Marketing attribution models in 2026: how each one credits a sale and which to use

An attribution model is a rule for distributing credit for a conversion across the touchpoints that came before it. A customer might click a TikTok ad on Monday, see an Instagram Reel on Wednesday and search your brand on Google on Friday before buying. The model decides whether TikTok, Meta or Google gets the sale, or how much of it each gets.

Multi touch attribution (MTA) is the umbrella term for models that split credit across several touchpoints instead of giving everything to one. Linear, time decay, position-based and data-driven are all multi touch. Last click and first click are single touch.

A simple example makes the difference visible. One €100 order came after four touches: a TikTok ad, a Meta ad, a newsletter click and a Google brand search.

The time decay column is an example split, the exact weights depend on the half-life a tool uses. Data-driven attribution has no fixed split. It is calculated from your account's own conversion paths.

Google retired four rule-based models in 2023: first click, linear, time decay and position-based. Google Ads now offers data-driven and last click, and data-driven is the default for most conversion actions (Google Ads Help). Conversion actions that used the retired models were moved to data-driven attribution. When Google announced the change in April 2023, it said fewer than 3% of conversion actions in Google Ads used those four models (Search Engine Journal, 10 April 2023).

GA4 removed the same four models by November 2023 (Google Analytics Help). The reporting attribution model in GA4 can be data-driven, paid and organic last click, or Google paid channels last click. Lookback windows are 30 days by default for acquisition key events (7 days optional) and 90 days for all other key events (30 or 60 days optional).

If you want to see first touch in 2026, you need your own data: UTM parameters stored at first visit, a first-party pixel or a CRM that records the lead source. Google's own measurement has moved in another direction. At Google Marketing Live on 20 May 2026 Google announced that Meridian, its open-source marketing mix model, is coming into Google Analytics 360 (Google, 20 May 2026).

Meta uses its own attribution settings, and in 2026 the default for new ad sets that optimise for website conversions is 7-day click, 1-day engage-through and 1-day view. Meta's help page lists these options without naming a default; the default is documented by practitioners such as Jon Loomer (10 March 2026). Two changes shaped this.

• On 12 January 2026 Meta stopped returning 7-day view and 28-day view data in the Ads Insights API, as announced on its developer blog in October 2025. Click windows of 1, 7 and 28 days, 1-day engaged view and 1-day view remained available there (Meta for Developers, 16 October 2025). Ads Manager itself offers 1-day or 7-day click and 1-day view for website conversions (Meta Business Help Center).
• On 3 March 2026 Meta announced that click-through attribution for website and in-store conversions counts only link clicks, rolling out to advertisers at different times from later that month. Likes, shares and saves moved to engage-through attribution, the new name for engaged-view attribution, and the video threshold dropped from 10 to 5 seconds (Meta, 3 March 2026).

Meta also offers incremental attribution, which estimates the conversions an ad caused rather than all conversions that followed it. It can be used for reporting or as an optimisation setting (Meta Business Help Center, Social Media Today, September 2025). It is a model trained on Meta's lift studies, so it is closer to causation than click windows, but it is still Meta grading its own work.

Each platform only sees its own touchpoints and credits itself whenever it touched the buyer. Meta cannot see the Google search. Google's data-driven model splits credit between Google ads, not between Google and Meta. GA4 sees website sessions but not ad views. Add the three reports and you count the same order several times.

We see this in live data. A connected car app has run its advertising with the same agent system behind Marketing AIgency since 14 September 2026, across 9 markets. In Germany in September 2026, Meta, Google and TikTok together claimed 2.2 times the revenue the company actually made. Google alone reported €31.3k against €23.4k in real revenue. The full breakdown and the formula for your own over-counting factor are in our guide to true ROAS.

Research has long pointed the same way. A study of 15 large advertising experiments at Facebook found that common observational measurement methods often failed to match the effect measured in randomised tests (Gordon et al., Marketing Science, 2019).

Attribution tracks individual paths. Marketing mix modelling (MMM) and incrementality tests estimate what your spend caused. They answer different questions, so strong setups use them together.

• Conversion lift: Meta splits people into exposed and holdout groups (Meta). Google's user-based Conversion Lift needs a minimum budget of $5,000 and at least 1,000 observed conversions (Google Ads Help).
• Geo tests: switch ads off in some regions and compare, for example with Meta's open-source GeoLift (GitHub).
• MMM: open-source models such as Google's Meridian (GitHub) and Meta's Robyn (GitHub) estimate channel contribution from spend and revenue over time.

Pick one source of truth for revenue, use platform attribution for day-to-day decisions inside each platform, and test before large budget shifts. The model setting in Google or Meta matters less than this discipline.

1. Define the company number. Net revenue from your shop, or qualified leads and won deals from your CRM. This is what budgets are judged against.
2. Leave Google on data-driven and Meta on its default, unless you have a reason to change. Consistency beats the theoretically best model.
3. Track first-party data. UTM parameters on every ad, a first-party pixel on your own domain and server-side events such as Meta's Conversions API, set up as described in our Conversions API guide.
4. Ask customers. A one-question post-purchase survey (how did you hear about us?) captures podcasts, word of mouth and TikToks watched on another phone.
5. Compare weekly. Put each platform's reported revenue next to the company number and track the over-counting factor.
6. Test before big moves. Before shifting a large share of budget, run a lift or geo test on the channel you plan to cut or grow.

Running Google's Performance Max? The same logic applies there, see how to measure Performance Max.

Marketing AIgency's agents steer on the company number, real revenue or real leads from Shopify, a CRM, a Google Sheet or a first-party pixel, instead of platform ROAS. Every suggestion goes into an approval queue with the data and reasoning behind it. The Attribution add-on adds a first-party pixel, a post-purchase survey, purchase data sent back to Meta through the Conversions API and geo-lift tests, so you can see which channel really drives sales and give Meta better signals. Plans have fixed monthly prices with no percentage of ad spend, see pricing.