GA4 (not set): Where the Traffic Source Goes Missing
A report arrives with a large block of sessions under (not set) and another block under direct. The client reads that as campaigns failing. Most of the time the campaigns are fine and the collection is not.
GA4 uses several placeholder values for data it never received or cannot place. They look alike in a table and each one has a different cause. Treating them as one problem is how a team spends a week on the wrong layer of the stack.
Four placeholders, four failures
(not set) means Analytics got no value at all for that dimension. Unassigned means a value arrived but no channel rule matched it, so the traffic falls outside every channel in the group. Direct with a medium of (none) means no campaign parameters and no usable referrer survived the journey. The (other) row comes from a row limit rather than from missing data.
Google also reports (data not available) in some tables. That value usually cannot be acted on, unlike (not set), where the cause is generally findable and fixable.
Session source and medium come from one event
Session source and medium are written when session_start fires. If session_start is missing, the session has nothing to attribute and the report shows (not set). The traffic then appears as Unassigned in the default channel group, because no channel rule matches a missing source.
The usual cause sits in GTM. The GA4 config tag is triggered on Page View instead of Initialization, so the session start never arrives. Changing the trigger type fixes new sessions.
Confirm it before touching the container. Open DebugView, load the page in a fresh tab, and watch for session_start in the stream. If page_view arrives without it, the trigger is the first suspect.
Consent mode can remove the session start
Two implementation errors produce exactly this symptom.
- Using the default command to set every consent value, including the update after the visitor interacts with the banner. The user_engagement event is then not sent, and the initial event carrying session_start is lost. Consent updates belong to the update command.
- Sending more events with denied consent after the visitor granted it. Those events are treated as a separate session, and that session has no session_start.
Both are worth checking on any site where the banner was added after the tags. In the UK the second one is also a compliance problem, because the consent state recorded in the tag no longer matches the choice the visitor made.
Landing page and content group
Landing page shows (not set) when a session has no page_view event at all. A session that opens with a custom event and never fires page_view has no landing page to report.
Content group shows (not set) when you combine it with automatically collected events such as session_start and first_visit. Those events do not accept the content_group parameter, so the combination has no value to show. An empty content_group string produces the same result.
Campaign and Google Ads dimensions
On the advertising side, four causes cover most cases:
- The Google Ads account and the Analytics property are not linked.
- Auto-tagging is off, so no gclid arrives on the click.
- The gclid is present on the final URL, but the click came from a Google Ads account that is not linked to this property.
- Destination URLs are manually tagged with UTM parameters that are incomplete or incorrect.
The last one catches people. A URL carrying utm_campaign but no utm_source or utm_medium is incomplete, and incomplete is the same as missing for attribution. It also happens on links you did not build: a partner, a newsletter sponsor or a social team can tag a URL badly and send the traffic into the same row.
Custom parameters need a day
A newly registered custom parameter reports (not set) for the first 24 hours. If it was registered this morning, wait before debugging the tag.
There is a second trap on session_start and first_visit. Those events keep the parameters from the first event collected, so a parameter that only appeared later in the session shows (not set) against them even though the tag is working.
Direct traffic that was never direct
Direct means the visitor arrived from a saved link or typed the URL. Real direct traffic exists. A sudden block of it on a paid campaign usually means the campaign data was lost between the click and the landing page.
- A redirect between the ad and the landing page drops the query string. Test the chain and compare the final URL with the one you paid for.
- The referrer never arrives. Email clients, mobile apps, PDFs and links that move from http to https often send none.
- A cross-domain journey runs without the linker, so the second domain starts a new session with no source.
- Two implementations run side by side, for example a hardcoded gtag call plus the GTM tag, or a CMS plugin sending its own hits. Google's guidance is explicit that a property should not run both a server-side and a standalone client-side implementation on the same page.
Once traffic is misfiled as direct, the channel report loses the campaign and hands it fewer conversions than it earned. The paid team then gets judged on numbers the tracking produced rather than on the media.
Ordering rules that keep the session intact
Google's tagging guidance puts the config command or Google tag first on the page, before any event, and asks for all page configuration to be in that first call. Custom events fired before config get batched with session_start, and the config can then change user and session identity for the rest of the page, which disconnects the page view and the events after it from the session start.
Three settings cause trouble when they are customised:
- A different cookie prefix per section or subdomain splits the client ID. Users cannot be tied across sessions and events fall out of sessions.
- A linker initialised late adopts the previous page's identity after some events have already gone out, and the short session before it is discarded.
- Manually setting client_id or session_id breaks the assumptions the tag and the processing layer make about sessions. Use user_id when you need a custom identity.
(other) is a row limit, not missing data
The (other) row appears when a table holds more rows than its limit. Analytics keeps the most common values and condenses the rest. Google's own example is a Pages and screens table with a 100,000 row limit against 150,000 unique pages, where the least common 50,000 end up under (other). You may see it written as other or as (other). Both mean the same thing.
Cardinality drives it. More than 500 unique values in a day makes a dimension high cardinality, and the 500 figure is guidance rather than a hard limit. Analytics applies cardinality control beyond 50,000 values. Limits also vary by property type, by report, and by query complexity, so a two-dimension report with a filter reaches the limit well before a single-dimension one.
The fix is to stop registering identifiers as custom dimensions. A per-user custom dimension guarantees the (other) row and breaks the reporting it was meant to improve. Send high-cardinality data as event parameters or user properties without registering a definition, and it stays reachable in BigQuery, audiences and segments without pressing on the property limits.
What the row does to the reports around it
A (not set) row is not a segment you can exclude and forget. It inflates the denominator of every rate in the table. Conversion rate by channel drops because the conversions are sitting in a row the chart never flags. Engagement rate by landing page drops because a session with no page view cannot be engaged.
Excluding the row makes the numbers look tidy and hides the defect. The honest version is to report the gap as a collection fault with an owner and a date, then let the fixed property show a clean month. Clients accept that when it comes with a cause. They do not accept a chart where the biggest row is called (not set) and nobody can explain it.
A triage order that finds the cause
- Break the affected sessions down by landing page, hostname, country and device. A single hostname or landing page narrows the fault to one template or one redirect.
- Open DebugView on the affected page and confirm that session_start arrives with its source and medium parameters.
- Check the config tag's trigger type in GTM.
- Run the redirect chain from the ad URL and compare the query string at the end of it.
- Search the page source for a second Google tag or a hardcoded measurement ID.
- Read the consent calls in order: default first, update after the choice, and no events with denied consent after a grant.
- Check the cross-domain path for the linker parameter on outbound links.
Each step takes minutes and eliminates a layer. The order matters less than finishing the list, because two faults commonly sit on the same site and fixing one leaves the row in place.
A report full of (not set) and nobody to chase it? Book a free tracking audit and we will trace each row to the tag, the redirect or the consent call behind it.
What changes after the fix
Nothing that already happened. (not set) rows are not rewritten when the cause is corrected, so the historical report stays wrong and the next one is right. Say that before you start, or the fix looks like it failed when last month's chart does not move.
One saved exploration catches it early: landing page, session source and medium, and hostname, filtered to sessions where the source is (not set) or direct. Run it weekly. A jump in that number is a collection fault, and it shows up there before anyone reads a campaign review.