Rules and ethics
Part of Marketing attribution rules and ethics: UK checks
A versioned disclosure process for attribution claims, from label to withdrawal
Build a versioned England attribution disclosure process for methods, limits, commercial ties, placement, corrections and withdrawals before a claim goes live.
A disclosure policy should control how an attribution statement is assembled, approved, displayed, corrected and withdrawn. It is not a generic notice and it cannot make weak evidence or unlawful processing acceptable. This procedure was researched on 6 September 2026.
Named UK advertising, consumer-law, privacy, PECR, statistics and commercial reviewers must approve its application; the text is not legal advice.
What to take away
- A disclosure policy must control assembly, approval, display, correction and withdrawal of attribution claims.
- Classify claims as observed, platform-attributed, modelled or causally estimated and record model details and uncertainty.
- Separate sign-offs are needed for evidence, advertising, consumer protection, data protection, PECR, accessibility, IP and commercial relationships.
- Set re-review triggers for model changes, new data, revised audience, altered visual, supplier relationship, complaint, legal update or expiry.
- If evidence or specialist decision is missing, hold publication rather than issue a softer disclosure.
Create a versioned disclosure record
Give every proposed claim an ID. Record the owner, intended audience, channel, publication date, evidence file, source geography and period. Then classify the result as observed, platform-attributed, modelled or causally estimated. Capture the model and version, lookback window, eligible population, outcome, exclusions, missing data and uncertainty.
Keep a separate field for a supplier, affiliate, sponsor, agency or other commercial relationship. State whether tools or data were supplied, whether funding influenced scope, and what remained editorially independent. Do not turn absence of declared funding into a claim of universal independence.
Worked example: a completed disclosure record. The figures are illustrative.
| Field | Example |
|---|---|
| Claim ID | MG-2026-014 |
| Classification | Platform-attributed |
| Model | Platform last-click model, version 4.2 |
| Window | 7-day click, 1-day view |
| Uncertainty | No confidence interval; platform does not report incrementality |
| Evidence file | MG-2026-014-evidence-v3.pdf |
| Owner | Head of analytics |
| Label wording | "Platform-attributed: 1,240 purchases credited to [platform] under its 7-day click and 1-day view rule. This is platform credit, not confirmed incremental sales." |
Draft for the likely impression
Write the headline, chart labels, body qualification and call to action together. ASA and CAP's misleading-advertising guidance, updated 12 March 2026, asks whether the marketing communication is likely to mislead. Its substantiation guidance explains that documentary evidence for objective claims should be held before publication.
A platform-attributed conversion should name the platform coverage and credit rule. A modelled estimate should state model scope and uncertainty. A causal lift statement needs a credible counterfactual and bounded population. Do not use a broad success headline with a remote caveat.
For example, the ASA banned a Ryanair ad in 2020 over a 'lowest CO2 emissions' claim, finding the evidence did not support the comparison.
Choose a label and placement
The CAP Code's recognition rules require marketing communications and commercial intent to be identifiable. Select clear wording for the actual relationship, then preview it where the audience encounters the claim. Record screen size, format, prominence and the reviewer who saw the final rendering. A label hidden behind another click may not create the intended understanding.
Sample label wording by claim class. The figures are illustrative.
| Claim class | Sample label wording |
|---|---|
| Observed | "Observed: 1,240 purchases recorded by our analytics between 1 and 30 June 2025. This counts tracked checkout events and does not show what caused them." |
| Platform-attributed | "Platform-attributed: 1,240 purchases credited to [platform] under its 7-day click and 1-day view rule. This is platform credit, not confirmed incremental sales." |
| Modelled | "Modelled estimate: 1,240 purchases, with an uncertainty range of 900 to 1,600. The model uses [listed inputs] and may overstate or understate the true effect." |
| Causally estimated | "Causal estimate: a 5.2% lift in purchases, based on a 10% holdout of the eligible audience, with a range of 2.1% to 8.3%. This estimates incremental effect under the test conditions." |
Privacy information is a different layer. Link the claim record to the applicable notice version and data-flow map, but do not treat an advertising label as a privacy notice. Likewise, a tracking choice must be reviewed under the ICO's final storage and access guidance. Disclosure does not replace a consent or exception analysis where one is required.
Add approval and expiry controls
Use separate sign-offs for evidence quality, advertising, consumer protection, data protection, PECR, accessibility, IP and commercial relationships. Record approve, reject or unresolved, with conditions and expiry. No reviewer may clear another speciality by implication.
Set triggers for re-review: model or window changes, new data source, revised audience, altered visual, supplier relationship, material complaint, legal update or evidence expiry. The CMA's unfair commercial practices guidance is relevant to consumer-facing omissions and representations from 6 April 2025; a qualified lawyer must apply it to the finished journey.
Monitor, correct and withdraw
Assign one contact to receive challenges and one decision-maker to pause publication. Preserve the original statement, evidence version, complaint, assessment and corrected wording. Publish a correction where the original audience needs it, rather than silently editing a record that has already circulated.
The policy register should show status, owner, last review, next review and withdrawal route. On publication day, reopen the cited regulator pages and inspect the final format. If the evidence or specialist decision is missing, the operational outcome is hold, not a softer disclosure.
Before you act
- Give every proposed claim an ID and record owner, audience, channel and evidence.
- Classify the claim and capture model, window, population and uncertainty.
- Draft headline, labels, qualification and call to action together.
- Choose clear label wording and preview it where the audience sees it.
- Obtain separate sign-offs for each speciality and set expiry.
- Assign a challenge contact and a decision-maker to pause publication.
Common questions
What should a versioned disclosure record contain for each proposed claim?
A record should also show the version number, approval date, expiry date and the reviewer who signed each speciality. Store it with the evidence file and the final rendered label. If a supplier provides the data, add the contract or data-sharing terms. That lets an auditor reconstruct what was published and why.
How should a platform-attributed conversion be described?
Check the platform's coverage and credit rule before drafting. State the lookback window, the event counted and whether the platform includes view-through credit. Compare the platform figure with an independent source where possible. If the two differ, explain the gap or label the figure as platform credit only.
What triggers a re-review of an approved disclosure?
Add a change log entry for each trigger and name the person who must decide within a set period. A model refresh, new attribution window, revised audience or supplier change can alter the claim. A complaint or regulator query should create an immediate review, not wait for the next scheduled date.
