"Make it sound more human" is one of the most common requests for any chatbot project - and one of the vaguest. In practice, it breaks down into a handful of concrete, fixable things.
Corporate phrasing and long, hedged sentences are the fastest way to make a bot sound like a bot. Real people speak in short sentences and use plain words - the writing behind your chatbot should too.
A chatbot that's formal in one answer and casual in the next feels inconsistent in a way people notice immediately, even if they can't say why. Pick a tone - warm, professional, playful, whatever fits the brand - and keep it consistent across every response.
People do this naturally in conversation ("Good question - let me check that") before giving the actual answer. A chatbot that jumps straight to a dry answer every time skips a small social cue that humans rely on.
A bot that never says "I'm not sure" isn't more capable - it's just more likely to confidently give a wrong answer. Being willing to hedge, ask a clarifying question, or hand off to a human is one of the most human things a chatbot can do.
Repeating a question the user already answered a moment ago is one of the fastest ways to break the illusion of a real conversation. Basic context-tracking across a session goes a long way.
Even a well-written, well-behaved chatbot is still limited by the medium: text alone can't carry tone of voice, expression, or timing. This is where an interactive avatar changes the equation rather than just polishing it - the same well-tuned AI, but delivered through a voice and a face that can pause, react, and make something closer to eye contact.
This isn't an either/or choice. The writing and behaviour work above still matters just as much once there's a face attached - arguably more, since a bad answer delivered by an avatar feels more jarring than the same answer in a text box.