Which attribution model should I use in Google Ads?

A customer often clicks more than one of your ads before they buy. An attribution model decides how the credit for that sale is shared among those clicks, and so which keywords, ad groups and campaigns the conversion is counted under. Google Ads offers two:

• Data-driven shares the credit by what your own account’s data shows each click contributed. It compares the paths of people who converted with the paths of people who didn’t, across clicks and video engagements on Search, Shopping, YouTube, Display and Demand Gen ads, and gives more credit to the steps that make a conversion more likely. It’s the default for most conversion actions.
• Last click gives all the credit to the last ad clicked and its keyword. Every click before it earns nothing.

First click, linear, time decay and position-based are gone. In June 2023 Google stopped letting you choose them for an action not already on one, and from September 2023 it moved every action still on one to data-driven. Last click and external attribution, where another tool has already decided the credit, were left as they were.

The model is set on each conversion action, and it decides what that action adds to the Conversions and All conv. columns. Every bid strategy that learns from the Conversions column, such as Target CPA or Target ROAS, learns from those numbers. Only your primary actions reach that column; the conversion tracking guide says how to check which those are.

Under last click, the generic searches that start customers look like they rarely convert, and the brand searches that end the journey look like they do all the work. Smart Bidding believes the numbers it’s given, so it bids down the searches that find new customers. Google’s own Model comparison report exists to find those keywords and campaigns: the ones undervalued on a last-click basis.

Every conversion action can use data-driven attribution, whatever its volume. Google recommends at least 200 conversions and 2,000 ad interactions in 30 days for it to share the credit accurately. With less, it still runs, and on a small account it can give the same result as last click.

1. 01Go to Goals › Attribution, and select Model comparison in the page menu.
2. 02Choose Campaigns, Ad groups or Keywords from the Dimension menu.
3. 03Set Compare to Last click and With to Data-driven, and pick a date range that leaves out the last few days.
4. 04Read Cost / conv. or Conv. value / cost under each. The rows whose cost per conversion falls, or whose return rises, under data-driven are the ones last click undervalues.

The report’s totals won’t match the Campaigns table. It leaves out conversions from Search partners, Gmail and App campaigns, and an offline conversion uploaded more than 7 days after it happened, and it dates each conversion by when it happened rather than by the click.

1. 01Go to Goals › Conversions › Summary.
2. 02Select the conversion action’s name, then Edit settings.
3. 03Select Attribution model, and choose Data-driven. Last click is the other choice there.
4. 04Select Save, then Done.

The setting is there for website and Google Analytics conversion actions. For conversions tracked across accounts from a manager account, it’s set in the manager account. The Switch to DDA tab under Goals › Attribution moves eligible actions to data-driven in one place.

The change counts conversions from then on; the ones already recorded keep the credit they had. To see how past conversions would read under the new model, add the current model columns from the Attribution section of the Columns menu: Conversions (current model), Cost / conv. (current model) and Conv. value / cost (current model) among them.

Google says to expect three changes in the Campaigns table after a switch to data-driven:

• Decimals. A conversion’s credit is shared among clicks, so the Conversions column shows fractions.
• A dip in the last few days. Google Ads dates a conversion by the click, and data-driven credits earlier clicks too, so the most recent days fill in more slowly. Wait until the Avg. days to conversion in the Path metrics report have passed before judging it.
• Credit moving between campaigns, ad groups and keywords that count the same action.

A target CPA or target ROAS was set in the old model’s numbers, so leaving it where it was makes a campaign that gains credit overbid or one that loses it underbid. Google’s method: over a period that leaves out the last couple of weeks, compare Cost / conv. with Cost / conv. (current model) for each campaign, and move its target CPA by the same percentage. For a target ROAS, compare Conv. value / cost with its current model column the same way.

Each conversion action has its own windows. A conversion after the window is never recorded: it isn’t in your reports, and bidding never learns from it.

• Click-through conversion window: 30 days unless you change it, and from 1 day up to 30, 60 or 90 depending on where the conversions come from. Google recommends at least 7, because a longer window gives a richer set of conversion data.
• Engaged-view conversion window, after someone engages with a video ad: 3 days unless you change it, up to 30.
• View-through conversion window, after an ad was seen and not clicked: 1 day unless you change it, up to 30 days.

To choose one, read how long your customers take: Goals › Attribution, then Path metrics, and its Avg. days to conversion. A window shorter than most customers take undercounts every campaign, and most of all the ones that reach people early. The attribution reports’ Lookback window control is a different thing: it only changes what the report looks back over.

1. 01Go to Goals › Conversions › Summary, select the conversion action, then Edit settings.
2. 02Under the Settings tab, expand Click-through conversion window and choose the number of days.
3. 03Select Save, then Done.

Like a model change, a new window counts conversions from then on: a sale that fell outside the old window isn’t counted after the fact. Smart Bidding counts and optimizes for whatever window you set. A Google Analytics conversion that credits paid and organic channels has its window changed in Google Analytics.

A campaign’s bid strategy adds up every primary action in the goals it bids on. When those actions are credited by different models, or counted over different windows, the total mixes numbers counted by different rules, and a campaign whose customers take longer, or click more ads first, reads worse on one than on the other. Read each primary action’s settings side by side.

Give them the same model, data-driven unless you have a reason, and the same click-through window, the one that fits how long your customers take, unless one action really has a shorter or longer path to it.

Every day Goldbeater reads the settings of each conversion action your bidding learns from: enabled, primary, and in a goal an enabled campaign bids on. It judges the ones measured on your site or imported to it, whose settings are yours; Google sets the model itself for actions it measures, such as directions and calls from Maps, YouTube’s and an app’s. It raises:

• An action that credits the last click alone: one that recorded conversions in the last 30 days on last click, or still on first click, linear, time decay or position-based, which Google hasn’t moved. It advises switching it to data-driven.
• A click-through window under 7 days on such an action. It advises lengthening it to 7 days at least, or to Google’s default of 30, unless it’s short on purpose.
• Actions measured different ways under one strategy: a campaign bidding on two or more that use different models or click-through windows, each action’s model, window and conversions in a table. Where only the windows differ, it drafts each window set for all of them, the one most of them use first, and recommends none, since how long a customer takes to book or to buy is yours to know.

The finding is the table’s first row. The audit checklist has every conversion line with where to find it in Google Ads.

What attribution models does Google Ads have?

Two you can choose for a conversion action: data-driven, the default for most, and last click. First click, linear, time decay and position-based were retired in 2023, and the actions on them moved to data-driven.

Is data-driven attribution better than last click?

Google made it the default: it credits every click that helped, where last click credits only the one before the sale, so bidding stops cutting back the searches that find customers. Google recommends 200 conversions and 2,000 ad interactions in 30 days for it to be accurate, and with less it can give the same result as last click.

Does changing the attribution model change my past conversions?

No. It counts conversions from then on. The current model columns, such as Conversions (current model), show how past conversions would read under the model you’ve chosen.

Why are there decimals in my Conversions column?

Data-driven attribution shares each conversion among the clicks that led to it, so a campaign can be credited with part of one.

What conversion window should I use?

One at least as long as most of your customers take from a click to a conversion, which the Path metrics report shows. Google starts every action on 30 days and recommends at least 7.

Is the attribution window the same as the conversion window?

Usually that’s what’s meant: the conversion window on a conversion action, which decides which conversions are recorded at all. The attribution reports also have a Lookback window control, which only changes how far back a report looks.