Death to Chatbots

I’m ready to retire the word “chatbot." To refer to ChatGPT and its kin as a "chatbot" seems, at best, a mild insult and, at worst, like calling a tiger a "house cat."

The best language we have for what’s really going on in the chat space right now are technical words: Older chatbots are “deterministic” (following explicit pathways to resolution that have been predefined and pre-approved) while newer ones are “probabilistic” or “generative.” But the average person doesn’t know what that means.

However, the average person can feel the difference: old chatbots can’t write haikus (semi-accurately), create parodies in the style of Monty Python, or devise choose your own adventure tales. When a classic chatbot pops up in the corner of most websites (“Hi! Can I help you?”), it feels like those overeager sales staff hanging out at the front of the store (maybe it’s the cusp-millennial in me, but please just go away).

Yet when ChatGPT arrived on the scene, 100 million people signed up in two months. There isn’t a chatbot in the world that saw that kind of adoption. Ever. So for brands willing to create these newer, generative chat experiences, making them appear to be classic chatbots undersells what they’re offering — and could have people hitting “X” before they even realize what it can do.

On the flip side, continuing to offer up deterministic chatbots is also a mistake. Those 100 million people who signed up to ChatGPT (back when it was GPT 3!) now know there’s something better. So when they’re presented with limited options and canned answers to everything, they’re going to stop engaging and—gasp—call a call center.

So here’s a proposal for 5 signals you should add to smart chat experiences so people know the difference between the dumb ones out there and your brand’s intelligent option.

1. Re-examine where chatbots should live. Don’t assume every chat capability should always be stuck in the lower right Clippy corner (Microsoft’s disastrous but meme-worthy chatty Office helper from the late 90s). Make it more contextually available around the topics it’s trained to know best. For example, if your experience offers conversational help, then think about embedding it in tool tips or onboarding. Or why not make your site search bar simply the start of a conversation?

2. Call out its “intelligence” at the start of the conversation. When generative AI is under the hood, it’s good to signal just how flexible and open-ended it can be with language like “Ask me whatever you want to know about X, Y or Z.” However, don’t ignore pre-set options altogether. Those will help people understand the parameters and boundaries of what the chat experience is offering and make it easy to engage if they’re not yet sure what they want. A blank text box is still daunting.

3. Stop following messaging best practices. Think beyond text. For example, if site search is how someone accesses a generative capability, think about all the ways the “results” could be presented. It could be someone’s own personal landing page, with smart modules that present the answer in a skimmable, more visually rich form. Or maybe it’s rich audio, such as a generated podcast or audio tour. Imagine that site browse filters could be described in the person’s own words instead of selected from dropdowns. The list goes on!

4. Always indicate it’s an AI. With dear old chatbots, people could tell pretty easily when something was human or AI. Those deterministic chatbots tended to talk too fast, too formally, and went on the fritz when someone asked an offtopic question (or an on-topic question in a way it wasn’t expecting). But now, it’s actually possible to fool someone without their knowledge. Simply by cautioning that something is AI gives a good signal to the end user that they’re dealing with the more generative, useful kind.

5. And, of course, don’t publicly call it a chatbot. That means it shouldn’t introduce itself as “Aardvark, your friendly chatbot.” But what can we call it? Maybe it’s a personal agent. A chat assistant. An answer engine. Just please, call it anything BUT a chatbot. Just like we outgrew the term “carriages” when cars came along or “electric typewriters” when personal computers arrived, it’s time to recognize “chatbot” for what it’s becoming: a relic of the past.

Agree? Disagree? I’ll be speaking on why we need to get past the narrow view of generative AI as chatbots in my talk “Removing Chatbots From the Conversation” at the Voice & AI summit in DC on October 30th. Join me!