Attribution settings
Attribution decides which touchpoint gets credit — and therefore which channels get budget and which get cut. This module covers how GA4 assigns credit, why Google retired most rule-based models in favor of data-driven, how lookback windows and scopes work, why GA4 and Google Ads never quite agree, and the hard line where attribution ends and incrementality begins.
What you will learn
Why attribution is the most argued-about setting
Attribution decides which marketing touchpoint gets credit for a conversion — and therefore which channels look good, which get budget, and which get cut. No setting in GA4 starts more arguments, because every model tells a different story from the same data. GA4 now defaults to data-driven attribution and has retired most of the old rule-based models. Understanding why is the difference between defending a budget and losing one.
Here is the uncomfortable truth that makes attribution political: it is a way of allocating credit, not a measurement of truth. A customer saw a display ad, clicked a Google search result, opened an email, and bought. Who caused the sale? Every model answers differently, and each answer reshuffles which team gets congratulated and which gets a budget cut. Treat attribution as a useful lens with known distortions, never as ground truth, and you will navigate these fights far better than someone defending their favorite model as ‘correct.’
Attribution is not incrementality.
Keep that line in your pocket for the rest of this module. Every attribution model answers ‘how should we divide credit among the touchpoints a converter saw.’ None answers the question that actually decides budget: ‘what would have happened if we had not run this channel at all.’ We will return to that gap at the end, because it is where attribution’s usefulness ends and incrementality testing begins.
What attribution actually does
Attribution takes a conversion and distributes credit across the touchpoints that preceded it within a lookback window. A single-touch model gives 100% to one touchpoint (the last click, say). A multi-touch model splits credit across several. Data-driven attribution uses your property’s own conversion patterns to decide the split algorithmically. The model you choose does not change what happened — it changes the story the reports tell about it.
Picture a relay race where only the finish matters but four runners contributed. Last-click hands the entire trophy to whoever crossed the line. First-click gives it all to whoever started. Linear splits it evenly. Data-driven studies thousands of past races to estimate how much each leg actually mattered. Same race, four different trophy ceremonies — which is exactly why people argue.
100% to the final touchpoint before conversion. Simple, stable, and the historical default — but it ignores everything that created the demand the last click harvested.
100% to the first touchpoint. Flatters top-of-funnel discovery and ignores everything that closed the sale. Now deprecated in GA4.
Credit split evenly across all touchpoints. Fair-sounding, but pretends every touch mattered the same, which is rarely true. Deprecated in GA4.
Heavy credit to first and last touch, the rest shared. Encodes an opinion that discovery and closing matter most. Deprecated in GA4.
More credit the closer a touch is to conversion. Reasonable for short cycles, misleading for long ones. Deprecated in GA4.
Algorithmically assigns credit based on your property’s actual conversion patterns — which touchpoint sequences truly move the needle. Now GA4’s default and recommended model.
The models GA4 still offers
GA4 deliberately narrowed the field. In 2023 Google deprecated first-click, linear, time-decay, and position-based attribution, leaving data-driven (the default) and last-click as the standard reporting choices. The reasoning: rule-based models encode a guess about how credit should flow, while data-driven infers it from real behavior. Fewer models, less arguing about which arbitrary rule to use.
This surprised and annoyed a lot of practitioners, because favorite models vanished overnight. But the logic holds up: every rule-based model is a hardcoded opinion (‘the first touch matters most,’ ‘recent touches matter most’) dressed as analysis. Data-driven replaces the opinion with an estimate from your own data. You can still see last-click as a baseline, and that pairing — data-driven for decisions, last-click as a familiar sanity check — is what most mature teams settle on.
GA4 is deprecating the first-click, linear, time-decay, and position-based attribution models. Only data-driven attribution and last-click models will be available after this.
Claim: In 2023 Google removed first-click, linear, time-decay, and position-based attribution from GA4 (and Google Ads), leaving data-driven and last-click. Source: Search Engine Land. Context: The change pushed the whole ecosystem toward data-driven attribution as the standard for assigning conversion credit.
Data-driven attribution, demystified
Data-driven attribution (DDA) uses machine learning to assign fractional credit based on how each touchpoint actually changes conversion likelihood in your data. It compares the paths of converters and non-converters and estimates each touchpoint’s real contribution — so credit reflects your customers’ behavior, not a generic rule. It is GA4’s default and recommended model, and for most accounts it is the right choice.
The intuition without the math: DDA effectively asks, ‘when this touchpoint appeared in a journey, how much more likely was a conversion than when it did not?’ Touchpoints that reliably move the needle earn more credit; ones that merely show up along the way earn less. That is a real improvement over hardcoded rules — but it is not magic. It still only sees touchpoints GA4 can observe (cookied, consented, click-based), and it still divides credit among people who converted, which is the limit we close this module on.
I never present a single attribution number. I show the same conversions under data-driven and under last-click, and I point at the difference, not either figure.
Where data-driven gives a channel far more credit than last-click, that channel is an assist engine — it creates demand someone else closes, and last-click was quietly starving it of budget. Where the two agree, the channel is a genuine closer.
That delta is the actual insight. It turns ‘which number is right’ into ‘what role does each channel play,’ which is the conversation worth having.
Lookback windows
The lookback window is how far back GA4 looks for touchpoints to credit before a conversion. GA4 lets you set acquisition-related conversions to a 30-day window and other conversions up to 90 days, with shorter options available. Set it too short and you starve long-consideration channels of credit; too long and you credit touches that had nothing to do with the sale. It should reflect your real buying cycle.
Match the window to reality. A $20 impulse purchase has a buying cycle of minutes — a 90-day window would absurdly credit a blog visit from three months ago. A $40,000 B2B contract has a cycle of months, and a 30-day window would erase most of the journey that created the deal. The default is a compromise; the right move is to know your sales cycle and choose the window that captures it without inventing credit.
Cross-channel vs Google paid channels
GA4 separates two attribution scopes. Cross-channel models distribute credit across all channels — organic, email, direct, paid, social — and are what you use to judge your whole marketing mix. Google paid channels (the ‘Ads-preferred’ setting) bias toward giving Google Ads the last click when it appears, which aligns GA4 with what Ads reports. Know which scope a report uses before comparing channels, or you will compare apples to oranges.
This trips up nearly everyone reconciling GA4 against Google Ads. If GA4 is set to a cross-channel model and Ads uses its own, the same conversion can be credited differently in each tool — not because either is broken, but because they are answering different questions. For an honest view of your full mix, use cross-channel. For aligning to Google Ads’ own accounting, understand the Google-paid-channels scope. The professional skill is knowing which lens is in front of you.
Conversion paths and model comparison
Two reports in the Advertising section make attribution concrete. The conversion paths report shows the actual sequences of touchpoints leading to conversions — early, mid, and late roles each channel plays. The model comparison view (where available) shows how credit shifts between models. Together they move you past arguing about a single number toward understanding the journey’s real shape.
The conversion-paths report is the antidote to attribution dogma. Instead of trusting one model’s verdict, you watch what really happens: maybe paid social almost always appears first, email almost always appears last, and organic search shows up everywhere. That picture tells you each channel’s job in the funnel — assist, closer, or both — which is a far richer basis for budget than any single attributed number. Read the paths before you read the model.
Last-click makes upper-funnel channels look worthless because they rarely deliver the final tap. Teams cut them, and conversions mysteriously sag a month later.
Before any cut, I pull the conversion-paths report and compute, per channel, how often it appears as an assist versus a last touch. A channel that assists far more than it closes is creating demand someone else harvests — an assist ratio above one is a flag, not a candidate for the chopping block.
It turns ‘this channel has low ROAS’ into ‘this channel feeds three others,’ which is the conversation that protects good budget.
Why GA4 and Google Ads disagree
GA4 and Google Ads will almost never show identical conversion numbers, and that is expected, not a bug. They differ on attribution model, conversion-counting rules (Ads counts conversions at click-time; GA4 at event-time), time zones, lookback windows, and how each handles modeling and consent. Your job is not to force them to match — it is to explain calmly why they differ and which tool to trust for which question.
Walking a client through this builds more credibility than any dashboard. Google Ads attributes a conversion to the time and channel of the ad click; GA4 records the conversion when the event happens and may credit a different channel under its model. Add different time zones and windows and a 10–20% gap is normal. Use Google Ads’ numbers for in-platform bidding decisions and GA4’s cross-channel view for whole-mix strategy, and stop trying to reconcile two tools that were built to answer different questions.
Every GA4-vs-Google-Ads fight I’ve seen comes from never deciding, in advance, which tool governs which decision. So the same conversion gap gets re-litigated every monthly review.
I write it down once: Google Ads’ own conversions govern in-platform bidding and budget; GA4’s cross-channel view governs whole-mix strategy; incrementality tests govern ‘is this channel worth keeping.’ One page, signed off.
After that, a 15% gap between the tools is a footnote, not a crisis — because nobody is asking the wrong tool the wrong question anymore.
- Which tool should I trust for conversion counts?
- For optimizing within Google Ads (bidding, budgets), trust Google Ads’ own conversions. For understanding your whole marketing mix across all channels, use GA4’s cross-channel attribution. Neither is ‘more correct’ — they answer different questions.
- Why does Google Ads show more conversions than GA4?
- Common causes: Ads counts at click-time (and can count multiple conversions per click) while GA4 counts at event-time; different attribution models; time-zone and lookback differences; and different handling of modeling and consent.
- Should I make the two tools match?
- No. Document the structural reasons they differ, pick the right tool per decision, and monitor the gap for stability rather than chasing an impossible exact match.
The honest limit: attribution is not incrementality
Every attribution model shares one blind spot: it only divides credit among people who converted, and only across touchpoints it can see. It cannot tell you whether a channel caused conversions that would not have happened otherwise. That question — incrementality — is answered by controlled experiments (holdout tests, geo experiments), not by any attribution setting. Mistaking attribution for incrementality is the most expensive error in marketing analytics.
The classic trap: a brand-search campaign gets huge last-click and even data-driven credit, so it looks like the best channel — but most of those people would have searched the brand and bought anyway. Attribution happily credits the campaign; an incrementality test (turn it off for a holdout region and watch) often reveals it added far less than it claimed. GA4’s attribution is the right tool for understanding journeys and allocating reported credit. It is the wrong tool for proving causation. Use it for what it is, and reach for experiments when the question is ‘did this actually cause new revenue.’
Brand search always tops attribution — data-driven included — because the people who search your name were going to convert anyway. The model credits the campaign; it cannot see the counterfactual.
So before scaling brand search on its attributed glory, I pause it in one matched region for a week and watch whether conversions there actually fall. Often they barely move, which means most of that ‘credit’ was harvested demand, not created demand.
One quiet region for one week answers a question no attribution model can: what would have happened anyway.
Where attribution goes wrong
Attribution mistakes are usually conceptual, not technical: treating a model as truth, using a lookback window that ignores the buying cycle, comparing reports built on different scopes, expecting GA4 and Ads to match, and — the big one — mistaking attributed credit for proof of incrementality. Each leads to confidently wrong budget decisions, which is the most expensive kind.
Defending last-click (or any single model) as ‘correct’ ignores that every model is a lens with built-in distortions.
A 7-day window on a 3-month B2B sale erases the journey; a 90-day window on impulse buys invents credit.
Comparing a cross-channel report to a Google-paid-channels report and concluding a channel changed is a phantom finding.
Chasing an exact reconciliation between two tools built to answer different questions wastes days.
Assuming an attributed channel caused the revenue leads to overspending on channels that harvest demand they didn’t create.
Your attribution checklist
Sound attribution practice is captured here. Tick what is genuinely true of how this property’s attribution is configured and used.
Prove it. Earn your passcode.
Ten questions, CASE method (Context · Analysis · Strategy · Execution). Pass at 90% to unlock this module’s completion passcode — retake as many times as you like.