Where attribution applies in GA4
GA4 carries traffic-source dimensions at three scopes, and only one of them is touched by the attribution settings. User-scoped dimensions, prefixed "First user", record the source that first brought a person and never change. Session-scoped dimensions, prefixed "Session", record the source that brought each session and never change either. Event-scoped dimensions, with no prefix at all (Source, Medium, Campaign, Default channel group), are the ones the reporting attribution model fills in, and they are only meaningful when paired with key events or revenue, because they describe which touchpoints earned credit for an outcome.
This explains a confusion that costs many teams an afternoon: switching the model and seeing the Traffic acquisition report unchanged. That report is session-scoped and was never going to move. The places that do move are the Advertising section (attribution paths, model comparison and the channel performance views), the key events report when an event-scoped dimension is added, and any exploration built on unprefixed source dimensions. The AttributionAssigning conversion credit to touchpoints. settings are therefore less about "which channel brought traffic" and more about "which channels get credit for the outcomes we care about".
Two documented behaviours shape every model. Direct visits are excluded from credit unless a path is made entirely of direct visits, so a person who clicked an ad and later typed the address gives credit to the ad. And GA4 can only credit touchpoints it recorded: landing sessions with recognisable source information. Impressions, views without a click, offline contacts and sessions where consent was declined are invisible to it, which is why the attribution hub treats platform attribution as one input rather than the answer.
The reporting attribution model setting
The setting lives in Admin under the data display settings, and it needs the editor role on the property. At the time of writing Google offers a short list of models: data-driven, which is the default; paid and organic channels last click, which hands all credit to the final non-direct touchpoint; and Google paid channels last click, which hands credit to the last Google Ads click in the path when there is one and otherwise behaves like the paid-and-organic rule. The older rule-based models (first click, linear, time decay and position-based) were retired from the setting, and properties that used them were moved to data-driven. Google documents the current list on the attribution settings help page, which is the place to check before describing the options to a stakeholder.
Data-driven attribution builds a model from the property's own paths, comparing the key-event rate of paths that include a touchpoint with similar paths that do not. The data-driven attribution guide explains the mechanism, the data it needs and what it cannot tell you. Last-click rules are transparent and easy to explain, and they remain useful as a continuity model when a team's reporting history was built on them.
Google states that a change to the model is applied to historical as well as future data in the reports that use it, because credit is calculated when the report runs. In practice that means the choice is reversible, but also that a report pulled before the change and one pulled after are not comparable. Note the date of any change in the property's change log and in the reporting documentation that the attribution models comparison recommends keeping.
Picking a model is a reporting decision, not a measurement one. Choose data-driven when the question is "which channels contribute to outcomes" and the property has a steady flow of key events. Choose Google paid channels last click when the main consumer of the report is the Google Ads team and the aim is a view that leans the same way their platform does. Choose paid and organic last click when a simple, explainable rule matters more than nuance, for example in a board report that has always used it.
Lookback windows
The lookback window is the period before a key event during which touchpoints are eligible for credit.
GA4 keeps two windows. One applies to the acquisition key events (first_open and first_visit) and
offers a short and a long option. The other applies to every other key event, offers three options, and
defaults to the longest. Google lists the exact day counts on the attribution settings help page; the
numbers are less important than understanding what moving them does.
A shorter window concentrates credit in the touchpoints close to the outcome. A longer window lets an early campaign share credit with the final click, which is usually what a team running awareness or content activity wants to see. If the sales cycle is long, the longest window is the safe default; if purchases are impulsive and paths are short, the window rarely matters and the shortest option simply trims noise. The window is applied per key event, so a newsletter signup and a purchase on the same property share one setting even if their consideration periods differ.
Google's guidance is that a change to the lookback window takes effect from the change onward rather than being recalculated for past data, which is different from the model setting; confirm the current behaviour on the help page, because this detail has changed over time. Either way, a window change is a break in comparability that belongs in the change log.
The lookback window is also one of the documented reasons GA4 and Google Ads disagree. Google Ads applies the conversion window set on each conversion action, measured from the ad interaction, and GA4 applies its own window measured backwards from the key event. The Google Ads versus GA4 numbers comparison walks through that gap cause by cause.
Channel groups and custom channel groups
A channel group is a set of rules that turns raw source, medium and campaign values into readable channel names. GA4 ships a default channel group whose rules Google documents and periodically updates. The rules match on the medium value, on whether the source is on Google's lists of search or social sites, on the campaign name, and on the source platform for Google's own products. The table summarises the pattern; the help centre holds the exact expressions.
| Channel | What the default rules look for | Where tagging usually goes wrong |
|---|---|---|
| Direct | Source is "(direct)" and medium is "(none)" or "(not set)" | Untagged email and app links, stripped parameters |
| Organic Search | Source on the search-site list and medium is organic | Nothing to fix; it is the baseline |
| Paid Search | Source on the search-site list and medium matches the paid pattern (cpc, ppc, paid and similar) | Auto-tagging disabled, or a manual medium outside the pattern |
| Paid Social | Source on the social-site list and medium matches the paid pattern | Medium set to social for paid posts, which lands in Organic Social |
| Organic Social | Source on the social-site list, or medium is a recognised social value | Agency links tagged with a custom medium |
| Medium is one of the recognised email values | Medium written as newsletter or the platform name | |
| Referral | Medium is referral | Payment gateways and own subdomains without cross-domain setup |
| Display | Medium is one of the recognised display values (display, banner, cpm and others) | Programmatic partners using their own vocabulary |
| Affiliates | Medium is affiliate | Network links tagged as referral or partner |
| Cross-network | Campaign name carries the cross-network marker Google adds to its multi-network campaigns | Expecting these sessions under Paid Search |
| Unassigned | Nothing matched | Any invented medium |
The consequence is that channel reporting is only as good as the UTMURL tags identifying traffic source. discipline behind it. A medium outside the documented vocabulary pushes sessions into Unassigned or Referral, and the attribution model then faithfully credits a channel that does not exist. Fix the naming convention before judging the model; the measurement section's naming guide covers the vocabulary.
Custom channel groups let you define your own rules when the default grouping does not match how the business thinks. They are created in Admin alongside the default group, built from ordered conditions on source, medium, campaign, source platform and similar fields, with the first matching rule winning. The default group cannot be edited, but it can be copied as the starting point for a custom one. A custom group can be selected as the primary dimension in standard reports, explorations and the attribution reports, so a team that separates brand from non-brand paid search, or splits partners from affiliates, can see attribution credit under its own labels. Google documents a limit on the number of custom groups per property and on channels per group; keep to one or two well-named groups rather than one per stakeholder.
Attribution reports: conversion paths and model comparison
The Advertising section holds the reports where the settings become visible. The attribution paths report (previously called conversion paths) shows the sequences of channels that preceded a chosen key event, with each path split into early, middle and late touchpoints and the credit each position received under the current model. Alongside the paths it reports path length and the time from first touch to key event, which are useful inputs when choosing a lookback window. Filter the report to one key event at a time; a blended view across signups and purchases describes nobody's journey.
The model comparison report sets two models side by side for a chosen key event and shows the key events and revenue each channel receives under each, with the change between them. It is the fastest way to see how much the choice of model moves credit, and in most properties the answer is "less than the argument suggested": a handful of channels gain or lose a share, and the ranking barely changes. Use it as a decision aid when picking the reporting model, and as a conversation tool when a channel owner disputes their credit.
To apply rules the reports no longer offer, such as first click or position-based, the tutorial on building an attribution model in SQL orders each user's sessions from the BigQuery export and applies the credit rules directly.
A furniture retailer compares two models
A furniture retailer with 1,400 monthly purchases runs the model comparison for the purchase key event
over one quarter, setting paid and organic last click against data-driven. Under last click, Paid Search
holds 46% of credited purchases, Email 14% and Paid Social 9%. Under data-driven, Paid Search falls to
39%, Email rises to 17% and Paid Social to 15%, while Organic Search is almost unchanged. The attribution
paths report explains the shift: 62% of purchases over £600 have a Paid Social touch in the early
position, typically 11 days before the purchase, and last click gave those touches nothing. The team
adopts data-driven for channel reviews, keeps the Google paid channels last click view in the shared
dashboard for the Google Ads team, and records the switch date in the reporting log.
Read both reports with their limits in mind. They only contain touchpoints GA4 recorded, so a channel that works through impressions rather than clicks will always look weak, and a channel whose landing links are untagged will not appear at all. No model, data-driven included, measures cause; a IncrementalityConversions caused by marketing. test does, and the attribution hub explains when one is worth running.
Interaction with Google Ads attribution
Google Ads keeps its own attribution. Each conversion action in Google Ads carries an attribution model and a conversion window of its own, and when GA4 key events are imported into Google Ads as conversion actions, Google Ads attributes them with those settings, using its own click identifiers, cross-device matching and engaged-view data that GA4 never sees. The GA4 reporting model does not travel with the export. One purchase can therefore be credited to Paid Search in Google Ads and to Email in GA4 without either system being wrong, which is the central point of the Google Ads versus GA4 comparison linked above.
Three practical consequences follow. First, set the Google Ads model per conversion action deliberately, because it feeds bidding, whereas the GA4 model feeds reports; the GA4 and Google Ads integration tutorial covers that step. Second, if the Google Ads team wants a GA4 view that leans their way, give them the Google paid channels last click model in a custom report rather than switching the property setting for everyone. Third, when the two numbers are compared in a meeting, compare key events of the same definition, over the same closed period, and explain the remaining gap by cause rather than chasing agreement. The GA4 hub lists the related setup pages, from key events to the Google Ads link, in the order they are usually tackled.
When a large share of sessions lands in Unassigned, the diagnosis page for GA4 unassigned traffic maps each source and medium pair to its tagging or collection fault.
Common mistakes
- Expecting acquisition reports to change. They are session-scoped and user-scoped; the model only moves event-scoped dimensions.
- Judging the model before fixing tagging. Unassigned and misplaced channels are a naming problem the model cannot repair.
- Changing settings without a log. Model and window changes break period-over-period comparisons; write the date down.
- Reading attribution paths as the whole journey. Impressions, offline contacts and consent-declined sessions are missing.
- Treating data-driven credit as proof of cause. It redistributes credit among recorded clicks; incrementality testing measures lift.
- Trying to reconcile GA4 with Google Ads to the unit. Different models, windows and inputs; explain the gap, do not force it closed.
- One custom channel group per stakeholder. Keep a small number of well-defined groups the whole team understands.
