Designing chatbots that actually help: Lessons from the frontlines of Conversational UX

What does it really take to build a chatbot that’s helpful, human, and on-brand?
In a recent conversation with Marjorie Allen, Conversation Design Lead at Campfire AI, we unpacked the principles behind great chatbot and voice assistant experiences. Drawing on her extensive work in the field - from telco to transport - Marjorie shared what separates clunky bots from those that genuinely support users.

If you’re a product manager, CX lead, or digital channel owner wondering how to elevate your chatbot strategy, here are six sharp takeaways from our session.

1. Start with purpose, not prompts

“AI is not the chocolate sauce on your sundae. It’s the ice cream.”

The biggest mistake? Starting with technology instead of intention. Whether you’re designing a script or setting up a generative AI bot, your first task is to define the purpose: what does this conversation need to achieve, and why? Skip this, and you’re just decorating a broken experience.

Why it matters: Without a clear goal, your bot will confuse more than help. Teams need to align early around what success looks like - before diving into tooling.

2. Don’t write, orchestrate

“It’s not just about bubbles of text anymore - it’s orchestration logic.”

Whether you’re scripting classic decision trees or using generative AI, the core remains the same: piece together the puzzle of user needs, inputs, and context. But with GenAI, you’re no longer writing responses line by line - you’re designing instructions and fallback structures that adapt on the fly.

Why it matters: Teams need new skills. Copywriters become prompt engineers. Designers think in flows, not lines. It’s a shift in mindset.

3. Tone is a tool - use it wisely

“Define your tone, but match it to the moment.”

Marjorie insists on clarity when setting tone of voice: choose three adjectives, a name, pronouns, and a one-sentence mission. Then, adapt. If a user is frustrated, don’t hit them with emojis. If a situation is serious (like a suspended account), ditch the fluff.

Why it matters: Consistent tone builds brand trust. Adaptive tone builds user trust. Great bots do both.

4. “Not understood” isn’t a failure - if handled right

“Fallbacks aren’t failure - they’re a second chance to help.”

Every chatbot should have a fallback hierarchy: detect gibberish, shield from trolls, suggest relevant articles, or escalate to a human. But the real danger? False negatives - when the bot says “I don’t know,” but actually does know. That breaks trust fast.

Why it matters: Good fallback design saves face and solves problems. It’s not just about what the bot says - it’s how it recovers.

5. Channels are not copy-paste zones

“You wouldn’t run your TV ad on Instagram. Same logic applies here.”

Voice isn’t text. WhatsApp isn’t your website. Each channel has its quirks: latency on voice, brevity on chat, interruption handling everywhere. Design accordingly - or lose users in friction.

Why it matters: A seamless experience across channels means tailoring, not cloning. Your users expect it, even if your stack doesn’t make it easy.

6. Accessibility and common sense are non-negotiables

“Design for all, or risk fines - and users.”

With the Accessibility Act coming into force, teams must think beyond colours and buttons. Can users navigate your chatbot with a keyboard? Is your web widget screen-reader friendly? Accessibility is compliance, yes - but it’s also care.

Why it matters: Inclusive design is good design. It’s also smart business.

Wrapping Up

From playful emojis to fallback flows, the conversation with Marjorie reminded us that great chatbots aren’t born, they’re iterated into existence. They start with a purpose, learn from user signals, and adapt across channels.

The big question for your team: is your bot designed to truly listen, or just respond?