Marketing attribution models: how they change your results and how to log a switch

What are marketing attribution models?

A marketing attribution model is the rule a platform uses to split the credit for one conversion across the clicks and touchpoints that came before it. The model does not change how many people bought; it changes which campaign, keyword or channel the report says brought them.

Models fall into two families. Rules-based models apply a fixed formula, such as giving everything to the last click. Data-driven attribution looks at converting and non-converting paths in your own account and estimates how much each interaction changed the chance of a conversion.

Multi-touch attribution is the general name for any model that gives credit to more than one touchpoint. Inside Google's tools, data-driven attribution is now the only multi-touch option.

Which attribution models can you still use in Google Ads, GA4 and Meta?

Google Ads offers two models today: last click and data-driven. GA4 offers data-driven, paid and organic last click, and Google paid channels last click. Meta works differently: it uses attribution windows instead of models.

Google Ads

Google's help page says data-driven is the default for most conversion actions. First click, linear, time decay and position-based stopped being selectable in June 2023, and conversion actions still using them were moved to data-driven from September 2023. You change the model per conversion action: Goals > Summary, click the conversion name, Edit settings, Attribution model, then Save and Done.

GA4

• Data-driven: credit based on data for each key event, specific to each property and key event.
• Paid and organic last click: ignores direct traffic and gives 100% to the last channel (also called last non-direct click).
• Google paid channels last click: 100% to the last Google Ads channel clicked; if there is none, it falls back to paid and organic last click.

You set it in Admin > Data display > Events > Attribution settings, with at least the Marketer role. The same screen holds the key event lookback window (30 or 7 days for acquisition key events, 90, 60 or 30 days for the rest) and the channels that can receive credit.

Meta Ads

Meta reports conversions inside click, view and engagement windows chosen on the ad set. According to Dataslayer's summary of Meta's announcement, since March 2026 click-through only counts link clicks, and likes, shares, comments and short video views moved into a new one-day "engage-through" bucket. If you compare Meta results across that date, part of the gap is definitional.

Why does changing the attribution model look like a performance change?

Because a model switch moves conversions between campaigns and channels overnight, while real sales stay the same. A report read without that context shows winners and losers that did nothing different.

Three details make this worse:

• Google Ads changes the bidding target. The model applies to the Conversions column, and Google says bid strategies that use it, such as Target CPA and Target ROAS, will optimize based on the model you pick. A model switch is therefore also a bidding change.
• Google Ads only changes future counting. Data before the switch keeps the old model in the Conversions column, so a monthly chart mixes two models. The "current model" columns show how history would look under the new one.
• GA4 rewrites history. A new reporting attribution model applies to historical and future data. A report you exported last month will not match the same report today. Lookback window and channel changes apply only going forward.

Data-driven attribution in GA4 can also reattribute conversions for up to 7 days after they happen, so recent days can shift without anyone touching a setting. Read last week's numbers with that in mind before you call it a trend.

The same logic explains why GA4 and Google Ads rarely agree; see why GA4 and Google Ads conversions don't match.

How do I compare attribution models before switching?

Compare the models on the same data before you switch, and write down the expected shift. Then, when reports move after the switch, you already know how much of the move is the model.

1. In GA4, open Advertising > Attribution > Attribution models. The report puts key events and revenue side by side for two models and shows a % change column per channel.
2. Switch the dimension from primary channel group to campaign or source/medium to see which campaigns gain or lose credit.
3. In Google Ads, add the "current model" columns to see how conversions would look under the model each action uses now.
4. Write down the channels with the biggest % change. Those are the ones your client or boss will ask about.
5. Pick a switch date at the start of a reporting period, so one month is not split between two models.

GA4 notes that Google Ads conversions based on key events always use last click, so when you compare against Google Ads, use paid and organic last click in GA4.

How do I isolate which change caused the shift: the model or the campaigns?

Put the attribution change on the same timeline as every other change from the days before the shift, rule out tracking first, then test one explanation at a time. If total conversions stayed flat while credit moved between channels, the model is the likely cause.

1. Rule out tracking
2. Check total conversions
3. Line up all changes
4. Compare models on same dates
5. Test one hypothesis

• Total versus split. A model change moves credit; it rarely moves the total in GA4. If the total also fell, look for a tag, consent or site change too.
• Same dates, both models. Run the attribution models report for the period after the switch. If the old model reproduces the old pattern, the campaigns did not change.
• Bidding side effects. In Google Ads, a model switch changes what Smart Bidding optimizes toward; real performance can move in the weeks after. See the Smart Bidding learning period.

With every change on one timeline, you would see that GA4 moved to data-driven on the 3rd and that branded search "lost" conversions that same day, while nothing in the account changed. In Changeline, Google Ads edits are imported automatically and the team logs the GA4 and Meta settings next to them.

Example: A client asks why paid social "dropped 30%" in GA4 in May. The agency lines up the changes from late April: no new creatives, no budget cuts, but the reporting attribution model went from paid and organic last click to data-driven on 1 May. Total key events were flat. The attribution models report showed paid social at last month's level under the old model. The answer: credit moved, sales did not.

How should you log an attribution model change?

Log it like any change that moves numbers: date, tool, setting before and after, who changed it and why. Without that entry, the switch looks like a performance swing in every report that follows.

Also log changes that look similar: new conversion actions, a primary or secondary goal change, lookback window edits and Meta attribution window edits. Add a note in GA4 itself too; see GA4 annotations. A shared marketing change log keeps these entries with your campaign edits, so the next person reading a report has the context.

Which attribution model should you choose?

For most advertisers, data-driven is the sensible default because it is the default in Google Ads and the only multi-touch option left. Last click is fine when you need a simple, stable rule or have to match another system that uses last click.

• Keep reporting and bidding consistent. If Google Ads bids on data-driven but the client report uses last click, the two will tell different stories.
• Change rarely. Each switch breaks comparisons with earlier periods. Pick a model, record it, and keep it for at least a few quarters.
• Tell people before the report moves. A one-line note in the monthly report saves a call later. If a client still asks why a channel fell, use the steps in how to explain a conversion drop to a client.

No model shows true incrementality. Attribution splits credit for conversions that were tracked; it does not prove a channel caused them. Use holdout or lift tests when you need that answer.