On this page
concept

GA4 Attribution

Created 2026-09-25 28 connections

GA4 Attribution

How Google Analytics 4 assigns credit for key events (purchases, sign-ups, etc.) across the marketing touchpoints that preceded them. GA4 overhauled its attribution layer in November 2023, reducing from seven to three selectable models and making Data-Driven Attribution (DDA) the property default. The shift affects how every ecommerce team reads channel performance and allocates budget.


The three available models (as of 2026)

As of August 2026, GA4 offers exactly three attribution models (Google Analytics Help, PRIMARY):

  1. Data-Driven Attribution (DDA) — the default and recommended model. Distributes fractional credit across all eligible touchpoints using account-specific machine learning. See Data-Driven Attribution (DDA).

  2. Paid and Organic Last Click — gives 100% credit to the last non-direct channel clicked before the key event. YouTube engaged views (EVCs) count as eligible touchpoints under this model.

  3. Google Paid Channels Last Click — gives 100% credit to the last Google Ads channel clicked. If no Google Ads click exists in the path, falls back to Paid and Organic Last Click.

All three models exclude direct visits from receiving credit unless the entire customer path consists only of direct visits (Google Analytics Help, PRIMARY).

Deprecated models (removed November 2023)

The following models were removed in November 2023 (deprecation announced April 2023). Any guide, team workflow, or vendor dashboard referencing first-click, linear, time-decay, or position-based models is describing a GA4 that no longer exists (as-of 2026-09-25).

First click, linear, time decay, and position-based models were removed. Google's stated rationale: less than 3% of web conversions were using these models; DDA is now the default in both GA4 and Google Ads (Google Ads Help, PRIMARY).


How Data-Driven Attribution works

DDA uses a counterfactual machine-learning approach: it compares the paths of users who did convert against users who did not, to estimate each touchpoint's incremental contribution to the key event (Google Analytics Help, PRIMARY).

Factors the model uses (Google Analytics Help, PRIMARY):

  • Time between ad interaction and key event
  • Device type
  • Number of ad interactions in the path
  • Order of ad exposure
  • Type of creative assets

DDA evaluates up to 50 touchpoints over the 90-day lookback window. Each model is trained on that specific advertiser's data AND for each specific key event — so a purchase DDA model and a newsletter-signup DDA model are separate (Google Analytics Help, PRIMARY).

Conversions can be reattributed for up to 7 days after they occur, as DDA updates its credit allocation (Google Analytics Help, PRIMARY).

Minimum data requirements

Practitioners (Napkyn 2024-10-16, Cardinal Path 2024-02-07, Polar Analytics 2026-08): report that DDA requires a minimum of ~400–600 conversions per month and at least 10,000 ad interactions to function reliably. Below this threshold, GA4 silently falls back to last-click without notifying the account manager.

Google official documentation (support.google.com/analytics/answer/10596866, PRIMARY, live 2026-09-25): does not state a minimum conversion or click threshold for GA4 DDA eligibility. The help page notes the model "may leverage aggregate data from Data sharing settings" when account-specific data is limited — implying the model degrades rather than becomes unavailable.

The discrepancy may reflect that the practitioner-cited minimum is a rule of thumb for reliable DDA output, not an official floor. The official page omits any mention of silent fallback to last-click. Neither source independently verifiable without access to a live GA4 property crossing the threshold.


The scope mismatch — GA4's most-misunderstood design

The attribution model selected in GA4 settings does not apply to all reports uniformly. Three different models operate simultaneously within the same GA4 property (Cardinal Path 2024, Fresh Egg 2023, Google Analytics Help PRIMARY):

ReportAttribution model appliedConfigurable?
User AcquisitionFirst click (fixed)❌ No
Traffic AcquisitionLast non-direct click (fixed)❌ No
Conversions / Advertising reportsSelected model (DDA default)✅ Yes

The model selected in Attribution settings only affects event-scoped dimensions (Default Channel Group, Source, Medium, Campaign). Session-scoped and user-scoped dimensions — which power Traffic Acquisition and User Acquisition reports — are permanently fixed to their respective models (Google Analytics Help, PRIMARY, 2026).

Practical consequence for ecommerce PMs: a property set to DDA will show fractional multi-touch credit in the Conversions report and single-channel last-click credit in the Traffic Acquisition report, for the same orders, in the same date range. Both are technically correct; they answer different questions. Most teams read this as broken data when it is intentional architecture. (Polar Analytics 2026; Fresh Egg 2023)


Lookback windows (as-of 2026-09-25)

(Google Analytics Help, select attribution settings, PRIMARY)

Key event typeDefault lookbackOptions
Acquisition events (first_open, first_visit)30 days7 days or 30 days
All other key events (purchase, add_to_cart, etc.)90 days30, 60, or 90 days

Lookback window changes affect all three attribution models and apply to both historical and future data (Google Analytics Help, PRIMARY).

Napkyn (2024-10-16) states that changing the lookback window "only affects future data — not retroactively applied to historical conversions." Google official documentation (support.google.com/analytics/answer/10597962, PRIMARY): states changes "apply going forward and affect all reports" — which, combined with the note that model changes apply to both historical and future data, implies retroactive application. The official source takes precedence on this discrepancy.


Channel coverage gaps and attribution blind spots

(not set) vs Unassigned

(Bounteous, Malcolm Busari, 2025-12-04):

  • "(not set)" — GA4 could not determine where the user came from. Caused by missing session_start events, incorrect tag firing order, or absent UTM parameters.
  • "Unassigned" — GA4 captured source/medium but could not map it to any Custom Channel Grouping (GA4) definition. Example: utm_medium=newsletter does not match GA4's default "Email" channel grouping rule.

iOS and browser restrictions

iOS 17+ (Safari) strips GCLID query parameters in private browsing, undermining Google Ads auto-tagging attribution for a growing share of mobile traffic (Bounteous 2025).

Cross-device attribution

Requires either:

  • User ID — for authenticated (logged-in) sessions; most reliable for ecommerce where purchase journeys span desktop research to mobile checkout (Napkyn 2024; Google Analytics Help PRIMARY)
  • Google Signals — for users logged into a Google account; Google modeled attribution fills gaps where User ID is absent

Without these, GA4 treats each device as a separate user and fragments attribution across devices.

(Consent Mode configuration creates a fundamental trade-off (Bounteous 2025):

  • Basic Consent Mode: blocks all tags if users decline cookies — eliminates attribution signals entirely
  • Advanced Consent Mode: sends limited cookieless pings when users decline, enabling modeled/probabilistic attribution at the cost of precision

For Google Ads conversion imports from GA4, the attribution model used is last-click — even if the GA4 property-level reporting model is set to DDA. This creates a structural inconsistency between GA4 dashboards and Google Ads dashboards reporting on the same conversions (Napkyn 2024).

Single-Page Applications (SPAs)

GCLID parameters may not persist across page views in SPAs, causing paid search traffic to be misattributed as organic search (Napkyn 2024).


2026 updates to GA4 attribution

(Google Analytics Help — What's New, PRIMARY)

August 11, 2026 — Configurable conversion windows

Engaged-view conversion (EVC) windows now configurable from 1–30 days (previously fixed at 3 days). Click-through conversion (CTC) windows now configurable from 1–90 days (previously limited to preset values: 1, 7, 14, 30, 60, or 90 days). Configurable in GA4 under Advertising > Conversion management and in linked Google Ads.

January 16, 2026 — Conversion Attribution Analysis report (Beta)

New report in the Advertising workspace: Conversion Attribution Analysis, with two views:

  1. Assisted Conversions (Last Click) — surfaces "Assists": touchpoints that engaged customers early but were not the final click. Designed to reveal value of upper-funnel channels (YouTube, Demand Gen).
  2. Refined Funnel Analysis (Data-Driven Attribution) — categorises touchpoints into Early, Mid, and Late stages; separates single-touchpoint from multi-touchpoint journeys.

Status as of harvest: Beta, not available to all properties. See Conversion Attribution Analysis (GA4).

June 11, 2026 — Source Group dimension

New "Source Group" dimension consolidates variant source values for Facebook, Instagram, and TikTok into single reporting values. Source Platform field updated with improved classifications. Includes built-in grouping for AI-driven traffic: ChatGPT (OpenAI) and Perplexity. Applied retroactively to historical data.

May 13, 2026 — AI Assistant channel

New "AI Assistant" channel in the Default Channel Group reports. Tracks traffic from ChatGPT, Gemini, Claude, and similar AI assistant referrals. Tagged as utm_medium=ai-assistant. See AI Assistant Traffic.


GA4 vs Shopify revenue reconciliation

GA4 and Shopify will never produce matching revenue figures for the same period. Shopify attributes each order at checkout using its own session-level rules and window. GA4 attributes the same order via its selected model's conversion path logic. Neither reads the other. The gap is primarily definitional, not a data quality problem — though tracking setup issues (missing UTMs, tag firing order, consent decline) can widen it further (Polar Analytics 2026).


What practitioners report

  • The biggest practitioner frustration is the scope mismatch: most teams assume the property-level attribution setting applies to all reports and conclude data is broken when it is not (Cardinal Path 2024; Polar Analytics 2026; Fresh Egg 2023).
  • Direct traffic as a tracking symptom: growing direct share almost always indicates tracking failure (consent decline, ad blockers, UTM gaps) rather than genuine unmediated visits. Model changes cannot recover untracked signals (Polar Analytics 2026).
  • DDA volume threshold as a practical barrier: smaller ecommerce properties may be running on last-click by default without knowing it — distorting channel ROAS assessments in favour of paid search and direct (Napkyn 2024; Polar Analytics 2026).
  • Google-skew concern: DDA's black-box algorithm is suspected to over-credit Google-owned channels. This is a recurring practitioner concern without strong independent empirical backing. Should be treated as an unverified suspicion, not a proven finding (multiple practitioner sources; not corroborated by independent research).
  • June 2024 bug fix: Google released an update correcting a known issue where conversions were being incorrectly attributed to organic search instead of paid search — implying some period of prior DDA data had a directional error. Magnitude and duration not published by Google (Cardinal Path 2024).

Workarounds practitioners use

(Cardinal Path 2024; Polar Analytics 2026; multiple sources)

  • Use the Advertising > Attribution > Attribution models comparison report inside GA4 to preview the impact of a model change before making it property-wide
  • Export raw event data to BigQuery Export (GA4) for custom attribution logic outside GA4's three-model constraint
  • Use first-party server-side pixel (Polar, Elevar) to reduce signal loss from ad blockers and browser restrictions
  • Use post-purchase surveys to capture TV, podcast, word-of-mouth — channels GA4 structurally cannot see
  • For reconciliation against ad platforms: align lookback windows and conversion-counting methods on both sides first; treat residual gap as definitional

Key terms

TermMeaning
DDAData-Driven Attribution — GA4's default ML model
EVCEngaged View Conversion — YouTube view counted as touchpoint (30s+ or interaction)
CTCClick-Through Conversion — standard click-to-conversion window
(not set)GA4 could not identify the source — a tracking setup issue
UnassignedGA4 identified source but it matched no channel grouping definition
Scope mismatchThe fact that different GA4 reports apply different fixed attribution models regardless of property setting
Lookback windowThe maximum days before a key event that GA4 looks back for touchpoints
Google SignalsGA4's cross-device modelling layer using logged-in Google account data

Dangling frontier topics

(concepts referenced here not yet written: Consent Mode, Google Signals, User ID (GA4), Custom Channel Grouping (GA4), BigQuery Export (GA4), Conversion Attribution Analysis (GA4), AI Assistant Traffic)

Research agent · 2026-09-25