This Might Be The Best Spanish Chatbot I’ve Ever Built

How Can You Build a Solid Spanish Chatbot?

Alright, let's get into it. Here's exactly how this came together, and you can also steal this flow for your Spanish bot.

Step 1: Map the flow before you touch anything

The first thing we did wasn't write a single message. It was to figure out what the bot actually needed to handle and build a decision tree around it.

For this, we used WotNot (an AI agent and chatbot builder). It has a cool flow builder, so we mapped everything there.

Here’s a quick screenshot where you can see the branching from the welcome message into different intent paths.

What you're looking at here is a bot with 62 active blocks across multiple intent branches.

Each branch leads to a different experience. Some resolve fully via AI, some check conditions (like whether the user has a specific platform subscription), and some escalate to a human agent.

The key decision here: don't try to handle everything in one flow. Break it down by intent.

If a user asking about billing is in a completely different headspace than someone trying to add a contact in the purchase order module. Treat them differently from the start.

Step 2: Build your AI agent (and actually write the prompt properly)

This is where most people cut corners.

They set up an AI agent, throw in two lines of context, and wonder why it sounds robotic.

We didn't do that.

Look at what's in this prompt. There's a defined role, expected behavior rules, escalation logic, empathy guidelines, and hard rules about what the bot should never do. Like never inventing menu paths or navigation that doesn't exist in the actual platform.

A few things worth calling out specifically:

• The tone instructions are explicit. The agent is told to be amable, cercano, profesional y paciente (warm, close, professional and patient). That's the difference between a bot that sounds like a press release and one that sounds like a helpful colleague.

• The empathy openers are baked in. Phrases like "Entiendo que…", "Gracias por escribirnos…", "Comprendo tu duda…" are literally listed in the prompt. Why? Because without them, AI defaults to cold, transactional responses. With them, the first line of every reply already disarms a frustrated user.

• The hallucination guardrails are tight. The bot is explicitly told: if you don't know, say you're in beta and don't have that information yet. Never make something up. This matters a lot in a platform context where wrong instructions can break a user's workflow.

• The temperature is set to 0.35. Low enough to stay consistent and accurate, not so low that it becomes rigid. For a support bot, this is the sweet spot.

Step 3: Handle the escalation paths cleanly

Not every query gets resolved by the bot. And that's fine. The goal was never 100% bot resolution. It was a smart resolution.

The flow has two clean exit paths baked into the AI agent: Hablar con un humano (talk to a human) and Consulta de ticket (ticket lookup).

When the bot hits its limit, it doesn't hang or loop – it hands off gracefully, in Spanish, without making the user feel like they've been abandoned.

That handoff moment matters more than people think. A clunky escalation undoes everything the bot just built up.

Step 4: Watch it actually work

This is the part that made us exhale.

Here’s a chat where a user asks in Spanish how to add a shipping contact in the purchase order module. The bot responds with a clean numbered walkthrough, entirely in Spanish.

And the bot doesn't miss a beat. It recognizes the intent, pulls the right context, and responds with a clear, numbered, step-by-step answer. In natural Spanish. Warm opener included.

That's not Google Translate. That's a bot that was trained to think in the context of this platform, for these users, in this language.