The Welsh language and AI: where it works, where it doesn't, and what it means for North Wales businesses
Welsh is a living working language across North Wales. AI's handling of it has improved sharply, but with real limits. A practical look at what is and is not safe.
If you run a business in Gwynedd, Anglesey, Conwy or parts of Denbighshire, Welsh is not a heritage language - it is a working language. Customers prefer it, regulators expect it, and employees use it daily. Any AI tool you adopt has to make sense in that bilingual reality.
This article looks at what current AI actually does well in Welsh, where it falls down, and the practical implications for North Wales businesses.
What AI handles well in Welsh today
The big general-purpose models - ChatGPT, Claude, Gemini, Copilot - all handle Welsh comfortably for most everyday business writing. They can translate from English to Welsh and back, draft emails in either language, summarise Welsh content, and switch between formal and informal registers.
For routine bilingual content - a Welsh-language version of a website page, a customer email, a social post, a sign - they are useful starting points. The output will not be perfect every time, but it is good enough that a Welsh-speaking colleague can quickly polish it.
This is a meaningful shift from three years ago, when AI Welsh was usually word-substitution that read as machine-translated. The big models have closed most of that gap.
Where AI still falls short
Three honest failure modes to know about.
Mutations. Welsh has soft, nasal and aspirate mutations triggered by grammatical context. AI gets this right most of the time but not always. The mutation errors are rarely catastrophic but read as obviously wrong to a native speaker. The fix: a native review before public-facing material is published.
Place names and proper nouns. AI sometimes anglicises Welsh place names ("Caernarfon" turning into "Caernarvon", "Llandudno" surviving but "Yr Wyddfa" reverting to "Snowdon"). This is a brand-and-respect issue, not just a language issue. Always check.
Regional vocabulary. Welsh has regional variation - North Wales (Gwynedd) Welsh and South Wales Welsh use different words for the same thing. AI tends to default to a textbook standard register, which can sound formal or "off" in Caernarfon or Bangor. For audience-facing content, prompt for the regional register: "Welsh as spoken in north-west Wales".
What this means for your business
If you serve Welsh-speaking customers, three practical implications.
1. AI lowers the cost of being properly bilingual. Producing a Welsh version of every customer email, every product description, every social post used to mean a translation budget that small businesses could not afford. With AI as a first draft and a Welsh-speaker to polish, the cost falls to a fraction of that. Genuinely bilingual service is more achievable.
2. The "polish" step still has to be a real Welsh-speaker. This is not optional. Public-facing Welsh content with AI errors will be noticed, and the credibility cost is real. Cymdeithas yr Iaith and individual customers are quick to flag obvious machine-translation. AI accelerates the work but does not finish it.
3. Welsh customers can interact with AI tools in Welsh. If you set up an AI assistant for guest enquiries, customer support or scheduling, today's big models will handle Welsh-language input. The reply quality depends on your prompt setup. The five tourism prompts post has the structure - the same approach works in Welsh, with the key step of telling the AI to reply in Welsh by default.
Practical patterns
Three patterns that work for North Wales businesses today.
Bilingual-first content production. Write your draft in your stronger language. Ask AI to produce a Welsh equivalent. Have a Welsh speaker review. Publish both versions. This produces consistent bilingual content faster than writing twice.
Welsh-by-default chatbot setup. If your audience is mainly Welsh-speaking, set the AI assistant to greet and reply in Welsh, switching to English only if the customer does. The reverse pattern is normal for English-first businesses; the Welsh-first pattern is rare and a real differentiator in places like Caernarfon and Pwllheli.
Translation memory, not just translation. As you produce content, save your reviewed Welsh versions. Over time you have a corpus of correct, on-brand Welsh that you can paste into AI as reference for new content. This is how you keep terminology consistent across years of material.
Where this sits in the regional picture
AI in Welsh is not a separate technology - it is a feature of the same general-purpose models. That means it benefits from every general AI improvement automatically. Three years ago, Welsh was a niche capability. Now it is mainstream. In another two years it will likely be at parity with English for everyday business work.
For now, the practical line is: use AI as a tool to make bilingual operation cheaper and faster, but do not let it be the final voice of your Welsh-language service.
Where to go from here
If you would like to talk through what AI plus Welsh could look like for your business - whether that is an AI-assisted bilingual content pipeline, a Welsh-speaking chatbot, or just a sensible policy for your team - that is the kind of work a discovery call can scope. The regional economy post sets the broader scene.