeling is doing a lot of work to make the dataset look complete. The danger isn't that modeling is wrong. The danger is that nobody knows it's there, so the reported numbers feel like ground truth when they're partly inference.

You can check your modeled conversion ratio in GA4's reporting identity settings, and you should. Not because the answer needs to be zero, but because anyone making decisions on the numbers should know how much of the dataset is observation versus estimation.

3. Consent Mode v2 quietly reshapes the dataset

Consent Mode v2 is the system that lets GA4 and Google Ads keep functioning when users decline tracking. When a user rejects cookies, the platform sends pinged signals without identifiers, and Google's modeling fills in the missing behavior. Across European traffic this can affect 30 to 60 percent of sessions depending on the consent banner.

The problem isn't that Consent Mode is dishonest. It's transparent about what it does. The problem is that the modeled portion of the data behaves differently from the observed portion in ways that are invisible in standard reports. Conversion windows shift. Attribution shifts. Audience composition shifts. And teams comparing year-over-year performance often don't realize they're comparing two different datasets — pre-Consent-Mode and post-Consent-Mode — and attributing the difference to campaign performance rather than measurement methodology.

Three quiet symptoms

Conversions in your top channels look unusually stable while traffic fluctuates — modeled conversions are smoothing variance.

Geographic patterns shift suddenly when consent banner copy changes — the modeled fill rate is consent-dependent.

Year-over-year comparisons show patterns that don't match what your media plan actually changed — you're comparing different measurement regimes.

4. Default events fire on things that aren't conversions

GA4's default enhanced measurement turns on events for scrolls, outbound clicks, file downloads, video starts, and several other interactions automatically. These get counted as engagement, which feeds into engaged sessions, which feeds into conversion modeling, which feeds into attribution.

The result: a session where someone scrolled to the footer and clicked a LinkedIn icon registers as engaged. A session where someone read your homepage for forty seconds without scrolling registers as not engaged. The model treats these as comparable signals of interest. They aren't.

This isn't a fatal flaw, but it's a signal-quality problem that compounds. Engagement metrics inform audience definitions. Audience definitions inform Google Ads. Google Ads optimizes against signals it received from GA4. Bad inputs at the GA4 layer become bad audience signals at the activation layer, and the loop reinforces itself.

5. Cross-domain and cross-device patching is invisible

If your customer journey crosses domains — your marketing site, your storefront, your help center — or devices — phone research to laptop purchase — GA4 stitches sessions back together using its own logic. Some stitches succeed. Many fail. The ones that fail show up as new sessions, new users, and direct traffic, which then over-credits direct as a channel and dilutes the attribution of whatever channel actually started the journey.

The most common downstream symptom is a "Direct" channel that consistently shows the highest conversion rate in the property. That's almost always not direct traffic — it's broken cross-session stitching. And as long as Direct is winning the conversion rate ranking, every channel that actually deserved that credit is being undervalued.

The audit checklist

Why this is the hardest problem to fix

None of the issues in this teardown are GA4 bugs. They're defaults, designed reasonably for the median user, that quietly become liabilities at enterprise scale. The fix isn't replacing GA4. It's treating measurement as an ongoing engineering discipline — not a setup task, not an analyst's hobby, but a function with a budget, a roadmap, and an executive owner.

The organizations that win the next decade of marketing aren't the ones with the most sophisticated dashboards. They're the ones whose decisions are made on data the team can actually defend. That's the work. And it starts with knowing exactly how your current measurement is biased — so you can stop being surprised by the budget meetings where the numbers don't match the reality. If you want to see the financial value of compressing the latency between "the data shows we should change this" and "the change is live," the Cost Per Decision calculator puts a dollar figure on the cycle GA4 is supposed to feed.

The numbers in your dashboard tomorrow are the numbers from the GA4 setup you have today. If you haven't audited it, the numbers are telling you something — but it isn't necessarily what you think they're telling you.