Four attribution methods for fitness creators and the limits of each

Fitness creators attribution methods answer different questions. Tagged links describe a recorded route, promotion codes identify eligible redemptions, platform models allocate credit and controlled evaluation estimates a causal effect. None is best for every decision, and the first three do not prove a health benefit.

This unranked desk comparison was researched on 5 September 2026. Inclusion required a reproducible rule, a current authoritative or first-party description and relevance to an England creator operation. It excludes vendors, prices and unsupported benchmark claims. No accounts or campaigns were tested.

What to take away

Like-for-like criteria

Each approach is compared by population, numerator, denominator, unit, period, timezone, inclusions, exclusions, source, uncertainty, owner and update trigger. The common business question is: what eligible action can reasonably be linked to a named release during a declared period? Health causality is assessed separately.

Tagged landing sessions

Population: unique eligible landing sessions during the campaign window. Numerator: sessions carrying the agreed campaign values. Denominator: all eligible landing sessions. Report a percentage in the analytics property's disclosed timezone, with staff, test traffic, duplicates and sessions lacking required consent excluded. Google documents that UTM parameters populate source, medium and campaign reporting.

The analytics owner records taxonomy, filters and extraction time. Consent loss, cross-device journeys and inconsistent case create uncertainty. A tagging or consent change triggers a series break. Best use: route description. Limit: it does not show whether the visit would otherwise have happened.

Promotion-code redemption

Population: eligible settled orders in the declared sales period. Numerator: those using the unique creator code; denominator: all eligible settled orders, or report a count if the publisher cannot lawfully access the full order base. Unit, currency, Europe/London cutoff, refunds, cancellations, staff orders and code-sharing rules must be explicit.

The merchant's reconciled order ledger is the source and the commercial owner signs it off. Leakage, forgotten codes and offline influence create uncertainty. Refund reconciliation triggers revision. Best use: an auditable direct-response record. Limit: it assigns rather than proves incremental credit.

Platform-modelled credit

Population: key events eligible under the property's selected model and lookback window. Numerator: credit assigned to the named content or channel; denominator: all eligible modelled key-event credit. Report an event count or currency, never mix the two. The account timezone, included channels, direct-traffic treatment and model version belong in the note.

Google describes attribution as assigning credit to touchpoints and documents data-driven and last-click approaches. The platform owner captures settings and extraction date. Reattribution, identity gaps and an opaque or changing model are limitations. A model or window change starts a new series. Best use: comparing credit rules inside the same property, not claiming independent causation.

Controlled impact evaluation

Population: people eligible under a pre-registered England evaluation protocol. Numerator and denominator depend on the chosen validated outcome; the estimate compares intervention and credible counterfactual groups over specified time points. Report the effect unit, interval and analysis population, with exclusions, missing-data handling and departures from plan. Use the protocol's stated timezone.

The evaluation lead owns analysis and a registered health professional owns clinical interpretation. The government's impact-evaluation quality guide explains that outcome monitoring alone cannot identify cause. NICE also recommends defined objectives, baseline and valid process and outcome measures for behaviour-change evaluation. Attrition, spillover, confounding and measurement error remain uncertainties. Safety signals or protocol failure can stop the study.

Choose by claim strength

Use tagged sessions to diagnose routes, codes to reconcile eligible redemptions and modelled credit to understand a platform's allocation. Use a suitably designed evaluation only when the decision requires an incremental commercial or health effect. Keep all four results separately labelled. If the evidence stops at attribution, the conclusion must stop there too.

Tracking may involve storage or access on a user's device. The ICO's final storage and access technologies guidance requires a context-specific privacy assessment. A lawful data flow is a prerequisite, not another attribution method.

Before you act

Common questions

What does a tagged landing session actually show?

It shows a recorded route: sessions carrying the agreed campaign values as a percentage of all eligible landing sessions, in the analytics property's disclosed timezone. It does not show whether the visit would otherwise have happened, so it describes a route rather than proving an effect.

When is a controlled impact evaluation needed?

Use a suitably designed evaluation only when the decision requires an incremental commercial or health effect. It compares intervention and credible counterfactual groups under a pre-registered England protocol, with a registered health professional owning clinical interpretation. Outcome monitoring alone cannot identify cause.

What limits platform-modelled credit?

Reattribution, identity gaps and an opaque or changing model are limitations. A model or window change starts a new series. It is best used for comparing credit rules inside the same property, not for claiming independent causation, and the account timezone and model version belong in the note.