# Localise client work into 5 languages without paying a translator: the AI translation build Source: https://getstartedwithai.net/tutorials/ai-translation-build Site: Get Started With AI Tagline: Build AI automations that actually run your business. Audience: solo professionals, freelancers, consultants and AI implementers in training Contact: hello@getstartedwithai.co.uk This Markdown companion is generated from the prerendered page for AI crawlers. # Localise client work into 5 languages without paying a translator: the AI translation build Five minutes per document, five languages, output a native speaker can use without rewriting. The build below uses a Custom GPT or Claude Project configured against a glossary, runs the translation through a tone-preserving second pass, and produces production-ready output for most European pairs. AI translation crossed the line from useful for first drafts to good enough for production work in most language pairs during 2025; this is the practical setup for UK solo professionals serving European clients. Setup is 90 minutes including the glossary build. Run cost is £15 a month. ## What you need before the build starts - A ChatGPT Plus or Claude Pro account at £15 to £20 a month. Plus is fine for solo translation work; ChatGPT Business or Claude Team only when client docs are under NDA. - A glossary of 50 to 200 industry terms in your source language with the agreed target translation per language pair. - Your tone notes: formality level, sentence length preferences, the voice register the client expects. - A native speaker reviewer for the first 10 outputs in each language. The Project calibrates against their feedback. ## Step 1: Build the glossary asset (25 minutes) AI translation lives or dies on terminology. Five minutes spent on a glossary saves hours of rewriting on every output afterwards. 1. Open a Google Sheet with columns: source term, definition, target language 1, target language 2 and so on. 2. Populate with the 50 to 200 industry terms specific to your work. Each sector has its own list, whether that is legal or medical or creative work. 3. For each term, agree the target translation with a native speaker once. The Project uses these as fixed mappings, never to translate freely. ## Step 2: Build the translation Project (20 minutes) The Project runs the translation pass and the tone-preserving review pass in one workflow. 1. Create a Project named Translation [language pair]. 2. Upload the glossary, the tone notes and 5 sample translations the client previously approved. 3. System prompt: translate UK English source into [target language]. Use the glossary for industry terms. Match the formality and sentence length of the uploaded examples. Output the translation only, no explanation. 4. Add a follow-up prompt for the tone pass: review the translation against the uploaded examples. Flag any phrasing that reads as machine translation. Output the corrected version. ## Step 3: Test against native speaker review (15 minutes) The first 10 outputs in each language go to a native speaker reviewer. Their feedback calibrates the Project for the rest. 1. Run 10 sample documents through the Project in each target language. 2. Send to the native speaker reviewer with the question: which phrases feel translated, and which feel native? 3. Add the feedback to the tone notes and re-upload. The Project learns the patterns the reviewer flags. 4. After 10 outputs the reviewer typically signs off; subsequent translations need only spot-checking. ## Step 4: Production workflow (10 minutes) The day-to-day workflow for a solo professional running translation as part of client delivery. 1. Paste the source document into the Project. Run the translation prompt. 2. Run the tone pass on the output. Read the result aloud if you have working knowledge of the target language. 3. For confidential client work, redact identifying details before paste; reinsert after the tone pass. 4. Send to the client or to a final native review for high-stakes documents (legal contracts, regulatory filings). ## Step 5: First month of operation Run the build across 20 documents in each target language. Track time per document, native reviewer feedback and any client edits. Most solo professionals see translation time drop from 45 minutes per document to 5 minutes, with native quality intact in 90 percent of cases. ## What you ship A glossary asset that grows over time. One Project per language pair, calibrated to the client tone. Translation time at 5 minutes per document. Native speaker reviewer signed off on the first 10 outputs in each language. A workflow that turns multilingual delivery from a freelance subcontract into an in-house capability. ## What to build next Pair this with the WIRE Method to package translation as a paid service for peer practitioners. For Asian language pairs (Japanese, Korean, Mandarin), follow this with a DeepL routing layer for the first pass and the Project for tone correction. ## Related builds More tutorials in Marketing & Sales. Each one runs in 60 to 90 minutes and ships a finished asset. ### Send 50 cold emails a week that book meetings: the AI build for outbound 30 to 60 minutes. Solo Pro / Implementer. [Open tutorial](/tutorials/ai-cold-email-outbound-build) ### Publish a weekly newsletter in 90 minutes: the AI build for solo creators 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/ai-newsletter-publish-weekly-build) ### Run a LinkedIn presence without burning Sundays: the AI build for solo professionals 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/ai-linkedin-presence-build) ### Synthesise customer interviews into product moves: the AI build for founders 90 minutes. Implementer. [Open tutorial](/tutorials/ai-customer-interview-synth-build) ## Questions people ask before they start ### What is the best AI for translation? For most European pairs, GPT-5.5 inside a Custom GPT and Claude Sonnet 4.6 inside a Project tie within margin of error in 2026. For Asian pairs, DeepL still has the edge on idiom and politeness register; route those through DeepL and use the GPT or Claude pass for tone correction. ### Is AI translation accurate? For business documents in major languages, yes, typically 95 percent or higher after a glossary is loaded. Marketing copy and legal contracts still need a native review pass; the build keeps a human in the loop on those rather than shipping output unchecked. ### Will AI replace translators? For first-draft commodity work, mostly yes by 2027. For specialist legal, medical and literary translation, no, and probably not within a decade. The build frees solo professionals from outsourcing the first 80 percent of the work, which is where the budget tends to disappear. ### What about confidentiality on client documents? Use Pro-tier accounts where inputs are not used for training. Document the AI use in the data-handling agreement with the client. For documents under NDA, run the translation against a redacted copy and re-insert the sensitive details in the human review pass. ### How does this work for UK English to US English? Configure as any other language pair: glossary of UK-specific terms (estate agent, solicitor, A&E), tone notes on the formality difference, sample documents from the client. UK to US is the easiest pair the build handles and runs at near-zero cost per document.