Ad Attribution: How Conversion Credit Is Assigned and Where It Falls Short

Ad attribution is the method that decides which ad interactions receive credit for a conversion, and how much each one gets. In Google Ads today that means choosing between data-driven attribution and last click. Either way, the result tells you how credit was divided among the interactions that were observed, not whether the ads caused the conversion.

This guide covers which attribution models remain in Google Ads and Google Analytics 4, how conversion windows, interaction types and modeled conversions shape the numbers, why the two products disagree, and how attribution differs from incrementality testing.

What ad attribution is and what it is not

Attribution is an accounting rule. A conversion happens, the system looks back over the ad interactions it can connect to that conversion within a set window, and it distributes one conversion’s worth of credit among them. Switching models within one system does not change how many conversions happened. It changes how they are split between campaigns, ad groups and keywords.

It helps to be clear about what attribution is not:

• Not proof of cause. A brand search click before a purchase gets credit even if the customer would have bought anyway.
• Not the whole journey. Each system sees only the interactions it can observe. Google Ads assigns credit among interactions with your Google ads, while an analytics tool sees the channels its tags record.
• Not a measure of traffic quality. A model can credit a campaign with conversions that later turn out to be junk leads. Quality needs its own review.
• Not a click count. Attribution starts after a click or view has been recorded and linked to a visitor, which depends on how ad clicks are tracked.

Attribution models in Google Ads today

Google’s page about attribution models lists two models:

• Last click gives all credit for the conversion to the last-clicked ad and its keyword.
• Data-driven distributes credit based on your past data for that conversion action. It estimates how much each ad interaction on the path actually contributed, instead of applying a fixed rule.

Google says data-driven is the default attribution model for most conversion actions. The same page confirms that first click, linear, time decay and position-based models are no longer supported, and that conversion actions which used them were moved to data-driven attribution.

The timing was set out in Google’s April 2023 notice that these rule-based attribution models were going away. From June 2023, advertisers could no longer select them for conversion actions not already using them. From September 2023, conversion actions still using them were switched to data-driven attribution, with last click available for anyone who preferred it. The notice states that data-driven attribution, last click and external attribution were not affected.

Google Ads also offers a Model comparison report that sets two models side by side, which is the safest way to preview a model change.

Attribution models in Google Analytics 4

Google Analytics has its own attribution settings, which apply to its reports. Google’s attribution overview for Analytics describes three models:

According to the same page, the first click, linear, time decay and position-based models have not been available in Analytics since November 2023. It also notes that conversions can be reattributed for up to 7 days after the conversion, so recent credit can still shift.

Conversion windows and types of ad interaction

A conversion window is the period after an ad interaction during which a conversion is recorded in Google Ads. Google gives these defaults:

• Click-through: 30 days by default, adjustable from 1 day up to 30, 60 or 90 days depending on the conversion source.
• Engaged-view: adjustable from 1 to 30 days.
• View-through: 1 day by default, adjustable from 1 to 30 days.

Google says changes to a window apply to conversions going forward, so changing a window does not rewrite history.

The interaction types matter as much as the windows:

• Click-through conversions follow a click on the ad.
• Engaged-view conversions apply to video ads. For a skippable in-stream ad, the viewer must watch at least 10 seconds, or the whole ad if it is shorter, and then convert within the engaged-view window. In-feed and Shorts ads use a 5-second threshold. Google says default engaged-view windows vary by conversion type, and that these conversions appear in both the Conversions and All conversions columns.
• View-through conversions occur, in Google’s words, when customers view but don’t interact with your ad and later complete a conversion on your site. Google’s guide to conversion tracking data says that on the Display Network the last viewable impression gets the credit. View-through conversions are not included in the Conversions column. They appear in the View-through conversions and All conversions columns.

When comparing campaigns, check which column you are reading. A video campaign judged on All conversions and a search campaign judged on Conversions are not measured alike.

Cross-device and modeled conversions

Some conversions cannot be observed directly. The same Google conversion data guide describes cross-device conversions as starting with a click on an ad from one device and ending with a conversion on another.

Google uses modeled conversions to estimate conversions it cannot observe directly. The model predicts whether an ad interaction led to a conversion that occurred, not whether a conversion occurred at all. Google lists these situations:

• Some browsers, such as Safari and Firefox, don’t allow conversion measurement using third-party cookies.
• Advertisers using consent mode will see modeling for users who did not consent.
• Conversions from ads on traffic affected by Apple’s App Tracking Transparency are modeled.
• Cross-device journeys may not be attributable to the ad interaction without modeling.

Google reports both modeled and observed conversions in the Conversions column. It says the modeling uses data that doesn’t identify the user, that it is checked with holdback validation, and that modeled conversions are included only when Google is highly confident they occurred. For analysis, this means the Conversions column is partly an estimate. Treat small week-to-week differences with caution, especially in low-volume campaigns.

Why Google Ads and GA4 report different conversion numbers

Different numbers for “the same” conversions are expected. Google’s conversion data guide and the Analytics attribution overview point to these main differences:

Within Google Ads itself, the Conversions column covers primary conversion actions. All conversions adds secondary actions and view-through conversions, so the two columns will rarely match each other, let alone Analytics.

Google says discrepancies of often up to 20% are expected because of different attribution models. It also notes that reporting can lag by 24 to 48 hours. Google Ads can report conversions by the time the conversion occurred, which makes comparisons with other tools fairer.

To reconcile the two, compare one conversion action at a time, over a range that excludes the last few days. Set Google Ads conversions by conversion time against the Advertising section of Google Analytics, which Google points to for fairer comparisons, and note which model each side uses. Mixing models or columns in one report is one of the classic advertising reporting mistakes.

Attribution versus incrementality

Attribution answers “how should credit be shared among the ad interactions we saw?” Incrementality asks “how many conversions would not have happened without the ads?” The two can give very different answers, for example for brand search or retargeting of people already close to buying.

Google describes incrementality experiments as the way to measure the causal impact of ads. In Conversion Lift, the audience is split into a treatment group that sees your ads and a control group that does not. The difference in conversions between the two groups is the lift. Similar logic underpins simpler holdout designs, such as pausing a campaign in a set of comparable regions while it keeps running elsewhere.

Experiments have limits too. They need enough volume and time for a clear result, they describe one period and one setup, and the control group forgoes conversions. A common pattern is to optimize day to day on attribution and run occasional lift tests to check that the attributed numbers are plausible.

An illustrative example: same conversions, different credit

The figures here are hypothetical and the split is invented for illustration. A real data-driven model derives its split from your account’s data. Imagine 100 conversions, all from people who clicked a generic search ad first and a brand search ad last.

The account total is identical under both models: $3,200 for 100 conversions, or $32 each. Only the split changes. Under last click, the generic campaign looks worthless and could be cut. Under the data-driven split, it looks like a useful contributor. Neither view proves that cutting the generic campaign would cost conversions, which is a question for an experiment. The same caution applies to any per-campaign figure that depends on the model, including the cost-per-result comparisons described in cost per click measurement.

Practical guidance

1. Choose one model as your reporting standard. Document it, and do not compare figures produced under different models.
2. Know which column you are reading. Conversions and All conversions answer different questions. Check whether view-through and engaged-view conversions are included.
3. Respect windows and lag. Leave the most recent days out of judgments, and remember that window changes apply only from the change onward.
4. Preview before you switch. Use the Model comparison report before changing a model, and annotate the date of any change.
5. Check attribution when results shift. A model or window change can move reported conversions overnight. It belongs early in any structured performance investigation.
6. Pair credit with quality. Attributed conversions still need to be real. Segment-level review of engagement and lead outcomes, as in ad traffic analysis, catches credit flowing to junk.
7. Test the big assumptions. Before large budget moves based on attributed results, consider a lift or holdout test.
8. Start from the click. Attribution can only allocate what tracking captured, so the basics of how Google Ads counts clicks underpin everything above.

Key takeaways

• Ad attribution divides credit among observed ad interactions. It does not measure whether the ads caused the conversion.
• Google Ads now offers data-driven and last click attribution. First click, linear, time decay and position-based models were retired in 2023 in both Google Ads and Google Analytics.
• Conversion windows, interaction types (click, engaged-view, view-through) and modeled conversions all shape the numbers you see.
• Google Ads and GA4 differ in dates, models, scope, counting and columns, so some disagreement is expected. Reconcile like with like.
• Use incrementality tests to check whether attributed results reflect real causal impact before making large budget decisions.