Earlier this year, I was handed an AI-generated content project with a deceptively simple goal: adapt email messages for international audiences.
This wasn’t my first time navigating global nuance. With an MBA in International Business and experience working on a global consulting project in Portugal, I’d already seen how messages land differently depending on culture, tone, and language. But this was my first time applying that lens to AI content generation in my MarTech AI role at HubSpot — and it was more complex than expected.
We already had an AI-generated email prompt that worked well in English—conversational, friendly, and context-aware. The challenge? Making it work in Spanish and French without sounding robotic, clumsy, or culturally off-base.
Sounds easy. It wasn’t.
The Hidden Complexity of “Just Localizing”
What we were really doing was asking an AI model — trained predominantly in English — to speak other languages as naturally as a native marketer would.
Our first attempts fell flat.
Example (original AI output in Spanish):
Here’s what we aimed for in English:
“I saw you were scoping around the platform and that you were interested in speaking with us. Would you like to meet on one of the following days?”
This is the original output in Spanish:
“Estuve revisando tus interacciones en nuestra plataforma y quería ofrecerme como tu punto de contacto.”
In English, it translates to:
“I reviewed your activity and wanted to become your point of contact.”
While grammatically correct, this sounded invasive in Spanish — like we were watching the user too closely. It didn’t feel natural. One reviewer called it “creepy.”
Here’s another example:
- Original English intent: “I noticed you’ve been exploring our platform and expressed interest in connecting with us.”
- Original Spanish output: “Me pareció interesante tu interés en nuestros servicios.”
- Translation in English: “I found your interest in our services interesting.”
Again, it’s technically accurate, but it’s redundant and robotic. It’s the kind of phrasing that makes a reader stop and go, “Did a bot write this?”
The takeaway: Even when the translation is accurate, the tone can be off. And tone is everything in marketing.
The Shift from Translation to Language-aware Prompt Design
At this point, I realized we needed more than AI outputs — we needed a system for guiding the AI to think like a multilingual marketer.
I built a language-portable prompt framework — a structured prompt that could adapt across languages while respecting each one’s unique grammar, tone, and cultural context.
Here’s What Changed
Instead of one static prompt, I broke the logic into variables:
- : Target language (e.g., Spanish, French, German)
- : Pronoun and tone level (“tu” vs. “usted”, “vous” vs. “tu”)
- : Inbox-friendly, conversational, professional
- : Direct vs. suggestive phrasing
- : Enforced where grammar allowed
We also added clear, language-specific rules.
Example (Spanish):
- Use tú consistently, never usted (too formal for our brand)
- Avoid gendered adjectives like interesado/interesada when possible
✅ “Mostraste interés en … ”
❌ “Estuviste interesado en … ”
Example (French):
- Always use vous, not tu, in B2B messages
- Avoid ambiguous endings like intéressé(e)
✅ “Vous avez montré de l’intérêt … ”
❌ “Tu t’étais intéressé(e) … ”
Why This Shift Mattered
In English, a friendly CTA might look like:
“Would you be available for a brief conversation on one of the following days?”
We tried directly translating it into Spanish:
“¿Quieres agendar 15 minutos para hablar sobre lo que estás buscando?”
It was grammatically correct, but it sounded too casual and unprofessional in a B2B context. Not pushy, just slightly off-tone.
So, we reworded it to be friendly but formal:
“Si te parece bien, podemos agendar una conversación breve esta semana.”
This translates to:
“If it works for you, we can schedule a short chat this week.”
Here’s another example in French:
- Original output: “Souhaitez-vous prendre rendez-vous pour en discuter ?”
(“Would you like to schedule a meeting to discuss this?”) - New version: “Auriez-vous 20 minutes pour voir comment HubSpot pourrait concrètement vous aider?” (“Would you have 20 minutes to see how HubSpot could practically support you?”)
The second version adds value to the CTA. Not just time — but purpose.
Backing It Up With a Stakeholder Questionnaire
Localization isn’t just a linguistic issue — it’s a business alignment issue.
To get it right, I created a simple stakeholder intake doc and shared it with marketing ops, regional marketers, and content leads. The goal was to align early on tone, content boundaries, and regional sensitivities.
These are some of the questions I asked:
- What level of formality is appropriate in your market?
- Should we avoid gendered terms?
- Can we reference the user’s company or product usage?
- How direct should we be in asking for action?
- Are there idioms, cultural references, or phrasings we should avoid?
We got some pretty interesting insights.
For example, in some regions, stakeholders preferred not to reference the recipient’s company type in the copy, even though that was common in English (e.g., “I saw that you help startups with HR”).
The localized alternative became more general:
“Entiendo que están buscando formas de mejorar sus procesos internos.” (“I understand you’re looking to improve internal processes.”)
The results of this survey helped create clarity between content, ops, and regional marketing teams — and dramatically reduced our revision cycles.
The Final Product: Human-sounding Emails at Scale
With the updated prompt and intake framework, the new outputs were instantly better.
Before:
- Original output: “Hola [FirstName], soy María de HubSpot. He visto que has navegado nuestra plataforma y parece que te interesa nuestro producto.”
- English translation: “Hi [FirstName], I’m María from HubSpot. I saw you’ve browsed our platform and it seems you’re interested in our product.”
After:
- Original 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 saw you were exploring the platform and wanted to learn more about how we can support your business.”
And stakeholders responded positively:
- “This finally sounds like someone from our team wrote it.”
- “Perfect tone — natural and local.”
- “No gender errors or weird formalities. We can actually use this.”
Even better, we didn’t need to write separate prompts for every campaign. The same core framework now powers AI-generated messages in multiple languages — with consistent quality.
Takeaways for Marketers
Whether you’re working on AI copy, global ads, or multilingual content, here’s what I learned:
1. Don’t just translate — localize for intent.
Literal translations will get you “technically correct” content. But only localization will make it land.
2. Use prompts like creative briefs.
Include tone, formality, CTA style, gender neutrality, and other language rules as variables. Don’t leave nuance to chance.
3. Build language-aware templates.
Languages behave differently. Plan for things like verb conjugations, pluralization, and sentence rhythm upfront.
4. Get feedback early.
Use a stakeholder intake doc before generation, not after. You’ll avoid rework and misalignment later on.
5. Aim for a real, human tone.
If your AI output doesn’t feel like something you would write to a customer, it won’t convert. Read it aloud. Would you hit send?
AI localization is a marketing skill now.
This project taught me something that has stuck with me since: The future of global marketing isn’t just about scaling content — it’s about scaling context.
The companies that succeed with AI won’t be the ones who generate the most content. They’ll be the ones who generate the most resonant content because they know how to prompt for it. And that starts with understanding the languages your customers speak — in more ways than one.