Rules and ethics
Part of Social listening rules and ethics: keeping checks in their proper lanes
Social listening disclosure policy: what readers are told
A social listening disclosure policy for report readers in England: the method note, what the numbers count, automated labels, paid quotes and corrections.
A published listening report owes its readers more than a chart. They need to know what was counted, what a machine labelled, who paid for a quoted post and how to challenge a quotation. Each of those is a separate disclosure, and none of them replaces the others.
This is the reader-facing side of a disclosure policy. The privacy notice for people whose personal data is used is its own layer. Build each reader layer from the data flow and the intended publication, and route it to reviewers with relevant privacy and advertising expertise. No template can guarantee that a particular disclosure is sufficient.
What to take away
- A reader-facing disclosure is several short statements placed next to the claim they qualify, not one badge.
- A method note must give collection dates, sources, counting rules, classification methods and limitations.
- Automated labels need a name, a version, a validation method and known error types.
- Payment, gifting or commission behind a quoted post must stay visible to readers.
- Quoted or classified people need a route to challenge the text and a record of the correction.
Put a method note beside every finding
A reader should be able to judge what a chart can show without asking the analyst. A method note beside the finding supplies that context: collection dates, sources, counting rules, classification methods and limitations. A note hidden in an appendix leaves the headline number alone.
Do not call accessible conversation representative of England. Avoid publishing search syntax if that would expose security controls or enable harmful targeting. Give the useful boundary without revealing personal or confidential material, so the reader sees what the finding covers and what it cannot cover.
A worked format uses placeholders rather than monitoring results.
Collection period: [start date] to [end date], [time zone]. Query: [version]. Sources: [authorised sources]. Unit counted: [unit]. Deduplication: [rule]. Exclusions: [rule]. Geography: [England inclusion rule and unknown quantity]. Classification: [methods]. Numerator and denominator: [values]. Limitations: [known gaps]. Updated: [date].
Fill in each field for the specific finding and remove any field that does not apply.
Say what the numbers count
Every number needs its unit and its base. Say what one count stands for, which items were removed as duplicates and what a percentage is a share of. Counting errors that survive into a published chart are covered under measurement mistakes.
| Field | What to state | What the reader can then judge |
|---|---|---|
| Collection period | Start, end, time zone and query version | Whether two charts cover the same window and the same search |
| Sources | Authorised sources and inaccessible spaces | Which conversations the finding could never see |
| Unit counted | The thing one count stands for | Whether totals are comparable |
| Deduplication and exclusions | The rule applied and what it removed | Whether repeats inflate the result |
| Geography | The England inclusion rule and the quantity left geographically unknown | How much of the finding is placed in England |
| Numerator and denominator | Both values, and any missing data | What a share is a share of |
| Limitations | Known gaps, uncertainty and update date | How far to trust the conclusion |
Label automated work and its limits
Name every automated topic, sentiment or risk label and say whether a person checked the original context before action. Give its version, validation method and known error types. Where validation is too weak, label the output exploratory or withhold it. Never publish an invented accuracy score.
Each of the AI applications used in listening has its own test, so say which job the label does. If an output feeds a decision with legal or similarly significant effects on a person, assess whether that decision is solely automated and which safeguards apply.
The Data (Use and Access) Act 2025 (DUAA) changes the UK GDPR regime for certain solely automated decisions. Check the amended UK GDPR provisions, including Articles 22A to 22D, and the ICO's guidance. The ICO's current automated-decision summary describes information, representation, intervention and contest safeguards for covered decisions.
Keep commercial relationships visible
If a quoted post, review or creator relationship involved payment, gifting, commission or another incentive, keep that fact visible next to the quote. A general corporate interests page may be insufficient if it is remote from the claim.
The CMA's guidance for brands on social media endorsements says any form of incentive or reward, including money, commissions, discounts and gifts of any products, is payment.
For reviews, use the CMA's Online reviews and endorsements: guidance for businesses. For influencer advertising, see the ASA and CAP Influencers' guide to making clear that ads are ads, plus CAP Code rules 2.1 and 2.3 on identifying marketing communications and making commercial intent clear where it is not obvious.
Illustrative label wording: "Ad: [creator] was paid by, received a gift from, or earned commission from [brand] for this content." Use only the accurate relationship and place the label where readers can see it.
Disclose supplier sponsorship, affiliate interests and editorial conflicts attached to the listening report itself. The advertising rules cover what happens when listening evidence is then used inside an advert.
Give quoted people a way to challenge
Decide first whether aggregation or paraphrase can meet the purpose, because quoting is the harder path. Where text or imagery is reproduced, document the platform permission and the IP analysis. GOV.UK's copyright guide explains licence and exception routes; attribution alone is not a universal licence.
Provide a route to challenge a quotation, a location label or an inferred category. Log the original evidence, the review, the outcome, the correction date and the downstream reports that were updated. Remove identifiers from public examples when they add no decision value.
The ICO says data obtained from public sources may still engage the right to be informed, so a quoted person may be owed privacy information as well as a correction route. The legal checks themselves sit in the wider compliance review.
Keep each disclosure current
Update the policy whenever purpose, source, model, recipient, retention or commercial relationship changes, and stop the activity when disclosure cannot be made accurately. A disclosure that was true at launch can be wrong a quarter later.
Reopen every authority and platform record on publication day, especially ICO pages marked under DUAA review. Have each layer reviewed by people with appropriate UK privacy, advertising, IP and research-ethics competence, and record who approved it and when. Run that review with the other checks before release.
Before you publish
- Place a method note with dates, exclusions and limitations beside each finding.
- State the unit, numerator, denominator and the quantity left geographically unknown.
- Name automated labels, their validation method and known error types.
- Keep payment, gifting and commission visible on quoted material.
- Offer a route to challenge quotations and inferred categories, and log each correction.
- Record the accountable reviewers and their approval before release.
Common questions
What should a method note beside a chart contain?
It should state collection start and end, time zone and query version, then the authorised sources and inaccessible spaces. Add the unit counted, deduplication and exclusions, the England inclusion rule, classification methods, numerator and denominator, missing data, limitations and the update date.
Do automated labels have to be disclosed to readers?
Name any automated topic, sentiment or risk label, with its version, validation method and known error types. Say whether a person reviews the original context before action. Where validation is too weak, label the output exploratory or withhold it rather than publish an invented accuracy score.
What should happen when a quoted person disputes a label?
Log the original evidence, the review, the outcome and the correction date, and update the downstream reports that used it. Remove identifiers from public examples when they add no decision value, so the correction does not leave the person more exposed than the original quotation did.
