The Google Analytics Attribution Report: A 2026 Field Guide

TL;DR

The Google Analytics attribution report is not the thing it was three years ago. Google deleted first click, linear, time decay, and position based models in late 2023, leaving data driven attribution plus two flavours of last click. In April 2026 it merged the model comparison and conversion paths reports into one tabbed Attribution report, recalibrated the model, cut the default acquisition lookback window from 90 days to 30, and reset some properties to data driven attribution without telling anyone. That model also needs 400 conversions for a key event and 20,000 across all events before it activates, and it falls back to the last click silently when you miss that bar. Read the report as a directional claim about which channels appear in journeys, not as proof of what caused revenue, and validate the expensive decisions with an incrementality test.

Key Takeaways

• Four attribution models are gone. The Google Analytics attribution report now offers data driven attribution, paid and organic last click, and Google paid channels last click. Nothing else.
• DDA needs at least 400 conversions for the specific key event and 20,000 total conversions across all events inside the lookback window. Below that it silently reverts to the last click.
• The April 2026 update merged the attribution reports into one, recalibrated the model, and changed the default acquisition lookback window from 90 days to 30. Comparisons that straddle April will show swings that have nothing to do with your campaigns.
• Changing the attribution model applies retroactively to your reports. Changing the lookback window does not. That asymmetry catches people out constantly.
• Since January 2026 you can set the model and lookback window per key event, so a newsletter signup no longer has to be judged the same way as a large purchase.
• Attribution now sees roughly 30 to 60% of touchpoints, down from over 90% before privacy changes. Treat the output as a tactical signal, not a source of truth.
• In the best documented comparison available, an attribution style estimate reported return on ad spend above 4,100% while a randomised experiment on the same spend returned minus 63%. Only experiments make causal claims.

Introduction

If you opened GA4 in the last few months and found the Model comparison report missing, you are not going mad. Google merged it into a single Attribution report with tabs in April 2026, having already moved the settings in Q3 2025 and deleted four models in 2023. The version of this post that used to live here described an interface that has been rearranged three times since it was written, which is why it stopped being useful.

The deeper problem is that most guides to the Google Analytics attribution report explain where the buttons are without explaining what the numbers can and cannot support. That gap costs real money. A channel's conversion count in GA4 is not a measurement. It is the output of a model, plus a lookback window, plus a consent state, and changing any of those three produces a different number from identical customer behaviour. People cut budgets over swings that were configuration changes.

So this rewrite covers what the report contains in 2026, what changed in April, how data driven attribution actually assigns credit, and where attribution stops being able to answer your question. If you want the reporting layer underneath it, our guide to machine learning and business intelligence in GA4 covers that ground.

FAST FACT: Google retired the first click, linear, time decay, and position based attribution models in November 2023. They cannot be restored in any GA4 property. (Source: Optimize Smart, 2026)

What does the Google Analytics attribution report actually show?

Since April 2026 it is one report with tabbed views, found under Advertising. Two things live inside it.

Model comparison, the first tab, puts your available models side by side for the same date range, so you can see how credit shifts when the logic changes. This is the single most useful screen in the whole product, and it is underused because people treat model choice as a decision to make once rather than a comparison to run regularly.

Conversion paths shows the actual sequences that preceded a key event, along with how much credit each position received. This is where you find out that organic search opens most journeys and paid search closes them, which is the sort of thing that changes budget conversations.

One naming point that still trips people up: GA4 renamed conversions to key events on 21 March 2024. In current GA4 language, a conversion means specifically a key event that has been imported into Google Ads. If a report or a colleague uses the two words interchangeably, check which they mean before you compare numbers.

Which GA4 attribution models still exist?

Three, and the shortlist matters because the removed ones were the ones many teams used for a neutral baseline.

• Data driven attribution, or DDA. The default. Distributes fractional credit across touchpoints based on how much each appears to have moved the outcome, using your property's own data rather than a global average.
• Paid and organic last click. All credit to the final touchpoint before the key event, paid or organic. Direct traffic is ignored when a prior touchpoint exists.
• Google paid channels last click. All credit to the last Google Ads click. Useful only when you deliberately want reporting that lines up with the Google Ads interface.

First click, linear, time decay, and position based went away in November 2023. Google's reasoning was that rule based models are outdated. The practical effect was that smaller advertisers lost the models they could actually use, because the surviving default has a volume requirement they cannot meet. If you need a linear or first touch view for planning, you now have to rebuild it yourself from raw event data in a warehouse.

FAST FACT: DDA requires at least 400 conversions for the specific key event and 20,000 total conversions across all events within the lookback window. Below that threshold GA4 falls back to last click without flagging it anywhere visible. (Source: 1ClickReport, 2026)

That silent fallback deserves a moment. A lot of teams believe they are running on DDA when they are getting last click results. There is a quick test: open model comparison and put the two side by side. If the two columns are identical, the model is not active for that key event. Most B2B lead generation accounts fail this test, and an account generating 25 leads a week has nowhere near the signal the model needs.

How does data driven attribution decide who gets credit?

It compares the paths of users who converted against the paths of users who did not, then estimates how much each touchpoint changed the probability of conversion. The underlying method is Shapley values borrowed from game theory, with a time weighting layer on top so that touchpoints closer to the conversion generally receive more credit.

Three properties are worth knowing:

• It evaluates up to 50 interactions per key event, against just four in Universal Analytics. Long consideration journeys are far better represented than they used to be.
• It is trained on your property, not on aggregate industry behaviour, which is why two similar businesses can get genuinely different credit distributions.
• Conversion modelling fills gaps caused by consent refusals and cross device behaviour, so even your observed conversion counts include modelled entries.

That last point unsettles people, and it should be understood rather than feared. In markets with consent requirements, a share of your conversions are estimates informed by the behaviour of users who did consent. The number on the screen is partly measured and partly inferred, and GA4 does not separate the two for you. Anyone presenting these figures to a finance team should say so out loud once, then move on.

A worked contrast makes the stakes obvious. Under last click, a journey that begins with an organic blog visit and ends with a branded paid search click gives everything to paid. Under DDA the same journey might split roughly 35 to organic and 45 to paid, with the rest to direct. Cut the SEO budget on the first view and you have defunded the channel that opened the journey. Our notes on customer segmentation techniques cover how to look at this by audience rather than in aggregate.

What changed in the April 2026 update?

Enough that any comparison spanning April needs a caveat attached. Four changes landed together.

• The Attribution paths and Model comparison reports were merged into a single Attribution report with tabbed views.
• The data driven model was recalibrated, so attributed conversions before and after the update are not directly comparable.
• The default attribution lookback window for acquisition key events changed from 90 days to 30. Other key events kept 90.
• Some properties were reset to DDA even where a model had been deliberately chosen, with no obvious notification.

FAST FACT: The April 2026 update reset some properties to DDA, recalibrated the model, and cut the default acquisition lookback window from 90 days to 30 days. (Source: GROAS, 2026)

If a Search campaign appeared to collapse in April while Display improved, rule this out before concluding anything about the campaigns. Upper funnel channels gained credit under the recalibrated logic and last touch channels lost it, which produces exactly the pattern people misread as a performance change.

The January 2026 change is the more useful one and got less attention. You can now configure the model and the lookback window independently for each key event, under Advertising and then Conversion Management. Judging a newsletter signup and a large purchase by the same rules never made sense, and now you do not have to. Set short windows and simple logic for micro conversions, longer windows and DDA for the events that matter to revenue. For teams running programmatic spend, this pairs directly with how you configure DV360 and Campaign Manager 360.

How should you set the attribution lookback window?

Match it to how your buyers actually behave, not to whatever Google shipped as a default.

The current defaults are 30 days for acquisition key events, which are first_open and first_visit, with 7 days as the alternative. For every other key event the default is 90 days, with 30 and 60 available. The window determines which touchpoints are eligible for credit at all: under a 90 day window, a blog visit 100 days before a purchase earns nothing and effectively did not happen.

A reasonable rule is to measure your median time from first touch to conversion, then set the attribution lookback window to at least twice that. For most B2B lead generation, 90 days is the right answer and anything shorter systematically flatters bottom funnel channels. For a low value ecommerce purchase with a same week decision cycle, 30 days is usually enough and a longer window over credits distant touches.

FAST FACT: Changing the attribution model applies retroactively across all reports, but changes to the lookback window apply going forward only. (Source: Google Analytics Help)

That asymmetry is worth writing on a sticky note. Switch models and your historical numbers move, which is useful for comparison but alarming for anyone who screenshotted last month's dashboard. Change the lookback window and nothing historical moves at all, so you wait months before the setting is fully reflected. Document both changes with dates, because the alternative is spending a quarter explaining variance you created yourself. Our notes on retroactive data analysis in Google Analytics cover what can and cannot be recovered when settings change mid quarter.

Why do GA4, Google Ads, and Meta never agree?

Because they are answering different questions with different windows and different self interest, and because none of them can see the whole journey any more.

Meta reports 60 conversions, Google Ads says 47, GA4 shows 38, and the actual number in your CRM is 41. Nobody is lying. Each platform counts conversions it believes it influenced, inside its own window, under its own model, and the platforms have every incentive to resolve ambiguity in their own favour. Summing them is the most common error in performance reporting.

FAST FACT: Identity coverage for multi touch attribution fell from over 90% to roughly 30 to 60% following Safari ITP, iOS App Tracking Transparency, and consent requirements. Browser side tracking now observes an estimated 25 to 40% of visitor behaviour. (Source: Lead Gen Economy, 2026)

A three tier framework resolves most of the argument. Use GA4 DDA as the neutral baseline for allocating budget across channels, because it is the only view that is not owned by a channel with something to sell you. Use each platform's own numbers for optimising inside that platform, because that is what the bidding algorithms consume. Use CRM and revenue data as the actual scoreboard. Three numbers, three jobs, and no adding up.

Worth noting that third party cookies did not disappear the way everyone planned for. Google confirmed in April 2025 that Chrome would keep them and would not ship a standalone prompt, and the Privacy Sandbox APIs were retired in October 2025. The signal loss came from Safari, iOS, ad blockers, and consent flows instead. The practical fix is the same either way: server side collection, enhanced conversions, and a first party identifier that survives independently of any browser's cookie policy. That work also improves paid advertising performance, since cleaner conversion signals feed better bidding.

What sits above multi touch attribution in 2026?

Two methods, and the honest answer is that you need all three layers rather than a favourite.

Marketing mix modeling uses aggregate spend and outcome data to estimate channel contribution without tracking a single individual, which makes it immune to consent and cookie problems. It used to be a six figure consulting engagement. Google released Meridian to everyone in January 2025 and Meta maintains Robyn, so the barrier now is data rather than budget: you need roughly two to three years of clean weekly spend and outcome data per channel.

Incrementality testing answers the question attribution cannot. Attribution tells you which touchpoints appeared before a conversion. An experiment tells you whether the conversion would have happened anyway. Only the second is a causal claim, and the gap between them is not a rounding error.

FAST FACT: In the best documented comparison available, an attribution style estimate reported return on ad spend above 4,100% while a randomised experiment on the same spend returned minus 63%. (Source: Koji, 2026)

Read that twice. It is the strongest argument in the field for treating attributed ROAS as a hypothesis rather than a result. The entry barrier has also collapsed: Google cut the minimum budget for incrementality tests from over $100,000 to $5,000 in 2025, and around 52% of US marketers now run them. Among US marketers, 27.6% rate mix modeling as their most reliable measurement method against 19.4% for attribution modelling.

The practical sequence, at any budget: fix the inputs with server side collection, use the Google Analytics attribution report as a directional compass for channel allocation, run one geo holdout on your largest channel for a causal anchor, and graduate to marketing mix modeling when spend and history justify it. Feed the outputs into media buying strategy and into the data visualization dashboards your leadership team reviews, rather than leaving them in a slide deck.

Summary

The Google Analytics attribution report in 2026 offers three models: data driven attribution, paid and organic last click, and Google paid channels last click. The other four were deleted in November 2023. DDA needs 400 conversions for a key event and 20,000 across all events before it activates, and reverts to last click silently when you fall short, so verify it by comparing models rather than trusting the setting. The April 2026 update merged the reports, recalibrated the model, and cut the default acquisition lookback window to 30 days, which means any comparison spanning April needs a caveat.

Set the lookback window to at least twice your median time to conversion, and remember that model changes apply retroactively while window changes do not. Then hold the whole thing loosely. Multi touch attribution now observes 30 to 60% of touchpoints, and a documented comparison found attributed ROAS above 4,100% where the randomised experiment on the same spend returned minus 63%. Use the report to decide where to look, controlled experiments to decide what is true, and mix modeling to decide where the budget goes.

Next Steps

Before your next planning meeting, do three things. Check which model each key event is using, confirm DDA is actually active by comparing it against last click, and write down the date of any settings change so future variance has an explanation. Then read the Google Analytics attribution report next to value rather than volume, using our guide to the Google Analytics lifetime value report and our notes on predictive analytics for customer lifetime value. For validating changes you make as a result, see our comparison of experimentation and testing tools, and for the wider picture, using data analytics for business success.

Not sure whether your attribution setup is telling you the truth? Book a measurement audit with the Lucrative AI team.

Frequently Asked Questions

Which attribution models are available in GA4 in 2026?

Three: data driven attribution, paid and organic last click, and Google paid channels last click. First click, linear, time decay, and position based were removed in November 2023 and cannot be restored. DDA is the default for new key events, and last click is now the opt in choice rather than the other way round. If you need one of the deleted models, you have to rebuild it from raw event data in BigQuery or use a third party tool.

Why did my GA4 conversion numbers change in April 2026?

Most likely because of the platform update rather than anything your campaigns did. Google recalibrated the data driven model, merged the attribution reports, cut the default acquisition lookback window from 90 days to 30, and reset some properties to DDA even where another model had been set deliberately. Check your attribution settings and lookback windows first, then compare model outputs, before drawing any conclusion about channel performance.

How do I know if data driven attribution is actually running?

Open the model comparison view and place it next to paid and organic last click for the same period. If the two columns show identical numbers, the model is not active for that key event and you are seeing last click results. The threshold is 400 conversions for that key event and 20,000 across all events inside the lookback window, and GA4 gives you no warning when you drop below it.

What is the best attribution lookback window for B2B?

Ninety days for almost every B2B lead generation event, because the default 30 day acquisition window systematically under credits early research touches in a long buying cycle. Measure your median time from first touch to conversion and set the window to at least twice that figure. Remember that lookback changes only apply going forward, so the setting takes months to be fully reflected in reports.

Should I add up conversions from GA4, Google Ads, and Meta?

No. Each platform counts conversions it believes it influenced, inside its own window and model, and the same conversion routinely appears in several of them. Summing produces a total well above reality. Use GA4 as the cross channel baseline, each platform's own figures for optimising inside that platform, and your CRM as the actual revenue record.

Is multi touch attribution still worth using?

Yes, but for a narrower job than it used to do. Coverage has fallen to roughly 30 to 60% of touchpoints, so it is a tactical signal for campaign level optimisation rather than a source of truth for budget allocation. Pair it with controlled experiments for causal evidence and mix modeling for portfolio level decisions.

What is the difference between attribution, MMM, and incrementality testing?

Attribution tells you which touchpoints appeared before a conversion. Marketing mix modeling tells you which spend correlates with aggregate outcomes. An incrementality test tells you what would have happened had you not advertised at all. Only the third supports a causal claim. Mature measurement programs run all three and treat disagreements between them as findings rather than errors.