How to use this page
Start from the symptom, not the suspicion. Find the row in the diagnosis table that best matches what the report shows, run the check in the second column before changing anything, and apply the fix only when the check confirms the cause. Two or three causes often coexist, so keep going after the first finding; a duplicate tag and an unmarked key eventA GA4 event marked as important to the business, the property's measure of success. can arrive in the same week and the second one surfaces as a new mystery once the first is fixed.
Before the table, apply the sanity checks that save most investigations. Is the period closed? The most recent day or two are still being processed and will change. Is the metric the one you think it is? Users in most reports means active users, and engagement time is not session duration. Has anything been released? A site deploy, a Tag Manager publish, a consent banner update or a new plugin is behind a large share of sudden changes, and the debugging and preview tutorial is the fastest way to see what the container is actually sending today. Only when the number is real, the metric is right and the release log is clean does the rest of this page apply.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Views or event counts roughly double sessions; page views per session jumped | Tag Assistant on a live page: how many GA4 tags fire for one page view, and from which installations | Two installations (hard-coded tag plus Tag Manager, a platform plugin plus Tag Manager, or two configuration tags) | Remove one installation; keep the one the team controls and document it |
| Page views double only on single-page app routes | Enhanced measurement settings for the stream; the container for a manual page view tag | Enhanced measurement's history-change page views plus a manual page_view on route change | Turn off one of the two; never send both |
| Key events show zero or far fewer than expected | Admin key events list: is the event marked, and does the name match the sent name exactly | Event not marked as a key event, a case or spelling mismatch, or marking done after the period | Mark the exact event name; marking applies from that moment onward |
| A custom parameter shows "(not set)" or is missing from the dimension picker | Admin custom definitions: is the parameter registered, at which scope, and since when | Parameter never registered, or registered after the data was collected | Register it at the right scope; expect data from registration forward only |
| A custom dimension shows values that make no sense against session metrics | The scope chosen at registration against how the parameter is sent | User-level attribute registered at event scope, or the reverse | Archive and re-register at the correct scope; document the date |
| Own domain or a payment provider appears as a referral source | Traffic acquisition with Session source as the dimension, filtered to the own domain and gateway names | Cross-domain measurement not configured, or the gateway not listed as an unwanted referral | Configure domains in the stream's tag settings; add gateways to the unwanted referrals list |
| Direct traffic is unusually high; Session source is "(not set)" | Landing page report for untagged entry points; redirects that strip parameters; consent flow timing | Untagged email or app links, redirects dropping parameters, events firing before the page view | Tag every owned link, preserve query strings through redirects, fix event order |
| Campaign names are inconsistent or split across rows | Session campaign dimension for variants of one campaign | Manual tagging that does not follow the naming convention | Enforce the convention and a link builder; see the naming guide |
| Engagement and conversion rates look too good on some pages | Pages report with a comparison for internal network ranges or a developer traffic filter | Internal or test traffic not excluded; data filter still in testing state | Define internal traffic rules and set the filter to active |
| Revenue or transactions do not match the shop | Ecommerce purchases report against the order list for one closed day | Duplicate purchase on reload, test orders, inconsistent value or currency, missing refunds | Follow the e-commerce tutorial's purchase and refund steps |
| A large "(other)" row, or rows missing with a warning icon | The dimension's distinct values per day; the reporting identity setting | Cardinality limit exceeded, or data thresholds under a blended identity | Reduce cardinality or use explorations and the export; switch identity or widen the range |
| GA4 key events differ from Google Ads conversions | Compare the same key event over a closed week in both systems | Different attribution models, windows, dates and inputs | Explain the gap by cause; do not force agreement |
| Interface totals differ from the BigQuery export | A single day's event count in both places | Modelled data, estimated counts and thresholds exist only in the interface | Expect a small gap; treat the export as the exact record |
Double counting from two installations
The classic symptom is event count sitting close to twice the number of sessionsA group of one user's interactions with a site or app in one visit, ended by inactivity. for page_view, with
engagement rateShare of sessions that were engaged; on social, interactions relative to reach. oddly high because every session now has at least two views. The cause is two routes
sending the same events: a hard-coded Google tag left in the theme after Tag Manager was introduced, a
commerce or CMS plugin with its own GA4 integration running alongside the container, or two configuration
tags in one container, each firing on all pages. Confirm it in Tag Assistant on a live page, where the
tags that fired are listed with their source; a second installation is visible in seconds. The fix is to
remove one route, keeping the one the team controls and can debug, and to write down which one that is.
Google tolerates some repeated installations of the same Google tag, but a second configuration with its
own settings or a plugin with its own logic still sends twice, so confirm rather than assume. On
single-page applications the same symptom has a narrower cause: enhanced measurement's history-change
page views and a manual page view tag on route change, where one of the two should be switched off.
Key events missing or misnamed
A key event report at zero, or a conversion rate that collapsed on a known date, usually means the event
was never marked as a key event, was marked under a slightly different name, or was renamed in the
container without anyone updating the property. GA4 event names are case-sensitive and must match
exactly; generate_lead and Generate_Lead are two events. Marking is not retroactive, so an event marked
on the tenth of the month shows key events from the tenth, and the first week looks like a failure. Check
the events report for the exact names arriving, then the key events list in admin for what is marked. The
fix is to mark the exact name that arrives, fix the container if the name itself is wrong, and set the
counting method deliberately. Where the key event feeds Google Ads, re-check the imported action, because
a renamed event breaks the import silently. The GA4 hub links the key events guide
that sets the marking and counting rules out in full.
Unregistered parameters
Sending a parameter and seeing it in DebugView does not make it reportable. Until a parameter is registered as a custom dimension or metric, it is stored but invisible to standard reports and explorations, and when it is registered it fills from that moment onward, which is why a new dimension shows "(not set)" for the earlier dates. The events and parameters guide explains the registration step, the documented limits on how many dimensions a property can hold, and why the limit argues for registering what the reports need rather than everything the site sends. Check the custom definitions list in admin for the parameter's exact name and its registration date; a name mismatch between the container and the registration is the second most common cause. The fix is to register the exact parameter name at the intended scope and to annotate the date, so that nobody later reads the "(not set)" rows as a tracking failure.
Wrong scope on custom dimensions
A custom dimension registered at the wrong scope produces numbers that are valid and useless. A user attribute such as membership tier registered at event scope only applies to the events that carried it, so a session count by tier covers a fraction of sessions; an event attribute registered at user scope overwrites itself with the latest value and rewrites history for that user. The symptom is a dimension whose values look plausible but do not add up to the totals, or that change when the date range changes. Check the scope on the custom definitions page against how the site sends the value: user properties for things true of the person, event parameters for things true of the moment, item parameters for things true of a product line. In practice a wrong scope cannot be edited; the dimension is archived and a new one registered at the correct scope, with the date logged. The scope table in the reports and dimensions guide is the reference.
Self-referrals and cross-domain gaps
When the site's own domain, a subdomain with a separate checkout, or a payment provider appears as a referral source, sessions are restarting mid-journey and the acquisition credit is moving from the real source to the domain the person came back from. Conversion rate by channel then understates every paid channel and shows Referral converting absurdly well. Check Traffic acquisition with Session source as the dimension and look for the own domain, checkout subdomains and the names of payment gateways. Two settings fix most cases: cross-domain measurement, configured by listing the domains in the data streamA source of data flowing into a GA4 property: one web stream for a site or an app stream for each iOS or Android app, each with its own settings.'s tag settings so the session identifier is carried across, and the unwanted referrals list in the same place, which tells GA4 to ignore the gateway as a source. Both apply going forward. Where a third-party checkout cannot carry the identifier, the remaining option is to move the key event to a page the site controls or to send it server-side with the original source information.
Campaign data lost on the way in
Direct traffic that is too high, Session source showing "(not set)", and campaigns split across several spellings share one root: the UTMURL tags identifying traffic source. parameters either never reached GA4 or arrived in forms GA4 cannot group. Links in email, apps and offline materials that carry no parameters land as direct. Redirects that drop the query string, link shorteners and consent flows that fire events before the page view all produce "(not set)" sessions. Manual tags written freehand produce five rows for one campaign. Check the landing page report for high-volume entry pages with no campaign, test the owned links by clicking them, and watch the redirect chain in the browser's network panel. The fix is partly technical, preserving parameters through redirects, and partly organisational: a naming convention with a link builder that everyone uses, as the naming conventions and UTM guide sets out, and a monthly look at the campaign dimension for strays.
Internal and test traffic
Engagement rate and CVRShare of visits or clicks that convert. that look too good on a handful of pages, a spike of key events on the day of a release, or a cluster of sessions from one city on the admin pages are the signatures of internal and test traffic. GA4 handles it in two steps that are easy to leave half done: an internal traffic rule in the data stream settings tags matching sessions with a traffic type parameter, and a data filter in the property's data settings then excludes that traffic type. The filter is created in a testing state so its effect can be reviewed with a comparison, and it only excludes anything once it is set to active. Many properties sit in testing for months. Developer traffic from debug mode has its own filter with the same two-state behaviour. Check the filter states, confirm the rule matches the office and VPN ranges actually in use, and remember that filters apply from activation onward, so a reported spike before that date stays in the history and needs an annotation.
Misread metrics
A fair share of "wrong" numbers are right numbers read with the wrong definition. Users in most standard reports means active users, not everyone who sent an event; sessions do not restart at midnight or on a campaign change; average engagement time is foreground time per active user, not the old session duration; and the session key event rate and the user key event rate are different conversion rates that will never match a shop platform's checkout rate. Comparisons with the previous generation of Google Analytics, with an advertising platform, or with a dashboard built on different definitions therefore fail for reasons that are not faults. Check the metric name in the report and the definition in the help centre before searching for a tag problem, and when two systems disagree, reconcile a closed period cause by cause rather than hunting for the one setting that will make them agree. Attribution settings deserve their own mention: switching the reporting model changes event-scoped source dimensions and nothing else, which looks like a bug to anyone expecting the acquisition reports to move.
The sessions versus users comparison sets out when each count is the right denominator, which settles most arguments over two conversion rates for the same week.
The (other) row and thresholds
Two mechanisms make rows disappear. The "(other)" row appears when a dimension's distinct values in a day exceed the documented cardinalityThe number of distinct values a dimension or field holds; high-cardinality dimensions, such as page paths with IDs, can exceed report row limits. limit; the long tail is rolled into one row and follows the data into every report that uses the dimension. Page paths carrying identifiers, site search terms and custom dimensions holding unique values are the usual culprits, and the fix is to reduce cardinality at the source, to use an exploration, or to query the export. Data thresholds are different: under a reporting identity that includes Google signals or modelled data, GA4 withholds small rows that could identify people and marks the report with an icon. The mitigations are a wider date range, fewer demographic dimensions, or a device-based reporting identity, which is a view-time setting that changes nothing in storage. Neither mechanism is a tracking fault, and both are cheap to recognise once the icon and the row name are known.
Revenue that does not reconcile
E-commerce figures that disagree with the shop are a troubleshooting subject of their own, with a short
list of usual causes: a purchase event that fires again when the confirmation page is reloaded or
revisited from a payment provider; test orders reaching the production property; a value sent with tax on
one template and without on another; a currency code that does not match the shop's; refunds recorded in
the back office but never sent as refund events; and a second integration sending the same orders.
Reconcile one closed day order by order against the shop's list, which usually names the cause in
minutes, then follow the purchase and refund steps in the
e-commerce tracking tutorial to fix the implementation rather than
the symptom. A revenue figure that is wrong by a stable share month after month is nearly always a value
convention; one that is wrong by a varying amount is nearly always duplicates or refunds. The
troubleshooting page on GA4 revenue that does not match orders
works through each of these causes with a fix, a verification step and a monthly reconciliation.
A subscription box company's doubled traffic
A subscription box company reports that sessions rose 9% and page views rose 94% in a week with no campaign change, while engagement rate climbed from 51% to 83%. Tag Assistant on the homepage shows two GA4 tags firing for one page view: the Tag Manager configuration tag the marketing team owns, and a second Google tag inserted by a theme update that shipped on the Tuesday of that week. The developers remove the theme's tag; the next closed day shows 1.9 views per session instead of 3.7, and engagement rate returns to 52%. The team annotates the four affected days in the reporting log, adds a weekly Tag Assistant check to the release checklist, and asks the theme vendor to make the built-in tag an option rather than a default.
Prevention
Most of these mistakes are cheaper to prevent than to find, and prevention is mostly about seeing changes early and writing things down.
- Set up from a checklist. The GA4 setup checklist covers streams, enhanced measurement, cross-domain, unwanted referrals, internal traffic, key events and custom definitions in one pass, which is where most of the table's rows are prevented.
- Keep one installation and name its owner. Document which route sends GA4 events and remove the rest when a platform, plugin or theme changes.
- Be on the release list. Site deploys, container publishes and consent banner changes should trigger a Tag Assistant walk of the main journeys the same day.
- Register before you report. Parameters and key events exist in reports only from the date they are registered or marked; do it when the event is built, not when someone asks for the report.
- Enforce the naming convention with a link builder. Campaign hygiene is a process, not a fix.
- Activate filters, then verify them. A filter in testing excludes nothing; check its state and its effect with a comparison.
- Reconcile monthly. A closed-period comparison with the shop, the CRM and the ad platforms, as the data quality guide describes, turns most of this page's symptoms into a line in a log rather than an investigation.
- Annotate everything. Every fix applies from its date onward; the history keeps the mistake, and the log is how later readers know why a chart bends.
Common mistakes
- Changing settings before checking Tag Assistant. The container shows what is actually sent; the report only shows the consequences.
- Reading yesterday as final. Recent days are still being processed.
- Assuming fixes are retroactive. Nothing in GA4 reprocesses history; annotate instead.
- Trusting a filter in testing. It excludes nothing until it is active.
- Comparing GA4 with another system to the unit. Different definitions, windows and inputs; explain the gap by cause.
- Fixing one cause and stopping. Run the whole table; a second cause often hides behind the first.
- Blaming the platform. The data reflects what the site sent and how the report was read; the cause is in one of those two places.
- Leaving the container to one person. The GTMTag management system. container, the property settings and the measurement log need a named owner and a deputy, or the next release repeats the last mistake.
