How do I read the GA4 attribution models comparison?

Read it one channel at a time. Usually you set data-driven against paid and organic last click, pick purchase, and read the % Change column. A channel that gains under data-driven tends to open or assist journeys; one that loses tends to close them. The total never changes, so every gain is another channel's loss.

The report is GA4's Attribution models report, under Advertising, then Attribution. Google's help page calls it the key event attribution models report. It sets two models side by side for each channel, with key events, revenue and the % change between them.

What one store's data shows

One store's anonymised GA4 export, 1 January 2024 to 21 August 2026. It holds shares of revenue only: no ad spend, no order counts.

A model comparison can only move credit inside journeys that had more than one touch. On the Journeys sheet, one-touch journeys hold 79.5% of revenue, and those buyers took 0.5 days. Every model gives a one-touch sale to that one touch.

That leaves 20.6% of revenue on journeys of two or more touches, spread over 3,656 distinct paths (Journeys sheet). In this one store, that is the ceiling. No switch of model could move more than about a fifth of revenue between channels.

Direct holds 57.7% of revenue under last click, first click and touched alike (Channels sheet). Three different rules, and its share did not move. GA4's own models behave the same way: Google says every model leaves direct visits out unless the whole path was direct.

What the export cannot show is which model is right. It holds shares of credited revenue, not the sales each channel caused.

Why does the % Change column mislead?

It is a percentage of the channel, not of your revenue. A small channel can jump a long way and move little money. A big one can slip a little and move a lot. For illustration, +80% on a channel credited €5,000 moves €4,000. A -5% on a channel credited €100,000 moves €5,000 the other way, for illustration.

Credit moves; it is not made. Data-driven attribution can split one sale across several touches, and Google's example shows the shares adding up to 1.0. Every key event is handed out in full under either model. So each gain in the table is paid for by a loss elsewhere in it.

All key events are added together by default. If you mark sign-ups as key events, they sit in the same columns as purchases. Pick purchase in the drop-down at the top left before you read a single row.

A gain is a position, not a verdict. A channel that gains usually sits before the last click on paths that end in a sale. A channel that loses usually closes paths that others opened. Neither tells you which one you could do without.

What can the comparison not tell you?

It cannot tell you what a channel caused. Both models divide credit for sales that happened, among the touches GA4 recorded. Google describes data-driven attribution as weighing converting and non-converting paths. Still, a touch GA4 never recorded gets no credit under any model, and a mislabelled one gets credit under the wrong name.

It cannot see past your lookback window. A touch older than the window gets no credit under either model, so a slow journey can lose its opener. Check the window in your attribution settings before you read a channel that starts long journeys.

It cannot bring back the old models. First click, linear, time decay and position-based left GA4 in November 2023. The drop-downs now offer data-driven, paid and organic last click, and Google paid channels last click.

It cannot always tell you what Direct means. In this report, Google also uses Direct for key events with no path data at all, such as imported ones.

To learn what a channel caused, test it: hold it back for a while in some regions and compare. The comparison tells you where to aim that test.

What to do this week

1. Find your ceiling. In GA4, open Advertising, then Attribution paths, pick purchase and set Path length to greater than 1. Google's paths page lists it as Key event attribution paths, under a Key events drop-down. Pass: you can name the share of purchase revenue on multi-touch paths. Fail: purchase is not in the list, so it is not a key event and no model can credit it.
2. Set Google's own model against last click. Open Advertising, then Attribution, then Attribution models. Choose Google paid channels last click and paid and organic last click in the two drop-downs. Pass: you can say how much revenue a Google ad touched while another channel closed. Fail: the Google Ads rows stay empty under both, so its clicks are not reaching GA4 as Google Ads traffic.
3. Write the model next to every number you share. In Admin, under Data display, open Events, then Attribution settings. Note the reporting model and the lookback window. Pass: every channel figure in your next report names its model. Fail: two people quote one channel under two models.

Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework

Sources, 1 October 2026: Key event attribution models report (Google Analytics Help); Get started with attribution (Google Analytics Help); Select attribution settings (Google Analytics Help); Key events attribution paths report (Google Analytics Help)

Related answers

• All GA4 how-to questions
• How to read GA4's attribution models report, step by step
• How should a food brand read GA4's model comparison?
• Incrementality testing