Planning multilingual social marketing around self-described needs and reversible decisions

A sound multilingual social marketing strategy does not begin by assigning languages to demographic groups. It begins with one task, an England-based eligibility rule and a decision that the organisation can reverse. People state the language and format they need for that task. The team then controls the source claim, target-language version, specialist reviews, channel release and evidence separately.

This guide uses a running planning case: an organisation weighs a target-language social message. It would help adults in one English service area start an application.

That is a planning object, not a real campaign report. No language, location, response or result is supplied. Each organisation must fill those fields from its own research and records.

What to take away

Start with a reversible decision and a defined task, not with languages assigned to demographic groups.
Ask eligible people to state their own language and format needs.
Freeze the approved source before any translation, localisation or transcreation work begins.
Treat language and accessibility as separate controls.
Compare routes only after fixing one source asset, one target version, one audience task and one period.

Define the decision before choosing a language

Write a one-sentence decision with two real alternatives. For example: should the organisation run a capped target-language pilot for the named application task, or keep the current supported route while it gathers better evidence? Record who may authorise either course, the decision date and what would cause withdrawal.

Define the decision

Decision: Run a capped target-language pilot for the named application task in the defined English service area, or keep the current supported route while gathering better evidence.
Authoriser: Head of service.
Decision date: One month from approval.
Withdrawal trigger: Inability to recruit eligible participants who can receive equivalent information.

This is a format example, not a live decision.

The audience boundary should use observable eligibility, such as adults who recently attempted the specified task while living or receiving the service in a defined English area. Ask eligible people to describe their own language and format needs. Names, postcodes, ethnicity fields, profiles and administrative labels do not establish proficiency, literacy, culture, vulnerability, consent or purchase intention.

The ONS Census 2021 language bulletin concerns usual residents aged three and over in England and Wales on 21 March 2021. It measures main language, not a current service task choice.

The ONS language-question development report records difficulties some bilingual respondents had with the question. These sources can inform inclusive research planning but they cannot select individuals, predict channel use or prove demand.

The ninety-day sequence at a glance

Days	Focus	Gate to pass	Outputs
1 to 30	Establish the decision	Participants can receive equivalent information and the evidence can change the decision	Decision card, evidence register, stop rule
31 to 60	Control production	Meaning, rights, access and fallback are sound	Frozen source, target version, reviewer records, synthetic test results
61 to 90	Decide on a capped release	Real materials and route accepted by specialists; cap in buyer-controlled units	Release plan, measure cards, pause authority
Day 90	Decision record	Proceed, modify or hold with evidence	Auditable record of approved meaning, usable delivery, independent gates, interpretable measures and continuing ownership

A live pilot is optional and begins only after each applicable specialist accepts the real materials and route. Pause on source change, target-language drift, inaccessible rendering, disputed rights, unreconciled tracking, unauthorised access, an unmanageable complaint or loss of the safe service route.

Build an evidence register, not a persona

Give every proposition an evidence class. This prevents a planning assumption from turning into an audience fact after several rounds of editing.

Evidence classes and permitted use

Evidence class

Population context: Identify broader coverage
Direct research: Answer declared question
Production evidence: Match target to source
Channel observation: Monitor chosen route
Outcome evidence: Test operational hypothesis

Permitted use

Population context: No individual inference
Direct research: No England-wide claim
Production evidence: No audience understanding proof
Channel observation: No causal business effect
Outcome evidence: No attribution beyond design

Boundary

Population context
Direct research
Production evidence
Channel observation
Outcome evidence

Build an evidence register

Evidence class	Example record	Permitted use	Boundary
Population context	Dated official table with geography and question wording	Identify where research routes may need broader coverage	No individual inference or demand estimate
Direct research	Eligible population, recruitment route, source and target materials, numerator, denominator and exclusions	Answer the declared research question within the sampled population	No claim about England unless the design supports it
Production evidence	Version log, translator record and review findings	Decide whether the target asset matches the approved source	No proof that an audience understands or prefers it
Channel observation	Delivered or displayed events under a documented platform definition	Monitor the chosen route	No causal business effect
Outcome evidence	Service record with an eligible population, period and comparison design	Test a defined operational hypothesis	No attribution beyond the design

The GOV.UK Service Manual recommends clear criteria and varied routes when finding research participants. Use that page as public-service research guidance, not as a guarantee that a sample represents England. Record invitations, eligibility, participants, withdrawals, non-response and access provisions by route.

Freeze the source before language production

The source owner approves the factual claim, intended action, legal qualifications, dates and prohibited interpretations. Translation starts only from that numbered version. A late English edit invalidates every target asset derived from it until the change is assessed.

Language service types

Service type

Translation: Transfers meaning from fixed source
Localisation: Adapts conventions for locale
Transcreation: Wider creative treatment

What it does

Translation: No creative changes
Localisation: Defined locale only
Transcreation: New claims review required

Key constraint

Translation
Localisation
Transcreation

Language work should have a stated service type. Translation transfers meaning from a fixed source; localisation adapts conventions for a defined locale; transcreation permits a wider creative treatment under a new claims review. Do not use the labels interchangeably in a brief or invoice.

A qualified native-level language reviewer examines meaning, register and local usage for the actual language and locale. A subject reviewer checks technical claims. Those are different judgements. Bilingual identity alone does not prove professional translation skill, and a language professional should not be made responsible for an unsupported legal, medical or financial interpretation.

GOV.UK's guidance on considering translations recommends evidence of user need and keeping translated content accurate and current. Its publishing rules apply to GOV.UK, but the planning principles are useful elsewhere. Retain a source-to-target register with owner, language, locale, source version, target version, translator, reviewers, approval time and expiry.

If machine assistance enters the workflow, record the supplier, service and version, input rights, retention, security decision, human reviewer and manual fallback. Do not describe output as reviewed until a named person has examined the actual target version.

Treat language and accessibility as different controls

A translated asset can still have unusable captions, poor reading order, unannounced language changes or inaccessible images. Ask about language and access needs separately. Test the rendered post, video, landing page and error route with the intended assistive technology and language settings.

The current GOV.UK page on accessible communication formats says accessible formats may still be needed in addition to an initially accessible document. It also calls for disability-aware quality assurance when disability-related material is translated. This is government guidance, not a complete accessibility determination for a commercial campaign.

Equality and accessibility review should therefore sit beside, not beneath, language approval. A named reviewer needs the actual audience task, channel rendering and service route. Where children or adults at heightened risk may encounter the content, add a safeguarding decision before research or release rather than inferring vulnerability from language.

Compare routes with one planning unit

Fix one unit before comparing routes: one approved source asset, one target-language version, one self-described eligible audience task in England, and one observation period.

Compare candidate distribution routes

Route

Public organic post: Viewed beyond intended population
Paid placement: Depends on platform targeting
Electronic direct message: Requires PECR analysis
Contracted creator: Adds relationship and claims

Key difference

Public organic post: Content owner
Paid placement: Media payer
Electronic direct message: Data owner
Contracted creator: Contract controls

Main control

Public organic post
Paid placement
Electronic direct message
Contracted creator

Candidate routes: a public organic post, a paid placement, an electronic direct message or a contracted creator distribution. Their event definitions and legal questions differ, so headline platform numbers are not interchangeable.

For each route, record:

audience eligibility and how it was established without proxy profiling;
account, content and data owner, plus the person who pays any media or partner charge;
source, translation, native-level, subject and accessibility work required;
platform or partner permissions, delivery definition and evidence export;
privacy, PECR, advertising, consumer, security and contract reviews;
the full cost field, left blank until a dated quote identifies currency, VAT status, unit, period, inclusions and exclusions;
pause, correction, withdrawal, archive and exit routes.

No single route is universally preferable. Public posts can be viewed beyond the intended population, paid placements depend on platform targeting and reporting definitions, and electronic messages require their own PECR analysis.

Creator content adds relationship, claims and contract controls. Apply the ICO's current direct marketing guidance and its separate electronic-mail guidance to the actual route, with a qualified UK privacy and PECR reviewer.

Define measures before a pilot

A platform dashboard is an observation source, not evidence that the activity caused a result. Make a measure card for each planned decision signal:

Field / Required entry

Event and eligible population: Exact event and who could enter the denominator
Numerator and denominator or unit: Queryable definitions, not a label such as engagement
Period and timezone: Start, end, late-event rule and reporting clock
Evidence: System, account, query, version and extraction owner
Exclusions: Tests, staff, duplicates, invalid records and missing data
Interpretation: Observation, rule-based attribution or causal estimate
Decision: Buyer-set threshold, action, stop rule and uncertainty note

An illustrative completed measure card entry: Event: Started application. Eligible population: Adults who attempted the specified task in the defined English service area during the observation period. Numerator: Unique users who started an application. Denominator: Eligible users who saw the target-language asset. Source system: Service database. 0. Exclusions: Staff, tests, duplicates. Uncertainty note: Small numbers may produce wide intervals.

For translation quality, use defect categories tied to the source claim and task. For accessibility, record the checks run and unresolved barriers. For service behaviour, define the eligible application events in the organisation's own system. Never substitute a view, click or favourable comment for comprehension, completion or incremental effect.

Keep approvals non-compensating

A promising early signal cannot cancel a failed legal or safety gate. Named specialists should review the actual materials and processing route where their discipline applies before release.

In this planning case, the route determines which gates apply. Any processing of personal data or electronic message triggers UK GDPR and PECR. Advertising claims fall under the CAP Code.

Copyright and other licences cover source and target material. Security review covers accounts, access, target files, audience data, suppliers, incidents and deletion. Contracts, procurement, employment status, tax and VAT each require their own owner and evidence.

The UK GDPR and Data Protection Act framework, including changes made by the Data (Use and Access) Act 2025, needs a purpose, roles, lawful route, transparency, rights and retention assessment. The ICO states that all DUAA data-protection provisions are now in force, while some ICO material remains under review. Recheck the operative pages on publication and launch dates.

PECR and later direct marketing remain a separate gate. The CAP Code covers non-broadcast advertising, sales promotions and direct marketing communications, including rules on recognition and substantiation. Platform approval is not UK legal clearance.

Copyright and other licences must cover source and target material, including any reuse by suppliers or creators. The GOV.UK overview of using somebody else's copyright explains that permission or a valid exception may be needed. A qualified IP reviewer should examine the actual rights chain.

Security review covers accounts, access, target files, participant or audience data, suppliers, incidents and deletion. The NCSC's supply-chain principles organise oversight around understanding risks, establishing control, checking arrangements and continuing improvement. Contracts, procurement, employment status, tax and VAT each require their own owner and evidence. They should not be collapsed into a general supplier-approved field.

Plan a reversible ninety-day sequence

During days 1 to 30, the decision owner fixes the task, England population, research question, source claim and stop conditions. Gate one stays on hold if participants cannot receive equivalent information or the decision cannot change. Research, privacy, language and accessibility leads examine the proposed evidence route. The phase ends on a written go, modify or hold decision.

It does not end because a calendar date arrived.

Days 31 to 60 cover source approval and non-production preparation. Commission the stated language service, review the exact target version, run accessible rendering and synthetic workflow checks, and prepare monitoring and withdrawal. Keep personal data out of test fixtures. Any unresolved meaning, rights, security or legal issue returns the asset to its owner.

Gate two fails if meaning changes, rights are unclear, access is excessive or the fallback cannot be used.

Days 61 to 90 may include a capped release only if every gate has evidence. A live pilot is optional. It may begin only after each applicable specialist has accepted the real materials and route. Define the cap in buyer-controlled units, not an invented benchmark.

Name the release owner, observation period, timezone and person who can pause without commercial approval.

Watch measure cards, support requests, complaints, rendering and source changes. Pause on source change, target-language drift, inaccessible rendering, disputed rights, unreconciled tracking, unauthorised access, an unmanageable complaint or loss of the safe service route. The owner records the event and preserves evidence.

Restart requires a corrected source or target version, repeated affected tests and approval at least equal to the original release authority.

Make the final decision auditable

At review, ask not whether multilingual activity felt successful, but whether evidence answers the original decision within limits. A proceed decision needs an approved source, a controlled target version, qualified reviews and usable delivery. It needs reconciled measures and sustainable ownership.

A modify decision names the exact defect and next test; a hold decision is correct when the organisation cannot establish audience task, rights, safe route or evidence quality.

The author, fact-checker and named specialists must recheck legal and platform material before publication or use. This draft gives a planning method, not legal, translation, accessibility, tax, procurement or financial advice, and it promises neither compliance nor campaign performance.

Before you act

Write a one-sentence decision with two real alternatives.
Record who may authorise each course and what causes withdrawal.
Ask eligible people to describe their own language and format needs.
Freeze the approved source before starting any language production.
Name separate language, subject and accessibility reviewers.
Define measures and denominators before running any pilot.

Common questions

How should an organisation define its audience boundary?

Use observable eligibility, such as adults who recently attempted the specified task while living or receiving the service in a defined English area. Ask eligible people to describe their own language and format needs.

What is the difference between translation, localisation and transcreation?

Translation transfers meaning from a fixed source, localisation adapts conventions for a defined locale, and transcreation permits wider creative treatment under a new claims review.

Do not use these labels interchangeably in a brief or invoice; state the service type instead. A qualified native-level reviewer examines meaning, register and local usage for the actual language and locale.

Can a platform dashboard prove that a campaign caused a result?

No. Make a measure card for each planned decision signal, with the exact event, eligible population, numerator and denominator. Platform event definitions and legal questions differ, so headline platform numbers are not interchangeable across routes.