Google Ads Has Two Attribution Models Left, and Most Advice Predates That

Attribution advice ages badly, and Google Ads attribution advice has aged particularly badly, because four of the six models people used to argue about no longer exist. First click, linear, time decay and position-based are gone. Google retired them and upgraded the conversion actions that used them to data-driven attribution. What is left is last click and data-driven, and the choice between two things is a different conversation from the choice between six.

This matters beyond trivia for two reasons. Clients still ask for models that cannot be selected, and historical reports in long-running accounts were built on models that were changed underneath them.

What is left

• Data-driven attribution.Google’s model, which distributes credit across the ad interactions on the path using your account’s own conversion data. Google describes it as the default attribution model for most conversion actions.
• Last click. All credit to the final ad click. Simple, predictable, and the only remaining alternative.

That is the list. If you are looking for a dropdown with six options, it is not hidden — it is not there.

What went, and what happened to the data

Google’s help documentation states that the first click, linear, time decay and position-based attribution models are no longer supported, and that conversions which used them have been upgraded to use data-driven attribution. The retirement was staged rather than overnight: Google’s announcements described the models as going away from early April 2023, with the ability to select them removed in the months after.

The phrase to sit with is “have been upgraded”. Those conversion actions did not stop working or start erroring. They were moved to a different model, which means the way credit is assigned to them changed, which means the numbers changed. For an account that had been running position-based attribution for years, the series has a seam in it.

The thing to understand about changing models

Attribution model changes in Google Ads are not reporting filters applied at read time. Changing the model changes how conversion credit is assigned, and that reassignment shows up in historical rows too. Which means:

• A year-on-year comparison that spans a model change is comparing two different measurement systems.
• Credit can move between campaigns, so one campaign getting better and another getting worse may be an artefact of redistribution rather than a change in performance.
• Fractional conversion counts are normal under data-driven attribution. A campaign showing 7.4 conversions is not a bug; it earned partial credit on several paths.

If you took over an account with a long history and the numbers in old client decks do not reconcile with what the interface shows now, the 2023 model retirement is a strong candidate before you go hunting for a tracking fault. Nothing is broken. The measurement definition changed.

When last click is still the right choice

Data-driven is the default and it is usually the better model, because it uses your own data rather than a rule somebody picked. But it is not automatically correct for every account, and there are cases where last click is the saner choice.

Thin conversion volume

Data-driven attribution is a model fitted to your conversion paths. An account with very low conversion volume gives it little to work with, and the output becomes harder to trust and harder to explain. Last click is crude but it is stable and nobody has to take its arithmetic on faith.

When you need the number to be explainable

Some clients need to be able to reconstruct how a number was produced, particularly if conversions feed invoicing, commission or a partner reconciliation. “The final click before the sale” survives that conversation. “Google’s model distributed credit across the path” does not always.

When you are reconciling against another system

If you are comparing Google Ads against a CRM, a last-touch analytics view or a platform that reports last click, matching the models reduces the number of reasons two systems can disagree. That is not an argument that last click is more true — only that fewer moving parts makes a discrepancy investigation tractable. The wider problem of systems disagreeing is covered in how much tracking discrepancy is normal.

For most accounts with reasonable volume and no reconciliation obligation, leave it on data-driven. The point is that it is a decision, not a default to accept without noticing.

What data-driven attribution is actually doing

It is worth being concrete, because “Google’s model decides” is the kind of explanation that makes people distrust a number they should probably trust.

Data-driven attribution looks at the paths that led to conversions in your account and the paths that did not, and assigns credit to interactions according to how much difference they appear to make. An ad click that shows up on converting paths far more often than on non-converting ones earns more credit than one that appears about equally on both. It is comparative rather than positional: it is not asking “where in the sequence did this happen” but “did this interaction seem to matter”.

That is why the removed models are not much missed by people who understand what replaced them. Position-based attribution gave 40% of the credit to the first interaction because somebody decided 40% sounded right, not because anything in your account suggested it. A rule applied uniformly to every advertiser is a guess wearing a percentage.

The trade-off is interpretability. A rules-based model can be recalculated by hand; a fitted model cannot. That is a real cost, and it is the reason the next section exists.

What happens when you switch, practically

If you are moving a conversion action between the two remaining models, expect three things, and tell the client about them before rather than after.

1. Reported history moves.The change is not applied only to new conversions. Credit gets reassigned, so last month’s campaign-level numbers will not match the screenshot in last month’s report. Save the old figures before you switch if anyone is likely to ask.
2. Smart Bidding has to adjust. The conversion signal it is optimising against has changed shape, so treat it as a bidding change with a settling period, not a reporting preference. Do not switch models and change targets in the same week, or you will not know which one moved performance.
3. Campaign-level winners and losers shuffle. Under last click, a campaign that closes traffic other campaigns warmed up takes all the credit. Under data-driven, some of that credit moves upstream. Brand search campaigns in particular often look less impressive after the switch, which is usually the model being more honest rather than the campaign getting worse.

None of that is a reason to avoid switching. It is a reason to switch on purpose, at a point in the month where a fortnight of unstable numbers does not land in the middle of a reporting cycle.

GA4 is a separate system, with different models

A frequent source of confusion: Google Analytics has its own attribution settings, and they are not the same list. GA4 offers data-driven attribution, paid and organic last click, and Google paid channels last click.

So “we use last click” is ambiguous across the two products, because GA4’s last-click variants differ in which channels are eligible for credit at all. Two systems both set to something called last click can still disagree, and that is before you reach the deeper reasons Google Ads and GA4 count differently, which are covered in Google Ads conversions vs GA4 key events.

If you are importing conversions from GA4 into Google Ads, the model matters at both ends, and the two choices are independent of each other.

What to actually check

1. Open each conversion action and note its attribution model. They are set per action, not per account, so a tidy account can still have a mix nobody intended.
2. Where a mix exists, decide whether it is deliberate. A lead action on last click and a purchase action on data-driven in the same account will produce numbers that cannot sensibly be added together.
3. If the account predates mid-2023 and anyone compares against that history, note the seam somewhere a future reader will see it. This is the kind of detail that is obvious while you are looking at it and completely lost six months later.
4. Check the GA4 property’s reporting attribution separately if conversions are imported from it, and do not assume the two match.

Attribution is not tracking health

Worth drawing a line here, because the two get mixed up constantly. The attribution model decides how credit for a conversion is distributed. Tracking health decides whether the conversion was recorded at all.

Changing the model moves numbers between campaigns. It never makes conversions appear or disappear in aggregate. So if your total conversions dropped, the model is not the explanation — go and look at whether the tag is still firing, whether the conversion action is still active, and whether consent or a site change intervened.

Taglertcovers that second category: it rechecks every connected Google Ads and Meta account every thirty minutes and emails you when a conversion action’s tracking status changes or events stop arriving. It has no opinion about your attribution model, which is a configuration choice rather than a fault. The distinction is the useful part — when a number moves, knowing whether the measurement broke or the measurement definition changed is most of the diagnosis.