Data-Driven Attribution
The default GA4 model since 2023. Fractional credit assigned by machine learning across the conversion path.
Read the entry →Stan Consulting · Marketing Atlas · Reference · Attribution
GA4 Attribution.
Google Analytics 4's system for assigning credit to marketing channels for conversions. The default attribution surface most operators read first and trust most.
Section 02 · Quick definition
Definition.
GA4 attribution is the system inside Google Analytics 4 that assigns conversion credit to marketing channels and touchpoints across a buyer's path. It runs on event-based data, supports four attribution models, and uses data-driven attribution as the default. The reports surface credit by source, medium, campaign, and channel grouping, blended across web and app data streams. The attribution model choice is set per property and applies to acquisition reports, exploration paths, and the conversions surface. The system reports a story; the operator decides whether the story matches the bank.
Section 03 · Why it matters
Why it matters.
GA4 is the first attribution surface most operators reach for, and the one most CFOs see when they ask where the marketing money went. The reports are clean, the channels are named, and the credit is assigned to a decimal place. The combination produces a settled feeling. The settled feeling is the problem. The reports describe what GA4 saw, filtered through what consent mode allowed, modeled where data was missing, and credited according to whichever model the property was set to last quarter.
The metric matters because the GA4 read routes the budget. An operator who reads GA4 channel revenue as the source of truth funds the channels GA4 credits and starves the channels GA4 misses. The misses are systematic: branded search inheritance, organic from AI surfaces, dark social, last-touch on email, and any path that started in a UTM-stripped redirect.
The practical stake is that GA4 is the best operator-side attribution surface available and is still wrong about specific channels in specific ways. Reading it well requires knowing which ways.
Section 04 · How it works
How GA4 calculates and reports attribution.
GA4 records events with associated source, medium, campaign, and content parameters captured from the URL, the referrer, or a stored client identifier. Conversion events are attributed across the path of touchpoints leading to the conversion, with credit distributed according to the configured model. The default lookback window for paid and organic search is 90 days; for other channels it is 30 days. The model is applied consistently across the property, but reports also expose direct path-level data for inspection.
-
Step one · event capture
Every page view, click, and conversion event fires with attribution parameters. The parameters come from UTM tags on the inbound URL, the document referrer, or the previously-stored campaign identifier on the client. If none of those are present, the event is recorded as direct.
-
Step two · channel grouping
Source and medium pairs are mapped to channel groupings: organic search, paid search, organic social, paid social, email, referral, direct, and others. The default channel grouping is opinionated; custom channel groupings can be configured per property to match the operator's mental model.
-
Step three · attribution model applied
Data-driven attribution distributes fractional credit using a machine-learning model trained on the property's own conversion paths. Last-click, first-click, and position-based models are also available. The model choice changes which channels appear to drive revenue without changing which channels actually did.
-
Step four · consent mode and modeling
When users decline analytics consent, GA4 fills the gap with modeled conversions estimated from the conversions it can see. The modeling is most aggressive on Google ad surfaces. The reported number is partly observed and partly estimated, and the report does not visually distinguish which is which.
The four steps run continuously per property. Changing the attribution model setting changes the reported credit retroactively. The underlying buyer paths do not change.
Section 05 · Common misunderstandings
What people get wrong.
-
“GA4 says paid search drove the revenue, so paid search drove the revenue.”
GA4 says paid search received the credit the configured model assigned. A property set to last-click credits paid search for any path that ended on a paid-search click. The same path under data-driven attribution might credit organic, email, or direct just as much. The model choice is the headline; the headline is not the truth.
-
“GA4 numbers and Google Ads numbers should match.”
GA4 and Google Ads count different events under different attribution windows with different models. Ads counts platform-attributed conversions on a click-through window; GA4 counts session-based conversions on its own model. A 10–30% gap is normal. A larger gap usually means a tagging or import issue, not that one platform is lying.
-
“Direct traffic is the customers who typed our URL.”
Direct in GA4 is the bucket where attribution failed. It includes typed URLs but also UTM-stripped redirects, app deeplinks, dark-social shares, and any session that started without a referrer or campaign tag. A growing direct share is usually a tracking-loss signal, not a brand-strength signal.
-
“Modeled conversions in GA4 are reliable because Google built the model.”
Modeled conversions are estimates that scale up the observed conversions to fill in for users who declined consent. The estimate is structurally biased toward Google-owned channels because the model has more visibility there. The number is not a fabrication, but it is not measured either.
-
“Switching attribution models will fix our reporting.”
Switching models reshuffles credit; it does not produce new information about buyer paths. If the underlying event capture is broken or the UTM hygiene is poor, the report is wrong under every model. Fix the data first; then choose the model that matches the question being asked.
Section 06 · Diagnostic questions
Questions a Stan Consulting diagnostic asks.
Which attribution model is the GA4 property set to, and when was that setting last changed?
What share of sessions land in direct, and how has that share moved over the last 12 months?
What is the gap between GA4-reported revenue and Shopify-reported revenue for the same window, and where does the gap concentrate?
Is consent mode v2 active, and what share of conversions in the report are modeled rather than observed?
Is the default channel grouping in use, or has a custom channel grouping been configured to match the business?
What is the rate of UTM-tagged inbound traffic versus untagged, and which paid-channel campaigns are missing tags?
Are conversions imported into Google Ads from GA4 or fired natively in Ads, and is the same conversion counted twice?
Section 07 · Related Atlas entries
Adjacent reading.
Section 08 · Five Cents
GA4 is the most-trusted attribution surface in operator marketing and the most-misread one. The model is sophisticated. The operator's read of the model usually is not. I have sat with founders who quoted GA4 channel revenue as if it were a bank statement, and the bank statement was on the next browser tab disagreeing by 18%. The disagreement was not a bug. GA4 was answering a question about touchpoints. The bank was answering a question about money. The two questions never had to produce the same answer. The fix is not to switch models. The fix is to know which question each surface is answering and to stop pretending one of them is wrong.
Stan · Marketing AtlasSection 09 · Sources
Sources.
- Google Analytics Help · About attribution and attribution modeling Official documentation on how GA4 defines attribution, what models are supported, and how lookback windows are applied to paid and organic channels.
- Google Analytics Help · Data-driven attribution Reference for how the default model in GA4 distributes credit, what data volume is required for stable training, and how the model handles cookieless paths.
- Google Analytics Help · Default channel group Documentation on how source and medium pairs are mapped to channel groupings and how custom channel groupings are configured.
- Search Engine Land · GA4 attribution models explained Practitioner reference on the differences between GA4 attribution models and the operational implications of switching between them.
- CXL · Marketing attribution methods compared Independent practitioner reference comparing platform-side, MTA, and incrementality-based approaches to marketing attribution.