Cardiff listening checklist for bilingual Welsh and English brand monitoring

What to take away

Step one: define the Cardiff decision the listening must serve

Start with the decision, not the dashboard. In Cardiff the decision is usually one of four. Is a Welsh-language criticism spreading? Did a campaign land in both languages? Does a service complaint need a response? Has a policy story crossed from English into Welsh media?

Write the decision down in one sentence before you buy or configure anything. A team that cannot name the decision will collect everything and answer nothing. This is the same discipline set out in our guide to defining a target audience for social listening before fieldwork begins.

Cardiff adds a language dimension to that audience work. The city is Wales's media and public sector centre: Senedd coverage, BBC Cymru Wales, Golwg360, Nation.Cymru and a dense third sector all publish in Welsh, often first. Your audience definition should say which of those outlets and communities matter to the decision.

Record who owns the decision and what would change their mind. If nothing would change, stop. Listening without a decision attached becomes an archive nobody opens.

Step two: build Welsh and English query term sets

Build two term sets and keep them separate until the reporting stage. English terms cover brand names, product names, English slogans and English-language service phrases. Welsh terms cover mutated forms, Welsh brand renderings and campaign vocabulary that never appears in English.

Welsh mutation is the first practical trap. A brand beginning with c, p or t may appear mutated after certain particles, so searching only the dictionary form misses posts. Add the common mutated variants as separate query terms rather than relying on stemming, which handles Welsh poorly.

Expect code-switching too. Cardiff posts often mix languages in one sentence, so a Welsh post may contain an English brand name and still be missed by an English-only sentiment rule. Treat language detection and brand detection as two separate jobs.

Use published statistics to size the Welsh-language opportunity rather than guessing. The Welsh language statistics on StatsWales give you population and speaker baselines for query planning, and the wider statistics and research collection from Welsh Government helps you place Cardiff against the rest of Wales.

Term set checklist:

Step three: align monitoring with Welsh Government campaign cycles

Welsh Government campaign cycles shape Welsh-language volume in a way English-only monitoring rarely anticipates. Announcements, consultations and awareness campaigns are published bilingually, and Welsh-language discussion often peaks around the Welsh version rather than the English one.

Map the year before you set alert thresholds. Budget statements, health campaigns, education changes and major consultations each produce a predictable rise in Welsh-language posting. A threshold tuned in a quiet month will fire constantly during a campaign week and be ignored.

The Welsh language policy context behind those campaigns is set out on the Welsh Government's Welsh language topic page, which is the reference point for what official bilingual publication looks like. Use it to anticipate which announcements will generate Welsh-language reaction rather than English.

Set your calendar so the team is not surprised. If a campaign is due in the autumn, brief the Welsh-language reviewer in advance and pre-agree which terms will be watched. Reaction that arrives before the reviewer is ready is reaction you will not classify properly.

Step four: test whether Welsh posts are missed or mistranslated

Run a missed Welsh post check before you trust any number. The check is simple: take a known Welsh-language conversation from the last month, run it through your setup, and see whether it appears, whether it is classified correctly and whether its sentiment is right.

Do this with real posts from Cardiff, not synthetic examples. Pull a sample from a Welsh-language community page, a Senedd-related thread and a Welsh media comment section. If any of the three fails to appear, your term set has a gap, not your tool.

Mistranslation is the second failure mode and it is harder to spot. Machine translation of Welsh sentiment is unreliable on idiom and on short posts, so a negative Welsh post can read as neutral after translation. Have a Welsh speaker review a sample every month, and treat their corrections as data, not opinion.

Keep the samples and the corrections. When someone asks why a Welsh post was missed in March, the answer should be a documented query change, not a shrug. This is the same testing discipline described in our quality checklist for releases leaving the team.

Step five: set retention and review gates for Cardiff data

Retention and review gates decide how long Cardiff data is kept and who signs off before it is used. Set both before collection starts, because retrofitting them after a complaint is expensive and visible.

The legal frame is UK GDPR and the Data Protection Act 2018. The Act is published in Welsh as well as English, and the Data Protection Act 2018 Welsh text is the version to cite when your Welsh-language policy documents need to match the law they describe.

Retention should be short and stated. Public social posts are still personal data when they identify someone, so keep raw posts only as long as the decision needs them and aggregate earlier where you can. Write the retention period into the collection brief, not a separate policy nobody reads.

Review gates are the human checkpoints. At minimum, gate the first release, any release naming individuals, and any release that will be shared outside the team. Research ethics guidance from the Market Research Society is a useful benchmark for how those gates should treat participants and their words.

The workflow from request to approved deletion is set out in our guide to a bounded request, which maps cleanly onto a bilingual setup where translations are also stored.

Step six: report bilingual coverage with its gaps shown

Report Welsh and English side by side, with the same measures and the same caveats. A single combined figure hides the thing Cardiff readers most need to see: whether Welsh-language conversation was captured at all.

Show coverage as a proportion, not a count. Welsh-language volume is smaller than English, so a raw count makes Welsh look negligible even when it carries the decisive criticism. Report the share of known Welsh posts captured against the share of known English posts captured.

Name the gaps in the report itself. If Welsh-language coverage is weaker in a particular week, say so on the page rather than in a footnote. A dashboard that shows its gaps as plainly as its numbers is more useful to a Cardiff communications team than one that looks complete.

Keep the table in every release so the comparison cannot be dropped when space is tight. If a month has no Welsh-language gap to report, say that too, because a clean month is evidence the term set is working.

Common questions