GA4 attribution decides which channel gets credit when someone clicks an ad, reads a post, opens an email and then buys, and it quietly shapes every report used to move budget. Google has also changed it repeatedly: models were removed in 2023, per-conversion settings arrived in beta in January 2026, and conversion windows became editable in August. This guide is for marketers, founders and growth teams who rely on GA4 and want to know exactly what it offers today. It covers which three models exist, where to change the settings, how lookback windows behave, which report answers which question, how to read Model comparison without fooling yourself, and where GA4 attribution stops being enough.
What Attribution Means Inside GA4
GA4 credits traffic in three different scopes, and mixing them up causes most "the numbers do not match" complaints.
- User-scoped dimensions, such as First user source, credit the channel that first brought the person.
- Session-scoped dimensions, such as Session source / medium in the Traffic acquisition report, credit the channel of that single visit.
- Event-scoped credit assigns a key event (GA4's term for a conversion) to a channel using the reporting attribution model.
The model setting mainly affects that third kind: the Advertising workspace and event-scoped source reporting. Changing it will not rewrite what the user-acquisition and traffic-acquisition reports show, which is why one purchase can appear under different channels in different reports without anything being broken.
The Three GA4 Attribution Models
| Model | How credit is assigned | Good for | Watch out for |
|---|---|---|---|
| Data-driven (default) | An algorithm spreads credit across touchpoints, based on data specific to the property and each key event | Journeys with several channels | Opaque: the weights are not disclosed |
| Paid and organic last click | 100% to the last non-direct click, paid or organic | A simple, auditable baseline | Over-credits the final channel, often brand search or email |
| Google paid channels last click | 100% to the last Google Ads click | Checking against Google Ads reporting | Ignores every non-Google channel |
Everything else is gone. Google's own help page states that the first-click, linear, time-decay and position-based models have not been available since November 2023. A rule-based view now has to come from outside GA4: exported path data, a spreadsheet model or a dedicated attribution tool. The Trackstack breakdown of marketing attribution models explains what each of those retired models did and when it fit.
Data-Driven Attribution in GA4: What It Does and Does Not Tell You
GA4 describes data-driven models as specific to each advertiser and each key event, built from the touchpoints that preceded conversions. In practice a purchase and a lead form on the same property can distribute credit differently. The model also reacts to the property's own mix: when a channel appears early in many converting paths, it gains credit that last click would have denied it.
Two cautions matter in daily use. First, GA4 can reattribute conversions for up to seven days after they occur, so channel numbers for the last week are provisional. Judging yesterday's paid-social result against last week's is a common way to make a wrong call. Second, the algorithm is a black box. It can show that a channel assists, not why, and it only sees tagged visits that reach the site. An ad that was seen but never clicked is invisible to it.
Lookback Windows: Matching the Window to Your Sales Cycle
A lookback window is how far back GA4 searches for touchpoints before a key event. By default, acquisition events use 30 days and other key events use 90 days; at property level the longer choices are typically 30, 60 or 90 days. The right value follows the sales cycle. A $30 impulse product rarely needs 90 days, and a B2B demo request can easily span two months of research.
Windows became more flexible on August 11, 2026. According to Google's release notes, click-through conversion windows can now be set to any whole number of days from 1 to 90, where earlier they were limited to preset values. Engaged-view conversion windows, previously fixed at three days, can be set from 1 to 30 days. These settings live in the Advertising workspace under conversion management.
One practical rule: treat every window change as a break in the trend line and write the date down. A shorter window lowers reported conversions for slow channels without any change in real performance, and a report that straddles the change will show a decline or a jump that never happened.
Per-Key-Event Attribution Settings
Since January 16, 2026, Google has been rolling out beta settings that let the attribution configuration be adjusted independently for every conversion, and since September 29, 2026 the same applies to app conversions. Google flags both as possibly unavailable for some properties, so conversion management in the Advertising workspace is the place to check before planning around them.
The value is straightforward. A fast checkout purchase and a slow enterprise lead no longer have to share one window and one model. A short window fits the purchase, a long one fits the qualified-lead key event, and the property-wide setting stays as the fallback. The cost is complexity: once settings differ per event, comparing reports across key events needs a note on which setting each one uses.
A Sensible Default Configuration
For most small and mid-sized properties, a short list of defaults avoids nearly all the trouble:
- Keep data-driven as the reporting model, because it is the only option that credits earlier touches.
- Choose a lookback window that covers the typical gap between first visit and purchase, which Conversion paths shows as average days to conversion, rather than a number copied from a blog post.
- Keep paid and organic last click in view as the comparison baseline.
- Keep a simple change log with the date of every attribution, window or key-event edit, so that a jump in a chart can be traced to a setting instead of a campaign.
If the property has only a few dozen conversions a month, the models will look alike, and simpler evidence works better, such as comparing periods with a channel on and off.
GA4 Attribution Reports: Where to Find What
| Report | Question it answers | Watch out for |
|---|---|---|
| Advertising snapshot | How are conversions split across channels at a glance? | High-level only; no path detail |
| All channels (Performance) | Which channels get credit for each key event under the reporting model? | Changes when the model setting changes |
| Conversion paths | Which sequences of channels precede conversion, and how many days and touches do they span? | Only as good as the source and medium tagging |
| Model comparison | Where do two models disagree about a channel? | Identical results can mean thin data or single-touch journeys |
| Conversion attribution analysis (beta) | Which channels assisted, and at which stage of the funnel? | Beta since January 16, 2026; may not be available for every property |
The newest of these reports has two views. Assisted conversions is based on last click and surfaces channels that appeared earlier in converting paths. Refined funnel analysis is based on data-driven attribution. Because the two views use different models, differences between them are expected, not a bug.
How to Read Model Comparison Without Fooling Yourself
Model comparison is the one report that turns the model choice from a setting into evidence. The table below is an illustrative example for one key event with 200 conversions, built for this guide rather than taken from a real property.
| Channel | Data-driven | Paid and organic last click | Difference |
|---|---|---|---|
| Paid Search | 61 | 66 | −5 |
| Organic Search | 48 | 52 | −4 |
| Paid Social | 33 | 14 | +19 |
| 28 | 31 | −3 | |
| Direct | 20 | 25 | −5 |
| Referral | 10 | 12 | −2 |
Four habits keep the reading honest:
- Look for the outlier, not the total. Both columns add up to 200. Only Paid Social moves materially, up from 14 to 33, which marks it as an assisting channel that last click starves.
- Treat a big gap as a hypothesis. It justifies protecting that budget while a holdout test confirms the effect; it does not justify scaling spend on its own.
- Be suspicious of identical columns. If data-driven and last click match for a key event, either most journeys are single-touch or the model has little to learn from. It does not mean the two models agree on a deep truth.
- Use enough time. Compare full weeks, leave out the last seven days because of reattribution, and never compare ranges that straddle a settings change.
Why GA4 Numbers Do Not Match Google Ads or Meta
Mismatched totals are the most common attribution complaint, and they usually have ordinary causes rather than errors:
- Different models and windows. Each platform counts inside its own rules, so totals cannot be expected to agree.
- Different dates. GA4 reports a conversion on the day it happened, while Google Ads by default attaches it to the day of the ad interaction.
- Different definitions. An imported key event and a native Google Ads conversion tag may not fire or count identically.
- Consent and blockers. Each tool models missing data in its own way.
- Reattribution. GA4 can still move credit for up to seven days.
For Meta the gap is larger, because Meta credits view-through and engage-through interactions that never create a GA4 session. Platform numbers are best used to optimize inside that platform, with a neutral source such as the CRM or store backend judging the total.
Where GA4 Attribution Stops
GA4 is excellent at what it can see and silent about the rest. It sees visits that reach the site, with source and medium tags, from consenting users. It does not see ad impressions, offline conversions such as calls and demos, or touches that happened on another device without a login. It also ships with only Google's cost data, so spend from Meta, TikTok or LinkedIn needs importing before a true ROAS comparison is possible.
Tagging discipline decides how much of that visible traffic is usable. Untagged campaign links land in Direct or Referral, where no model can rescue them, and inconsistent UTM spellings split one campaign into several rows. A short naming convention for source, medium and campaign, applied to every paid link and every email, does more for attribution quality than any switch in the settings. Auditing a month of Conversion paths for unexpected Direct or "(not set)" values is a fast way to find the leaks.
That is the point at which many teams outgrow GA4 alone. Once a budget is split across Google Ads, Meta, TikTok, LinkedIn and a CRM, the question becomes how to join spend, clicks and sales outcomes into one dataset and compare several models side by side. Better tagging helps first; a cross-channel analytics layer is what comes after.
A Setup Checklist Before Trusting Any Report
Attribution can only distribute credit for events that are measured properly. The foundations come first:
- Fire conversions once and correctly, using the Google Ads conversion tracking setup guide.
- For stores, send revenue-bearing events through the GA4 ecommerce events setup guide.
- For lead generation, track form submissions in GA4 without code and mark the result as a key event.
- Keep campaign context available with GA4 custom dimensions.
- Compare short and long journeys using useful GA4 audiences and segments.
- Bring sales outcomes in by connecting Google Sheets to the CRM.
FAQ
What attribution models does GA4 have?
Three: data-driven, paid and organic last click, and Google paid channels last click. First-click, linear, time-decay and position-based were removed in November 2023, so rule-based comparisons need an external tool or an exported dataset.
What is the default attribution model in GA4?
Data-driven attribution is the default reporting attribution model. Paid and organic last click remains available as a simpler baseline, and it is worth keeping an eye on in Model comparison.
How do I change the attribution model in GA4?
Open Admin, go to Data display, choose Events and then Attribution settings. There the reporting attribution model, the channels that can receive credit and the lookback window can be reviewed, then saved. The Marketer role or higher is required on the property.
What is the GA4 lookback window?
It is how far back GA4 searches for touchpoints before a key event. The defaults are 30 days for acquisition events and 90 days for other key events. Since August 2026, click-through windows can be set from 1 to 90 days and engaged-view windows from 1 to 30 days in conversion management.
Why do GA4 and Google Ads show different conversions?
The two tools use different models, windows, counting dates and conversion definitions, and GA4 can still reattribute credit for up to seven days. Small differences are normal. A large persistent gap usually points to a tagging or import problem.
Can GA4 attribute Meta and TikTok campaigns?
Only partly. GA4 can credit visits that arrive with campaign tags, but it never sees impressions, and view-through conversions stay on the ad platform. For cross-channel decisions, platform data combined with a neutral source and several models gives a more honest picture.
TL;DR
- GA4 offers three reporting attribution models in 2026: data-driven (default), paid and organic last click, and Google paid channels last click.
- First-click, linear, time-decay and position-based have been gone since November 2023.
- Conversion windows became editable on August 11, 2026, and per-conversion settings are in beta.
- Model comparison is the report that turns the model choice into evidence; look for outliers, not totals.
- GA4 only sees tagged visits from consenting users, so cross-channel budget decisions need data from outside GA4.
Originally published on Trackstack Editorial.
Top comments (0)