AI for Everyday Work: Writing, Research, Meetings & Data · Translation and personal knowledge · lesson 16 of 17 · 12 min

Translation and localization

Translation vs localization

Translation changes words from one language to another. Localization adapts the whole message for a specific audience: tone, examples, idioms, currency, dates, measurements, cultural references and legal requirements. A promotion that works in London might need a very different approach in Riyadh, Karachi or New York, even in English.

Modern AI assistants translate between many languages and can help localize, though quality varies by language and topic. Widely used languages usually get stronger results than less-resourced ones.

The translation prompt

Translate the text below from [English] into [Modern Standard Arabic / Urdu / French ...] for [audience, e.g. customers in Saudi Arabia].
Keep the meaning and tone [friendly and professional].
Keep these terms unchanged: [brand names, product names].
Flag any phrase that doesn't translate well and suggest an alternative.
[paste text]

"Flag any phrase that doesn't translate well" often catches idioms and jokes that would confuse or offend.

The localization prompt

Adapt this marketing email for [market]. Consider local holidays and seasons, currency ([SAR/AED/PKR/GBP/USD]), date format, cultural norms and examples that will resonate.
List every change you made and why.
[paste text]

Asking for a list of changes teaches you about the market and lets you check each adaptation.

Always involve a fluent human for important content

AI translation is excellent for understanding, drafting and internal use. For published, legal, medical, safety or customer-facing content, have a fluent speaker (ideally a native speaker familiar with the market) review it. They'll catch mistranslated terms, awkward phrasing, and cultural issues AI can miss.

Useful everyday uses

Understanding emails or documents in languages you don't read.
Replying to international colleagues and customers, with a quick back-translation check: "Translate your Arabic reply back into English so I can check the meaning."
Multilingual social captions (with native review for brand accounts).
Subtitles and scripts for videos (checked by a fluent speaker).
Learning: "Explain the difference between these two phrases and when to use each."

Worked example

A UK skincare brand launches in the UAE and Saudi Arabia. The marketing assistant uses AI to translate product descriptions into Arabic and localize the launch email: prices in local currencies, references to local seasons, and a tone suited to the audience. The AI flags a slogan built on an English pun that won't work in Arabic and suggests alternatives. A native Arabic-speaking copywriter reviews everything, adjusts two product terms to match how customers actually search for them, and approves. The launch copy feels local rather than translated.

Sample output check

For a translation, a good response keeps brand names unchanged, preserves numbers and dates exactly (or converts formats as you requested), and includes a short list of flagged phrases with alternatives. Back-translation is a quick sanity check: if the English back-translation says something different from your original, investigate before sending.

Building a glossary

If you translate regularly, build a simple glossary of approved terms: brand names, product names, job titles and key phrases with their approved translations. Paste it into your translation prompt each time ("Use this glossary for these terms") or store it in an assistant project. A glossary keeps translations consistent across people and months.

Regional and legal notes

Advertising rules apply in every market: claims, disclosures (such as #ad for sponsored content) and required information can differ by country.
Right-to-left languages (Arabic, Urdu) need proper layout in designs and documents; check how text displays.
Names and honorifics vary by culture; ask the AI about local conventions, then confirm with a local colleague.

Hands-on: translate, back-translate, localize

1. Translate with a glossary

Translate the text below from English into [Gulf Arabic for customers in the UAE / Urdu for customers in Pakistan].
Tone: warm and professional. Audience: [who].
Glossary (use exactly): [brand name] = unchanged; "free delivery" = [approved term]; "loyalty points" = [approved term].
Flag any phrase that doesn't translate naturally and suggest an alternative.
[paste text]

2. Back-translate to check meaning

In a new chat (so it can't copy the original), paste the translation and ask: "Translate this into English as literally as possible." Compare with your original. Differences in meaning, not wording, are what matter.

3. Localize, and list every change

Adapt this English promotional email for [market]. Consider: local holidays and seasons, currency and price format,
date format, weekend days, cultural norms, examples that resonate, and any local advertising rules I should check.
List every change you made and why, as a table.

Before: a direct translation of a UK spring sale email for Saudi customers keeps "Easter weekend deals", prices in pounds and a Saturday-Sunday "weekend".

After: the adapted email ties the offer to a relevant local moment, shows prices in SAR, respects the Friday-Saturday weekend, and includes a change table your Riyadh colleague reviews in ten minutes, correcting one word choice.

Live and voice translation

Voice modes in ChatGPT and Gemini Live can interpret conversations between two languages in real time on supported devices, and meeting tools increasingly offer translated captions. Useful for informal conversations and understanding; for contracts, medical or legal matters, use a qualified interpreter.

Measure it

Ask your fluent reviewer to mark each AI translation as "publish as is", "minor edits" or "rewrite". Track the ratio by language and content type; it tells you where AI drafts are good enough and where to invest in human translation.

Pitfalls

Publishing machine translations without fluent review.
Translating puns, idioms or humor literally.
Forgetting currencies, dates and units.
Pasting confidential documents into unapproved translation tools.

Video lecture: Translation and localization

Lecture coming soon · 11 chapters · about 8 minutes. Read the full transcript below.

Translation and localization
Why it matters
The tour guide analogy
The three-step method
Example 1: the customer reply
Example 2: the spring sale email
Watch me do it
Live and voice translation
When a human must review
Common mistakes
Recap and try this now

Lecture transcript

Translation and localization

Imagine sending a beautifully translated promotion to customers in Riyadh, with every word correct, and it still falls flat. The offer is tied to a holiday they don't celebrate. The prices are in the wrong currency. The weekend is on the wrong days. Nothing was mistranslated. It just wasn't localized. In this lecture you'll learn the difference between translation and localization, a three step method with a glossary, back translation and a change table, and when a fluent human has to be involved. You'll see two examples, and watch me localize a real email.

Why it matters

Why does this matter? Because AI translation has become genuinely good, especially for widely used languages. For understanding a message, drafting a reply or internal use, it's often excellent. But quality still varies by language and topic, and less widely used languages generally get weaker results. And even perfect words can carry the wrong meaning or the wrong cultural signal. When that happens in an internal note, it's a small embarrassment. When it happens in a public ad, a product label or a customer contract, it's public, and sometimes costly. So we use AI for speed and a clear process for safety.

The tour guide analogy

Here's an analogy. A dictionary gives you words. A local tour guide gives you meaning. The guide knows which café locals actually go to, which days shops close, and which jokes won't land. Translation is the dictionary. Localization is the guide. It adapts tone, examples, idioms, currency, dates, measurements, cultural references and even legal requirements. A promotion that works in London might need a very different approach in Riyadh, Karachi or New York, even when it's all in English. AI can do a good job of both, if you ask for both, and if you check.

The three-step method

Here's the method. Step one, translate with a glossary. Tell the AI the language variety and audience, like Gulf Arabic for customers in the UAE, give it a short glossary of terms that must be used exactly, like your brand name and approved product terms, and ask it to flag any phrase that doesn't translate naturally. Step two, back translate. In a new chat, so it can't just copy your original, paste the translation and ask for a literal English version. Compare meaning, not wording. Step three, localize. Ask it to adapt the message for the market, and to list every change it made and why, as a table. That table teaches you about the market and makes review fast.

Example 1: the customer reply

A simple example. A customer support agent in Dubai receives a message in Arabic about a delayed order, and her Arabic is basic. She asks her approved assistant to explain the message in English. Then she writes her reply in English, the delivery is now Thursday and here's the tracking link, and asks for a warm, professional reply in Gulf Arabic. Before sending, she back translates it in a fresh chat. The meaning matches, except one phrase came back as we promise Thursday, which is stronger than she intended. She softens it to we expect Thursday. Two minutes, and a customer who feels understood.

Example 2: the spring sale email

Now a realistic business scenario. An online homeware brand in Manchester wants to reuse its spring sale email for customers in Saudi Arabia. A direct translation would keep Easter weekend deals, prices in pounds and a Saturday to Sunday weekend. Instead, the marketing coordinator runs the localize prompt. The adapted version ties the offer to a relevant local moment, shows prices in Saudi riyals, respects the Friday to Saturday weekend, and simplifies an English pun that wouldn't work. The change table lists all seven changes with reasons. Their colleague in Riyadh reviews it in ten minutes and corrects one word that sounded too formal. And the coordinator checks the local rules on promotional claims before launch, because the AI flagged them as something to check.

Watch me do it

Let me show you. I have a short English welcome message for new customers of a gym in Karachi, and I want an Urdu version. I paste the translation prompt: Urdu for customers in Pakistan, friendly and motivating. Glossary: the gym's name stays unchanged, and personal training uses the term our Karachi team prefers. Flag anything that doesn't translate naturally. It returns the Urdu text and flags two phrases. One is an English idiom, crush your goals, which it suggests replacing with a more natural motivating phrase. Next, I open a brand new chat, paste the Urdu, and ask for a literal English translation. The meaning matches. Finally, I'd send it to a fluent colleague, because this will go to every new member.

Live and voice translation

Translation isn't only about documents anymore. Voice modes in assistants like ChatGPT and Gemini Live can interpret a conversation between two languages in real time on supported devices, and many meeting tools now offer translated captions. That's wonderful for a friendly chat with a supplier, understanding a customer on the phone, or following a meeting in a language you're still learning. But know the limits. Real time interpretation can miss nuance, names and numbers. So for contracts, medical conversations, legal matters, or anything where a misunderstanding has consequences, use a qualified interpreter. And if you're recording or captioning a conversation, the same consent rules apply as for any meeting.

When a human must review

So when must a fluent human review? For anything published or customer facing. For anything legal, medical or safety related, where a wrong word can hurt someone or create liability. And for less widely used languages, where AI quality can be less reliable. Ideally your reviewer is a native speaker familiar with the market, not just the language. Build a glossary together over time, with approved terms for your products and brand, and paste it into every translation prompt, or keep it in a project. It makes every future translation better and faster to review.

Common mistakes

Common mistakes. Translating when you needed to localize, so the words are right but the message is wrong. No glossary, so your brand and product terms change from one email to the next. Back translating in the same chat, where the AI can simply repeat your original, which tells you nothing. Skipping the fluent reviewer for public content because the translation looks fine to you, when you can't actually read it. And pasting customer personal details into unapproved tools just to translate a message. Remove what isn't needed, and use your work approved assistant.

Recap and try this now

Let's recap. Translation changes the words. Localization adapts the whole message. Use the three step method: translate with a glossary and ask for flagged phrases, back translate in a fresh chat to check meaning, and localize with a table of every change. Get a fluent human to review anything public, legal, medical or safety related. Here's your try this now. Take one short piece of real content, a welcome email, a product description or a social post, and run all three steps for a market you work with. Then show the change table to someone who knows that market. Their reaction will tell you a lot.

Key takeaways

Translation changes words; localization adapts the whole message: tone, examples, currency, dates, culture and rules.
Give a glossary of fixed terms and ask the AI to flag phrases that don't translate naturally.
Back-translate in a fresh chat to check meaning, and ask for a table of every localization change.
Have a fluent human review anything published, legal, medical, safety-related or customer-facing.

Try it

Translate and localize one real piece of content for a market you work with. Use back-translation to check it, and if possible ask a fluent speaker to review.