Attribution Models in Google Ads 2026: DDA vs Last Click

By autumn 2026 Google Ads has just two attribution models left: data-driven attribution (DDA) and last click. The other four — first click, linear, time decay and position-based — are gone, and conversion actions that ran on them were migrated automatically. In most accounts it happened quietly: no interface warning, no change to campaigns or budgets.

The problem is that attribution is not a reporting preference. Smart Bidding trains on attributed credit. Changing the model does not change numbers in a table — it changes bidding behaviour, on a delay rather than instantly. Below: what remains, what broke in your period-over-period comparisons, where to read current settings, and how to audit the whole thing in twenty minutes.

Attribution decides what lands in the Conversions column. The Conversions column decides what automated bidding learns from. So the attribution model is a bidding setting wearing a reporting setting’s clothes.

Which attribution models remain in Google Ads in 2026

Model	Status	Credit logic
Data-driven attribution (DDA)	active, default model	Distributes credit based on your past data for that conversion action
Last click	active	All credit to the last-clicked ad and corresponding keyword
First click	removed	—
Linear	removed	—
Time decay	removed	—
Position-based	removed	—

Google’s documentation is blunt about it: conversion actions using the deprecated models have been upgraded to use data-driven attribution. The practical meaning — there is no longer a choice among six approaches, only a choice between “machine distribution across your own data” and “everything to the last click”.

Timeline: when this actually happened

Mid-July 2026. The four models became unavailable for new conversion actions.
September 2026. Conversion actions still running on them were force-migrated — to DDA in accounts with sufficient volume, and effectively to last-click behaviour in the rest.

The change happened at conversion-action level with no change to campaigns, budgets or dashboards. That is why many accounts discovered it not through a notification but through an unexplained break in the trend line: upper-funnel campaigns “lost” conversions while brand campaigns “found” them. Nothing in the market had moved.

If you saw an unexplained shift on your charts between July and September 2026 and found no cause in your change history and performance drop diagnosis, check conversion-action attribution before you check the campaigns.

How DDA works and what it learns from

DDA differs from the others in that it applies no universal rule. It calculates the actual contribution of each interaction across the conversion path using your account’s own data, learning from the historical data of that specific conversion action and splitting credit in proportion to measured influence.

What matters about its scope:

Attribution models apply to website and Google Analytics conversion actions. In-app conversions may be attributed to App campaigns for reporting.
Both Google paid channels and organic channels can receive credit — DDA sees more than just your campaigns.
Accounts that only recently became eligible for cross-network attribution may see a delay before complete data appears.
Datasets with insufficient volume are filtered out automatically.

That last point sets up the next section.

The data threshold: why small accounts get last click in DDA clothing

DDA needs volume: to distinguish the contribution of touchpoints, the model has to observe enough distinct paths to conversion. The industry working estimate for reliable DDA is roughly 300 conversions per 30 days per conversion action. That is an orientation figure rather than a Google-published threshold, but it is a useful boundary.

What happens below it:

There is too little path data for the model to isolate the contribution of intermediate touches with confidence.
Credit distribution drifts toward the last click — the one touch that is always observed and unambiguous.
Your report says “DDA” while the behaviour is close to last click.
Upper-funnel contribution disappears from reporting, and budget decisions tilt to the bottom of the funnel.

This is the worst scenario because it is silent. The model never says “I lack data” — it simply returns an output that resembles last click, and you read it as machine learning. Hence the working rule: if a conversion action produces fewer than a few hundred conversions a month, do not draw conclusions about the value of reach formats from its DDA report. Those decisions need geo experiments and incrementality testing, or aggregated approaches like MMM and Meridian.

Under the hood: attribution → the Conversions column → bids

The chain looks like this.

Step 1. A user takes a multi-touch path: sees a Demand Gen ad, clicks a non-brand search query, comes back two days later through brand and buys.

Step 2. The attribution model decides how to split that one conversion across those touches. Under last click the whole conversion goes to brand. Under DDA it is divided fractionally across all three.

Step 3. The result lands in the Conversions and All conversions columns. A model change affects them going forward; historical data is not recalculated.

Step 4. Automated strategies — Target CPA, Target ROAS, enhanced CPC — use those same figures as their training signal. Credit changed, so the map of where the algorithm sees results changed, and it starts reallocating bids.

Two consequences worth pinning to the wall:

Changing the model is a bidding operation, not a reporting one. Google itself recommends testing a new model before rolling it out fully.
A volatility period follows a model change. Automated bidding re-learns on the new credit, and it behaves like the Smart Bidding learning period: one to two weeks of instability that manual edits only make worse.

Worked example: one account, two models

Take 100 conversions and three campaigns that appeared on the paths. Figures are illustrative.

Campaign	Conv., last click	Conv., DDA	Spend	CPA last click	CPA DDA
Search — brand	58	34	900	15.5	26.5
Search — non-brand	34	43	2,600	76.5	60.5
Demand Gen — reach	8	23	1,500	187.5	65.2
Total	100	100	5,000	50.0	50.0

Three things this table shows:

Total conversions and total CPA are unchanged. Attribution neither creates nor destroys conversions — it moves them. Any claim that “conversions dropped after the model change” at account level means the wrong things were compared.
Priorities invert. Under last click, Demand Gen is the worst campaign in the account and the first candidate for shutdown. Under DDA it is comparable to non-brand search. Same money, same outcome, opposite decisions.
Brand gets artificially cheap. Under last click it absorbs credit for demand other campaigns created. That is the root reason brand always looks like the best campaign in the account — and why any budget allocation tool pulls money toward it.

Practical takeaway: before acting on Recommended Investment Strategy and its Holistic and Growth modes, check which attribution model produced the numbers it is calculating on. Reallocating budget on last-click data systematically kills the upper funnel.

Why comparisons across the migration date lie

A model change applies forward and does not recalculate history. So in your month-by-month table, everything before the switch is one credit system and everything after is another.

Which means:

Campaign-level “September vs August” comparisons are meaningless if the migration fell between them.
Year-over-year comparisons are distorted too, if the model was different a year ago.
Client reports need a migration-date note, or you will spend time explaining a drop that never happened.

Minimum hygiene: annotate the migration date on every dashboard and in your custom columns and Report Editor views, and separate the periods visually in comparison tables. Two minutes of work that saves hours of explanation.

A twenty-minute attribution audit

Open the conversion actions list and add the attribution model column. If it is not shown, enable it in the column settings.
Write down the model per action. A mixed picture — some on DDA, some on last click — is normal after the automatic migration, but you need to know which is which.
Check volume per action. Conversions in the last 30 days. Anything well below a few hundred should be read as “DDA is behaving like last click”.
Screenshot the settings and a 30-day report. This is your baseline; it cannot be reconstructed later.
Annotate the migration date across all dashboards and reports.
Identify Target CPA and Target ROAS campaigns optimising toward the affected conversion actions. Those are the ones whose behaviour changed.
Check the conversion window on the same actions — a separate setting that is routinely confused with attribution.

The output of the audit is not “the correct model” but a clear view of the data you are currently deciding on. That alone prevents the expensive mistakes.

Attribution vs conversion window: two settings people conflate

	Conversion window	Attribution model
Question it answers	How long after a click a conversion still counts as yours	How to split a conversion across touches inside that window
Affects	How many conversions appear in reporting at all	Which campaign receives the credit
Changes apply	Can recalculate data	Forward only

In practice: if total account conversions dropped, look at the conversion window and tracking. If the total is the same but redistributed across campaigns, look at attribution. The first setting is covered in detail in the guide to conversion windows and conversion lag.

Attribution in PMax, Demand Gen and AI Max: what you will not see

In automated-placement campaigns, intra-campaign attribution is opaque to you. Credit arrives at the campaign, but the channel breakdown inside it is limited and per-audience or per-creative credit is essentially unavailable.

What follows for decision-making:

Comparing PMax and Search on CPA is only valid within one attribution model and with cross-campaign cannibalisation accounted for.
Evaluating individual placements inside automated campaigns through attribution is not possible — that requires experiments.
The signals you hand the platform affect credit indirectly: the richer your first-party data, the more complete paths the model observes. This is where attribution meets the Data Strength metric.

Attribution vs incrementality: different questions

Attribution answers “which of my channels should this conversion be booked to”. Incrementality answers “would this conversion have happened without advertising at all”. They are not refinements of each other; they are independent measurements.

Neither DDA nor last click answers the second question. Brand search can collect 60% of credit under DDA and still be almost entirely non-incremental: the user would have found you anyway. The only way to know is to switch it off — a geo experiment or a hold-out.

So the working division of labour: attribution for allocating budget between channels, incrementality for deciding whether a channel deserves money at all. Substituting the first for the second is the most expensive mistake in paid media measurement.

Changing the model without a crash: a four-week plan

Week 0. Freeze a baseline: campaign-level export for 28 days, screenshots of settings, dashboard annotations.
Week 1. Change the model on one conversion action, not all of them. Touch nothing in the campaigns — no bids, no budgets, no targeting.
Week 2. Observe. Automated bidding is re-learning; spikes and dips in this window are not signal.
Week 3. Review how credit redistributed across campaigns. You are looking for “where did credit move”, not “did it get better”.
Week 4. Revisit target CPA and ROAS: if credit shifted, your old targets no longer mean what they meant, and they need recalculating against the new map.

One caveat: never change the attribution model and your bid targets in the same week. Otherwise in a month you will not be able to say which change produced the result.

Seven mistakes

Treating attribution as a reporting setting. It feeds automated bidding.
Comparing periods across the migration date. Those are two different credit systems.
Trusting DDA at low volume. Below a few hundred conversions a month it behaves close to last click.
Killing reach campaigns on last-click CPA. You are switching off a demand source, not an unprofitable campaign.
Changing the model on every conversion action at once. You lose all control.
Reacting to week one after the switch. Bidding is re-learning.
Confusing attribution with the conversion window. One moves credit, the other changes how much credit exists.

Persistent misconceptions

“DDA inflates conversions.” It does not. Total conversion count is model-independent; only the split across campaigns changes.

“Last click is more honest because it is a fact.” Last click is also a model — just the crudest one. It assigns the whole result to the touch that is easiest to observe.

“Four models were removed, so DDA must be better in every case.” DDA is better with sufficient data. Below the threshold it loses its advantage, and that is not a reason to treat it as a universal answer.

“Attribution will show the real value of a channel.” It will not. Real value is incrementality, and that is measured by experiment, not by model.

FAQ

Which attribution models are available in Google Ads now?

Two: data-driven attribution (the default) and last click. First click, linear, time decay and position-based have been removed.

When were the four models removed?

They became unavailable for new conversion actions in mid-July 2026, and by September 2026 the remaining conversion actions using them were migrated automatically.

What happened to my conversion actions on the removed models?

They were upgraded to data-driven attribution. In accounts without sufficient volume, the model’s behaviour is effectively close to last click.

Does changing the model change historical data?

No. The change applies forward only, to the Conversions and All conversions columns.

How much data does DDA need?

There is no published official threshold; the industry working figure is around 300 conversions per 30 days per conversion action. Datasets with insufficient volume are filtered out automatically.

Does the attribution model affect Smart Bidding?

Directly: Target CPA, Target ROAS and enhanced CPC all train on attributed credit. Google recommends testing a model before full rollout.

Why did my brand campaign look more expensive after the migration?

Because under DDA part of the credit moved to the touches that created the demand. Under last click, brand collected all of it.

Which conversions do attribution models apply to?

Website and Google Analytics conversion actions. In-app conversions may be attributed to App campaigns for reporting purposes.

Does DDA account for organic channels?

Both Google paid channels and organic channels can receive credit, so DDA observes more touchpoints than your campaigns alone.

Should I switch back to last click?

There is one defensible reason: your data volume is too low and you need simple, reproducible logic for reporting. In every other case last click systematically undervalues the upper funnel.

How is attribution different from the conversion window?

The conversion window decides how many conversions enter reporting; attribution decides which campaign gets credit inside that window.

How do I measure a channel’s real contribution if attribution is not enough?

With an experiment: a geo test or hold-out where the channel is switched off in part of the market. For large budgets, aggregated models such as MMM.

What to do next

Attribution closes the loop on three decisions about account economics. First, where money goes between campaigns: Recommended Investment Strategy and its Holistic and Growth modes. Second, which event triggers a charge: pay for conversions in Google Ads. Third, who gets the credit — this article. Solving them separately makes no sense, because each one changes the numbers the other two are decided on.

A structured walk-through of measurement and conversion setup is part of the Google Ads guide, and if you need dependable infrastructure for running several accounts, look at Google Ads agency accounts from PPC Rebels.

Attribution Models in Google Ads 2026: DDA vs Last Click After the Forced Migration