Earlier this yr, I used to be handed an AI-generated content material venture with a deceptively easy purpose: adapt e-mail messages for worldwide audiences.
This wasn’t my first time navigating world nuance. With an MBA in Worldwide Enterprise and expertise engaged on a worldwide consulting venture in Portugal, I’d already seen how messages land in another way relying on tradition, tone, and language. However this was my first time making use of that lens to AI content material era in my MarTech AI position at HubSpot — and it was extra complicated than anticipated.
We already had an AI-generated e-mail immediate that labored effectively in English—conversational, pleasant, and context-aware. The problem? Making it work in Spanish and French with out sounding robotic, clumsy, or culturally off-base.
Sounds straightforward. It wasn’t.
The Hidden Complexity of “Simply Localizing”
What we had been actually doing was asking an AI mannequin — skilled predominantly in English — to talk different languages as naturally as a local marketer would.
Our first makes an attempt fell flat.
Instance (unique AI output in Spanish):
Right here’s what we aimed for in English:
“I noticed you had been scoping across the platform and that you just had been enthusiastic about talking with us. Would you want to satisfy on one of many following days?”
That is the unique output in Spanish:
“Estuve revisando tus interacciones en nuestra plataforma y quería ofrecerme como tu punto de contacto.”
In English, it interprets to:
“I reviewed your exercise and needed to develop into your level of contact.”
Whereas grammatically right, this sounded invasive in Spanish — like we had been watching the consumer too intently. It didn’t really feel pure. One reviewer known as it “creepy.”
Right here’s one other instance:
- Authentic English intent: “I seen you’ve been exploring our platform and expressed curiosity in connecting with us.”
- Authentic Spanish output: “Me pareció interesante tu interés en nuestros servicios.”
- Translation in English: “I discovered your curiosity in our companies fascinating.”
Once more, it’s technically correct, but it surely’s redundant and robotic. It’s the sort of phrasing that makes a reader cease and go, “Did a bot write this?”
The takeaway: Even when the interpretation is correct, the tone may be off. And tone is every thing in advertising.
The Shift from Translation to Language-aware Immediate Design
At this level, I spotted we wanted greater than AI outputs — we wanted a system for guiding the AI to assume like a multilingual marketer.
I constructed a language-portable immediate framework — a structured immediate that might adapt throughout languages whereas respecting every one’s distinctive grammar, tone, and cultural context.
Right here’s What Modified
As an alternative of 1 static immediate, I broke the logic into variables:
- : Goal language (e.g., Spanish, French, German)
- : Pronoun and tone stage (“tu” vs. “usted”, “vous” vs. “tu”)
- : Inbox-friendly, conversational, skilled
- : Direct vs. suggestive phrasing
- : Enforced the place grammar allowed
We additionally added clear, language-specific guidelines.
Instance (Spanish):
- Use tú constantly, by no means usted (too formal for our model)
- Keep away from gendered adjectives like interesado/interesada when potential
✅ “Mostraste interés en … ”
❌ “Estuviste interesado en … ”
Instance (French):
- All the time use vous, not tu, in B2B messages
- Keep away from ambiguous endings like intéressé(e)
✅ “Vous avez montré de l’intérêt … ”
❌ “Tu t’étais intéressé(e) … ”
Why This Shift Mattered
In English, a pleasant CTA would possibly appear to be:
“Would you be accessible for a short dialog on one of many following days?”
We tried straight translating it into Spanish:
“¿Quieres agendar 15 minutos para hablar sobre lo que estás buscando?”
It was grammatically right, but it surely sounded too informal and unprofessional in a B2B context. Not pushy, simply barely off-tone.
So, we reworded it to be pleasant however formal:
“Si te parece bien, podemos agendar una conversación breve esta semana.”
This interprets to:
“If it really works for you, we are able to schedule a brief chat this week.”
Right here’s one other instance in French:
- Authentic output: “Souhaitez-vous prendre rendez-vous pour en discuter ?”
(“Would you prefer to schedule a gathering to debate this?”) - New model: “Auriez-vous 20 minutes pour voir remark HubSpot pourrait concrètement vous aider?” (“Would you could have 20 minutes to see how HubSpot might virtually assist you?”)
The second model provides worth to the CTA. Not simply time — however goal.
Backing It Up With a Stakeholder Questionnaire
Localization isn’t only a linguistic problem — it’s a enterprise alignment problem.
To get it proper, I created a easy stakeholder consumption doc and shared it with advertising ops, regional entrepreneurs, and content material leads. The purpose was to align early on tone, content material boundaries, and regional sensitivities.
These are a few of the questions I requested:
- What stage of ritual is suitable in your market?
- Ought to we keep away from gendered phrases?
- Can we reference the consumer’s firm or product utilization?
- How direct ought to we be in asking for motion?
- Are there idioms, cultural references, or phrasings we should always keep away from?
We obtained some fairly fascinating insights.
For instance, in some areas, stakeholders most popular not to reference the recipient’s firm sort within the copy, regardless that that was widespread in English (e.g., “I noticed that you just assist startups with HR”).
The localized different turned extra common:
“Entiendo que están buscando formas de mejorar sus procesos internos.” (“I perceive you’re trying to enhance inside processes.”)
The outcomes of this survey helped create readability between content material, ops, and regional advertising groups — and dramatically decreased our revision cycles.
The Remaining Product: Human-sounding Emails at Scale
With the up to date immediate and consumption framework, the brand new outputs had been immediately higher.
Earlier than:
- Authentic output: “Hola [FirstName], soy María de HubSpot. He visto que has navegado nuestra plataforma y parece que te interesa nuestro producto.”
- English translation: “Hello [FirstName], I’m María from HubSpot. I noticed you’ve browsed our platform and it appears you’re enthusiastic about our product.”
After:
- Authentic output: “Soy María de HubSpot. Vi que estuviste explorando la plataforma y que querías saber más sobre cómo podemos apoyar tu negocio.”
- English translation: “I’m María from HubSpot. I noticed you had been exploring the platform and needed to be taught extra about how we are able to assist your corporation.”
And stakeholders responded positively:
- “This lastly feels like somebody from our group wrote it.”
- “Excellent tone — pure and native.”
- “No gender errors or bizarre formalities. We will really use this.”
Even higher, we didn’t want to write down separate prompts for each marketing campaign. The identical core framework now powers AI-generated messages in a number of languages — with constant high quality.
Takeaways for Entrepreneurs
Whether or not you’re engaged on AI copy, world advertisements, or multilingual content material, right here’s what I realized:
1. Don’t simply translate — localize for intent.
Literal translations will get you “technically right” content material. However solely localization will make it land.
2. Use prompts like artistic briefs.
Embrace tone, formality, CTA fashion, gender neutrality, and different language guidelines as variables. Don’t depart nuance to probability.
3. Construct language-aware templates.
Languages behave in another way. Plan for issues like verb conjugations, pluralization, and sentence rhythm upfront.
4. Get suggestions early.
Use a stakeholder consumption doc earlier than era, not after. You’ll keep away from rework and misalignment in a while.
5. Goal for an actual, human tone.
In case your AI output doesn’t really feel like one thing you would write to a buyer, it gained’t convert. Learn it aloud. Would you hit ship?
AI localization is a advertising talent now.
This venture taught me one thing that has caught with me since: The way forward for world advertising isn’t nearly scaling content material — it’s about scaling context.
The businesses that succeed with AI gained’t be those who generate essentially the most content material. They’ll be those who generate the most resonant content material as a result of they know find out how to immediate for it. And that begins with understanding the languages your prospects converse — in additional methods than one.
