The 7 GA4 Settings Marketers Should Fix Before Trusting Any Report
A working GA4 setup for marketers comes down to about seven settings that the install never touches: data retention, internal traffic filters, key events, unwanted referrals, cross-domain measurement, reporting identity, and consent. GA4 ships with defaults that keep only 2 months of event-level data and count your own team as customers. Fix those before you trust a single report.
I'll be honest about where I land on this. From what I can tell, most "GA4 is broken" complaints are really "GA4 was installed and then nobody configured it" complaints. The tag fires, numbers show up, and everyone assumes the numbers mean something. They might. But until somebody has walked through the admin panel with a checklist, I wouldn't put a GA4 chart in front of a CFO.
This piece is the checklist, in the order I'd do it. It's written for the marketer who owns the property but isn't a full-time analyst, and it skips the stuff you can safely ignore for now (which is a lot of it, honestly).
The two defaults that quietly delete your history
Start with retention, because it's the only mistake on this list you can't undo later. Per Piotr Litwa's setup walkthrough, GA4 keeps user-level and event-level data for just 2 months by default. The fix takes thirty seconds: Admin, Data collection and modification, Data retention, set event data retention to 14 months.
The catch is what that setting actually governs. Analytics Mania explains that the limit applies to Explorations, the custom report builder, while standard reports keep their aggregated data much longer. And the change isn't retroactive. If your property has been on the default for a year, the data older than two months is simply gone from Explorations, and a full 14-month window takes about a year to rebuild.
So anyone who says "we'll do a year-over-year funnel analysis in Q4" while sitting on the default setting is going to have a bad Q4.
The second default is the internal traffic filter. You can define your office IPs as internal, which most people do, and then GA4 leaves the filter in Testing mode. In Testing, the filter tags the traffic but doesn't actually exclude it. Analytify's guide recommends a sensible workflow here: leave it in Testing for a few days, compare the filtered and unfiltered numbers to make sure your IP rules are right, then flip it to Active. Just don't forget the flip. A filter that's been "testing" since 2023 is, functionally, no filter.
Benchmark for this section: if a team of 10 to 20 people checks the site daily and your property gets under 5,000 sessions a month, internal traffic can plausibly be a few percent of everything. Run the comparison. If the gap between filtered and unfiltered sessions is over 3%, that filter matters more than you think.
Key events: collecting something isn't the same as counting it
This is the setting that generates the most confused Slack messages. GA4 collects a lot of events automatically. None of them count as conversions until you mark them as key events (GA4's renamed "conversions"). As Litwa puts it, collecting an event and counting it as a conversion are two separate things, and if you skip the second step, Google Ads sees zero conversions.
That last part is where it gets expensive. If Smart Bidding is importing from GA4 and there's no key event, it has nothing to optimize toward. You end up paying for clicks with the bidding equivalent of a blindfold on.
My rule: mark between two and five key events, no more. One primary (purchase, qualified lead, trial start) and a couple of secondary signals that actually predict the primary. Marking page_view as a key event is a surprisingly common mistake, and it turns every visit into a "conversion" and makes the conversion rate column useless. If your GA4 conversion rate is above roughly 20% on a site that isn't a single-purpose landing page, I'd check that first.
If you run ecommerce, check the numbers against your store backend before you trust them. Analyzify puts healthy alignment between GA4 and Shopify revenue in the 95% to 98% range. Below that, something is leaking. One merchant in the Shopify Community forums described missing 10-20% of orders every single week through the Google sales channel app, with no pattern by country, device, or payment method. Nobody in the thread found a confirmed fix, which, frankly, is the most realistic part of the whole story. The action is simple though: run one real test order a month and trace it all the way into GA4.
Related reading: GA4's built-in helper won't catch most of this for you. We covered why GA4's Task Assistant misses the silent misconfigurations that show up in most audited properties.
Your payment processor is not your best traffic source
Here's a fun one. A customer clicks your Meta ad, lands on your site, adds to cart, gets sent to PayPal or Stripe to pay, then returns to your thank-you page. Without configuration, GA4 can credit that purchase to the payment processor as a referral. Your best "channel" becomes paypal.com.
The fix lives under your web data stream, Configure tag settings, List unwanted referrals. Optizent's guide notes you can list up to 50 domains per web data stream, and that GA4 already excludes your own domain, your subdomains, and anything in your cross-domain list. For page-level exceptions there's also an ignore_referrer parameter you can set in gtag or Tag Manager.
Cross-domain measurement is the bigger sibling of this problem. If your marketing site and your app or checkout live on different domains, GA4 starts a new session when the user crosses over, splitting one journey into two and handing the conversion to... your own website, as a referral. Add every domain you own to the Configure your domains list.
Quick audit: open Traffic acquisition, switch to session source, and sort by key events. If any payment processor, booking tool, or your own secondary domain shows up in the top 10 sources, you have work to do. It's a five-minute check and it fixes channel reporting more than almost anything else on this list.
Reporting identity and the thresholding trap
This one is less obvious, and I think it's where a lot of the "GA4 is hiding my data" frustration really comes from.
GA4 offers three reporting identities. PMG's breakdown lays them out: Blended (the default) uses User-ID, then Google Signals, then device ID, then modeling. Observed does the same minus the modeling. Device-based uses only the device ID, which is the closest match to how Universal Analytics counted users.
The problem is Google Signals. When it's part of the identity mix and your user counts are low, GA4 applies data thresholding, meaning it withholds rows from reports to protect privacy. Analyzify's thresholding explainer lists the usual fixes: widen the date range, switch reporting identity to device-based, or turn off Google Signals (which won't fix past data). There's no published minimum user count, so you can't really predict when it'll kick in. You'll just notice a little warning icon and rows that should be there aren't.
PMG's recommendation is the one I'd follow: if you haven't implemented User-ID, switch to Device-based. You keep Google Signals for remarketing audiences, but your reports stop getting thresholded. If you have User-ID wired up and logged-in users are a big share of traffic, Blended starts to earn its keep.
And the nice part: changing reporting identity doesn't alter the underlying data. It's a lens, not a migration. You can switch it and switch it back.
Consent is the setting that decides how much you see at all
Everything above assumes GA4 is actually seeing your visitors. If you run a cookie banner, a chunk of them aren't. MeasureU estimates a cookie banner may be hiding 30% to 40% of visitors without any warning in the interface, and points out that GA4 sends no alert when tracking breaks after something like a site redesign.
For EU traffic, Litwa is blunt that Consent Mode v2 is not optional. The common failure isn't the missing banner. It's the banner that looks fine but isn't actually wired to the tag, so consent choices never reach GA4 and modeling never kicks in.
Benchmark: compare GA4 sessions to your server logs or CDN request counts for the same pages over the same week. My rough rule of thumb: a gap of 10% to 20% seems normal-ish given ad blockers. A gap of 40% or more usually means consent configuration, a duplicate tag, or a template that dropped the tag entirely.
This also connects to a broader blind spot. AI assistants increasingly strip referrers, so a growing slice of traffic shows up as direct no matter how clean your setup is. We dug into that in GA4's AI Assistants channel and the 70.6% it can't see.
The settings I'd ignore for the first month
On paper, GA4 has dozens of things to configure. In practice, most teams try to do all of them at once, give up halfway, and end up with a half-configured property that's worse than a lightly configured one.
Things that can wait: custom dimensions beyond the two or three you actually report on, audiences you don't plan to export, BigQuery linking (great, but only if someone will write SQL against it), and custom channel groupings. Do those in month two.
The one thing that can't wait is UTM discipline. MeasureU's point that untagged email, social, and PDF links fall into direct is probably the most common version of this problem: direct traffic becomes a catch-all junk drawer. If direct is over 30% of your sessions and you're not a well-known brand people type in, assume it's mostly untagged links and start there.
And if you're using GA4 to evaluate influencer or creator programs, the same attribution limits apply in a different costume. Our piece on why influencer promo codes miss most sales walks through what to do when last-click undercounts.
The 45-minute version
If you only have one sitting, do these in order:
- Data retention to 14 months. (2 minutes.)
- Internal traffic filter defined and switched to Active after a quick comparison. (10 minutes, plus a few days of testing.)
- Two to five key events marked; page_view is not one of them. (5 minutes.)
- Payment processors and booking tools added to unwanted referrals; your own domains added to cross-domain. (10 minutes.)
- Reporting identity set to Device-based unless you've implemented User-ID. (2 minutes.)
- Consent banner verified as actually wired to the tag, and one real test conversion traced end to end. (15 minutes.)
My prediction, and I'd be happy to be wrong: if you audit any random 10 small-business GA4 properties, at least 6 will still have retention on the 2-month default. It's the cheapest fix in analytics and the one nobody gets around to.
FAQ
Does changing GA4 data retention recover old data?
No. The change only applies going forward. Standard reports keep aggregated data longer, but Explorations will only show history from the point you made the change, up to the 14-month cap.
Should I turn off Google Signals?
Not necessarily. Switching reporting identity to Device-based removes most thresholding while keeping Signals available for remarketing audiences. Turning Signals off entirely is the bigger hammer, and it doesn't fix historical reports anyway.
How close should GA4 revenue be to my store backend?
For a well-configured ecommerce setup, roughly 95% to 98% alignment is a reasonable target. A consistent gap bigger than that usually points to a checkout flow, consent, or duplicate-tag issue worth chasing.
None of this makes GA4 pleasant. It just makes it honest, which is probably the more useful of the two.
By Notice Me Senpai Editorial
