GA4 attribution models: which one your reports are using
Three models survive, the thresholds people quote for data-driven belong to other products, and switching model rewrites your history. What to check and change.
Short answer: open Admin → Attribution settings and read what's there before you argue with anyone about credit. GA4 has three models left — data-driven, paid and organic last click, and Google paid channels last click — after "the first click, linear, time decay, and position-based attribution models" were withdrawn in November 2023 (Analytics Help, checked 5 August 2026). Two settings on that screen matter more than the model choice: the lookback window, which you can only shorten from its default, and the fact that changing the model "applies to historical and future data". Switch it on a Tuesday and last quarter's numbers move too.
Three models, and their names describe their scope
The list is short enough to hold in your head.
Data-driven attribution "distributes credit for the key event based on data for each key event" and "uses your account's data to calculate the actual contribution of each click interaction" (Analytics Help, checked 5 August 2026). It spreads credit across the touchpoints in the path rather than handing it to one.
Paid and organic last click "ignores direct traffic and attributes 100% of the key event value to the last channel that the customer clicked through (or engaged view through for YouTube) before converting". This is the old last-non-direct-click behaviour under a clearer name.
Google paid channels last click does the same thing but only across Google's own paid channels. Anything else in the path gets nothing.
Read those descriptions against the report you're about to send someone. A model that ignores direct traffic will never show you a direct conversion, no matter how many people typed your URL in. That isn't a data problem, and no amount of tag debugging will change it.
Google's current help pages describe the three models without stating which one a property starts on, so don't assume yours is on the one you'd have chosen. Check the setting. It takes ten seconds and it's the only way to know what the numbers in front of you mean.
Nobody publishes a data threshold for GA4's data-driven model
This is the question everyone asks — do I have enough conversions for data-driven? — and the answers in circulation are borrowed from other products.
The often-quoted "minimum conversions" figures come from two places, neither of them GA4:
- Google Ads. Google recommends "at least 200 conversions and 2,000 ad interactions in supported networks within a 30-day period" for its own data-driven model, while stating that all conversion actions are eligible regardless of volume (Google Ads Help, checked 5 August 2026). That's a recommendation for the Ads model, not a gate on the Analytics one.
- Universal Analytics, retired. The legacy multi-channel funnels documentation required "400 conversions per conversion type with a path length of 2+ interactions" and "10,000 paths in the selected reporting view" within 28 days (Analytics Help, legacy article, checked 5 August 2026). That product no longer exists.
We could find no published minimum-data requirement for data-driven attribution in a GA4 property. What Google does say is that the model needs converting and non-converting path data, and that "conversions can be reattributed for up to 7 days after the conversion". If someone quotes you a specific GA4 threshold, ask them for the page it's on.
The practical consequence is that on a low-volume property, data-driven attribution isn't wrong so much as thin. It has few paths to learn from, and the credit it distributes reflects a small sample. That doesn't make last click more accurate — it just makes it more legible.
The lookback window is the setting you can only shorten
Two defaults, and both are longer than most people guess:
| Key event type | Default window | What you can change it to |
|---|---|---|
Acquisition (first_open, first_visit) | 30 days | 7 days |
| All other key events | 90 days | 30 or 60 days |
Every option is shorter than the default. There's no way to look back further than 90 days, which quietly caps what GA4 can tell you about any sales cycle longer than a quarter — B2B, high-value services, anything with a committee involved. If your average deal takes four months to close, the first touch is outside the window and simply isn't in the model.
Shortening the window makes early touchpoints disappear and pushes credit towards the end of the path. That's a legitimate choice for a fast-moving ecommerce business. It's a bad one made accidentally.
Changing the model rewrites history; changing the window doesn't
The asymmetry here catches people, and it's documented plainly.
"Changing the reporting attribution model applies to historical and future data." Change it and every historical report changes with it. Yesterday's board pack no longer reconciles with the same report pulled today, and nothing in the interface announces why.
"Changes to the lookback window apply going forward." So a window change creates a seam in your data: sessions before the change were attributed under the old window, sessions after under the new one, and any period spanning the change is a mix of both.
Which gives you two rules worth writing down somewhere your team will find them:
- Annotate model changes. They're retroactive, so the only trace they leave is a number that stopped matching an old screenshot.
- Don't change either setting mid-reporting-period unless you're prepared to explain the seam to whoever reads the report.
Attribution decides how credit is split among the touchpoints you measured. It can't do anything about the ones you didn't — and consent is what determines which clicks get recorded in the first place. The Google Ads consent checker reports whether your site passes ad_user_data and ad_personalization to Google, which is what decides whether an ad click is usable in attribution at all.
What attribution settings will not explain
If your problem is that GA4 and Google Ads report different conversion counts, none of this fixes it, and changing the model to chase the Ads number makes things worse. The two platforms count different things on different dates by design — which is why the two conversion numbers never match, and what to check before assuming either is broken.
Attribution also can't repair a path with missing links. A click that arrived while your tags were blocked, a session that lost its campaign parameters, a conversion recorded without a source: those are gaps in collection, and every model will distribute credit around the hole without telling you the hole exists.
What to do this week
Read your attribution settings and write down what they say. Check the lookback window against your actual sales cycle, not against the default. If you change the model, annotate the date, and warn anyone comparing reports across it.
Then spend the rest of the effort upstream. Attribution models argue over credit for the interactions you captured — and if the capture is leaking, the argument is about the wrong thing. Making sure the click, the consent signal and the conversion all survive the trip is the work we do.
Sources
- 1.Analytics Help — [GA4] Select attribution settings (models, lookback windows, retroactivity) · Checked 2026-08-05
- 2.Analytics Help — [GA4] Get started with attribution · Checked 2026-08-05
- 3.Google Ads Help — About data-driven attribution (the 200 conversions / 2,000 interactions recommendation) · Checked 2026-08-05
- 4.Analytics Help — [UA] MCF Data-Driven Attribution model application and limits (legacy Universal Analytics thresholds) · Checked 2026-08-05
- 5.Analytics Help — [GA4] Key events · Checked 2026-08-05