10 GA4 Setup Mistakes That Corrupt Your Data (and How to Fix Them)

The uncomfortable truth about analytics: a broken setup does not look broken. The dashboards render, the graphs move, the numbers feel precise — and the decisions made on them inherit every silent defect underneath. After auditing GA4 properties across hundreds of businesses, the same ten defects account for most of the corruption. Check yours against the list; most properties fail at least three.

1. Duplicate tag firing

The classic: GA4 installed via a plugin and Tag Manager and a hardcoded snippet nobody remembers. Every page view counts twice, sessions inflate, engagement rates distort. Fix: pick one installation path (GTM, ideally), remove the rest, and verify single-firing in DebugView and Tag Assistant.

2. Key events unmarked — or everything marked

Some properties mark nothing as a key event, so "conversions" reads zero forever; others mark scrolls and outbound clicks alongside purchases, so the conversion number means nothing. Fix: mark only the events that represent business outcomes — purchase, lead, signup — and report engagement events separately.

3. Internal traffic counted

Your team visits the site daily and converts on test forms weekly. Untreated, staff and developer traffic pollutes every metric, brutally so for smaller sites. Fix: define internal IP rules, activate the exclusion filter (it ships in testing mode and many setups never flip it), and filter known bot referrers.

4. Cross-domain journeys split in half

Site on one domain, checkout or booking on another — without cross-domain configuration, the purchase session starts fresh and attributes to "referral: yourotherdomain.com". Revenue detaches from the marketing that earned it. Fix: configure cross-domain measurement across every domain in the journey, then confirm the linker parameter survives the hop.

5. Payment gateways stealing attribution

The special case of the same disease: buyers bounce to a payment provider and return, and the conversion credits the gateway as referrer. Ad platforms and channels all underreport. Fix: add every payment domain to the unwanted-referrals list.

6. Ecommerce values missing or malformed

Events fire but carry no value, or the items array is malformed, or currency is inconsistent — so revenue reports zeros and product performance is a mystery. Fix: implement the ecommerce schema properly through the data layer, validate items and values in DebugView, and confirm currency codes on every event.

7. Sessions fragmented by engagement settings

Default session and engagement timers rarely suit long-consideration or content-heavy sites; users reading for eleven minutes fragment into multiple sessions or count as un-engaged. Fix: tune session timeout and engaged-session thresholds to your actual behavior patterns, and document what you chose.

8. UTM chaos

Facebook, facebook, fb and meta as four different sources; campaigns named by whoever felt creative that week. Channel reporting dissolves into an unmergeable mess. Fix: a written UTM convention (lowercase, fixed vocabulary, campaign naming pattern), enforced with a shared link builder — and channel definitions checked so your sources group correctly.

9. Consent mode absent or misconfigured

In consent-regulated markets, tracking that ignores the banner is a legal problem; tracking that dies entirely on rejection is a data problem. Both are common. Fix: implement Consent Mode wired to your consent platform, so declines degrade gracefully into modeled aggregates instead of silence — defensible and still informative.

10. Data retention left at the default

GA4 ships with event-level retention set to two months. Fourteen months into the future, someone asks for a year-over-year exploration and discovers the past was deleted on schedule. Fix: set retention to fourteen months today — it applies forward only, which is exactly why it cannot wait.

Verify like you mean it

The pattern behind all ten: setups configured once, assumed forever. Analytics deserves the treatment code gets — a test transaction traced end to end, a weekly glance at anomaly-shaped movements, and a re-verification after every site change. An afternoon of fixes converts most corrupted properties into ones worth trusting. Then, finally, the dashboards mean what they appear to mean.