The Day Three Languages Collided
One Monday morning, I opened the Google Sheet and nearly jumped out of my seat. The Hamsters bot's draft emails showed a pattern: English buyers received casual language like "It's really cool," while Chinese buyers got phrases translated in an overly formal, almost deferential tone. The consistency was completely shattered.
What about the Puppies bot? It was composing a Korean reply and suddenly started inserting emojis everywhere. For formal business communication, this was bizarre.
The Culprit: The Prompt Itself
After investigating, the issue became clear. My instructions to Claude Code didn't specify different tones for different languages. I never explicitly said "English: casual," "Chinese: formal," "Korean: friendly." The bot just mixed and matched whatever felt right in the moment.
The funniest part? Some emails destined for Chinese buyers still contained Korean sentences. The auto-translation feature kicked in, but got confused about what to translate.
**Discovery 1**: In multilingual automation, tone settings matter far more than language selection.
Organizing with a Single Regex Line
I decided to rebuild the bot's email generation logic. The system now determines which language each email uses and whether the correct tone applies to it.
On the Windows desktop in my main Claude Code environment, I added a simple regex pattern:
Language Detection: Auto-insert [EN], [ZH], [KO] tags at email start
Tone Verification: Check if language-specific keywords (formal, casual, friendly) appear
Error Alert: Flag mismatches as "Language Tone Mismatch"
What changed? The bot now self-validates before sending. If it starts in English, it maintains English tone throughout. Chinese emails use honorific language. Korean adjusts to context.
**Discovery 2**: About 80% of multilingual automation errors happen at language-switching moments.
What the Dream Team Learned
After this incident, our dream team (Hamsters and Puppies) established new guidelines:
• **Language Memory Separation**: Even one bot manages distinct language chains separately
• **Added Review Layer**: The generation bot self-reviews before any email gets sent
• **Feedback Loop**: When buyers flag awkward phrasing, we immediately update the prompt
The most surprising result? Our response rate improved noticeably. Trust grew because bot-generated emails became more consistent and professional.
Our bots still make mistakes sometimes, but we laugh and rewrite the prompts each time. That's the fun of automation.