Translation and localisation are not the same thing. Translation converts words from one language to another. Localisation adapts the meaning, tone, idioms, and cultural references so the copy feels like it was written for that market — not exported from somewhere else.
The difference matters more than most companies realise. UK copy that gets directly translated into US English still sounds British. American directness can read as abrasive in German markets. Australian audiences pick up on imported idioms immediately. A visitor who senses they're reading copy aimed at someone else doesn't stay long.
Maya, global content manager at a UK-based SaaS company expanding into the US, Australia, and Germany, had 340 copy blocks across the main marketing site — hero sections, feature descriptions, case study summaries, email templates. Getting a localisation agency to adapt all of them for three markets would have taken eight weeks and a budget she didn't have. Running it through a simple translation tool would produce output she'd be embarrassed to publish.
She used AI batch processing to do something in between: intelligent localisation at the speed of translation.
What localisation actually requires
Before building the spreadsheet, Maya listed the specific adaptations each market needed. This became the foundation of her prompt — localisation rules you haven't articulated can't be applied consistently.
Her notes, condensed:
- US English: spelling conventions (colour → color, organisation → organization), remove "quite" and "rather" (reads as hesitant), replace "brilliant" and "sorted" (too British), use "get started" not "get stuck in", pricing examples in USD, date format MM/DD/YYYY
- Australian English: spelling as UK English, conversational tone slightly more relaxed, replace formal constructions ("In order to" → "To"), avoid American idioms that have crept into the UK copy, pricing in AUD
- German: translate to formal German (Sie not du), more precise and less hyperbolic than English marketing copy — remove superlatives unless substantiated, feature descriptions should emphasise reliability and compliance, remove casual contractions that don't translate well
The German market required the most work — not just translation but a genuine shift in register. German B2B buyers respond to precision and credibility, not enthusiasm. Copy that says "supercharge your workflow" needs to become something closer to "optimise your processes reliably."
Building the spreadsheet
The structure was straightforward. One row per copy block, two columns: the source copy and the target market.
| copy_block | target_market |
|---|---|
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | US |
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | AU |
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | DE |
| No API keys. No pipelines. No code. | US |
| No API keys. No pipelines. No code. | AU |
| No API keys. No pipelines. No code. | DE |
340 copy blocks × 3 markets = 1,020 rows. Each row produced one localised output. The two-column structure meant she could filter the results by market and hand each column directly to the developer building the localised site variants.
She added a third column — copy_type — for the longer-running job: hero, feature, testimonial, email. This helped her review the output by content type rather than scrolling through 1,020 rows linearly.
Writing the prompt
The prompt had to carry the full localisation brief — every market rule she'd documented — and apply the right rules based on the target_market column in each row.
Prompt used:
You are a professional content localisation specialist. Localise the copy_block below for the target_market specified. This is not a translation exercise for English markets — it is a cultural and linguistic adaptation. For German, translate and adapt.
US English rules:
— Convert to American spelling (colour→color, organisation→organization, optimise→optimize, etc.)
— Remove British idioms: "sorted", "brilliant", "get stuck in", "quite", "rather", "keen on"
— Replace with American equivalents: "get started", "great", "ready to go"
— Use direct, confident tone — remove hedging language
— Any pricing examples: convert to USD
Australian English rules:
— Keep UK spelling conventions
— Slightly more conversational and direct than UK copy — reduce formal constructions
— Replace Americanisms that have crept in: "gotten" → "got", "reach out" → "contact"
— Keep the warmth of the original; AU readers respond well to approachable copy
— Any pricing examples: convert to AUD
German rules:
— Translate to formal German (Sie, not du)
— Reduce superlatives and enthusiasm: "supercharge" → "optimieren", "amazing" → drop or replace with factual claim
— Emphasise precision, reliability, and control — these resonate in German B2B
— Avoid contractions and casual constructions that don't translate naturally
— Keep sentences complete and specific — German readers distrust vague marketing claims
Output: the localised copy only. Same length as the original where possible. No explanation, no preamble.
Running the batch
1,020 rows on Claude Sonnet 4.6. The German localisation required genuine cultural intelligence — reducing hyperbole, shifting register, finding German equivalents that carry the intended meaning without sounding awkward. Flash-tier models handled the English market adaptations adequately, but the German output was noticeably better on Sonnet: less literal, more natural.
The batch completed in just under three hours. Total cost: £4.20.
Sample output
| copy_block (original) | target_market | AI responses |
|---|---|---|
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | US | Turn any spreadsheet into AI-powered output — at scale, while you focus on what matters. |
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | AU | Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. |
| Turn any spreadsheet into AI-powered output — at scale, while you get on with your day. | DE | Verwandeln Sie jede Tabelle in KI-gestützte Ergebnisse — in großem Maßstab, zuverlässig und automatisiert. |
| No API keys. No pipelines. No code. | US | No API keys. No pipelines. No code. |
| No API keys. No pipelines. No code. | DE | Keine API-Schlüssel. Keine technische Infrastruktur. Kein Code. |
A few things worth noting in the output. The AU adaptation of the hero line was unchanged — the original UK copy was already appropriate for Australian readers. The model correctly identified that no adaptation was needed and returned the source text rather than changing for the sake of changing. The German hero line dropped "while you get on with your day" (too casual) and replaced it with "zuverlässig und automatisiert" — reliable and automated — which is a genuine improvement for that market.
Reviewing AI-localised copy
Maya had a native speaker in each market review a 10% sample before publishing — roughly 34 copy blocks per language. The review revealed a few patterns worth knowing:
- The English market adaptations needed almost no corrections. The rules were precise enough that the model applied them consistently. US spelling, idiom replacement, and tone were handled well across the board.
- The German copy needed the most native review. Around 15% of German outputs needed light editing — mostly cases where a direct translation was technically correct but wouldn't have been how a native writer would phrase it. The model is good at avoiding errors; catching subtle unnatural phrasing requires a human reader.
- Ambiguous copy produced ambiguous localisations. A few UK copy blocks that contained metaphors the model couldn't reliably adapt — "getting your ducks in a row", references to specific UK cultural contexts — produced inconsistent results. The fix was to revise the source copy to be less idiomatic, then re-run those rows. The lesson: if you're planning to localise at scale, write source copy with localisation in mind.
What this cost compared to alternatives
A professional localisation agency for 1,020 copy blocks across three markets — including German translation — would typically run to £8,000–15,000 and a six-to-eight week timeline. The AI batch approach cost £4.20 in compute, a day of Maya's time to prepare the data and review the output, and a few hours from a native German speaker for the review pass.
The agency version would produce higher-quality German copy on average — experienced localisation specialists bring cultural depth that a well-prompted model doesn't fully replicate. But for a first pass that gets three market variants live quickly, that can be refined after real traffic data shows which pages need the most attention, the batch approach is a reasonable trade-off.