Few consumer experiences are more frustrating than chatting with online customer support. You know, in one of those little chat boxes. Too often it feels like you’re talking to a wall. Is it a language thing? Chat volume?

You’ve probably been airing your gripes to a bot. More and more, companies are saving on labor costs by staffing their website chat boxes with cut-rate algorithmic chatbots. And it shows.

But that’s changing.

Recent advances in large language models are poised to dramatically improve the performance of, well, everything. But, yes, customer-service chatbots.

Using large language models and tools like ChatGPT—you’ve heard of ChatGPT?— companies are stepping up their chat-bot game. Many startups have launched AI-driven chatbots that train themselves on a company’s own data—everything on the website, product specs, customer-service logs, and whatever else may be helpful—creating a customer service tool that not only talks like a human, but is able to dynamically suggest real solutions based on the company’s internal training data. These are not your grandma’s chatbots!

But as incredible as this may sound, if you’re considering hiring one of these AI chatbots to pick up some customer-service slack on your own website, there are a few things you need to keep in mind.

  1. Let users know they’re chatting with a bot: The robot revolution hasn’t arrived yet. Laws like California’s B.O.T. Act prohibit bots from peddling goods or services without disclosing their true robot identity. So, save yourself a legal headache by ensuring this is clearly stated. Need guidance on clear and conspicuous disclosure? The FTC has your back.
  2. Speaking of the FTC … they are concerned about the possibility of “Fake AI”: In fact, the FTC advises that pretending an AI chatbot is human may constitute an unfair or deceptive trade practice. In one such case, a dating website used fake “engager” profiles and pretended they belonged to actual humans. The FTC recommends transparency when using AI tools in customer interactions and data collection. So don’t play the AI masquerade game—solving customer problems is good; tricking them is bad.
  3. Choose your chatbot vendor wisely: There are a lot of different providers offering AI chatbot tools. Before you choose one, review their terms of service and privacy policy like your business depends on it—because it does. [Ed: This is a thing we can help with. Tell them it’s a thing we can help with.] This is a thing we can help with!  Before you sign with a vendor, understand their data collection, processing, and storage practices. Know who they share data with and why. Make sure the way they gather and use data is legal in your state and country. And remember, the devil’s in the details—understand their processes and commitment to monitoring and mitigating risk.
  4. Maintain regular oversight: Just like you wouldn’t buy a car and never take it for a tune-up (right?) don’t think of your chatbot as a set-it-and-forget-it tool. Review chat logs, monitor user interactions, and make adjustments as needed. Look for signs that users are oversharing sensitive information. If issues persist, consider making your AI disclosures and privacy notices more obvious. They should catch the user’s eye like a neon sign in a dark alley, not blend into the background like a chameleon. You should also, of course, make sure the chatbot reflects your company’s services and offerings accurately. If your chatbot starts sounding more like it’s selling second-hand cars rather than your SaaS product, it might need more data for training. Or you might have to hit the pause button and reconsider if a chatbot is the right answer to your customer service prayers. Not every problem needs a high-tech solution.
  5. Disclose how you use chatbot-collected information: As always, transparency is key when dealing with user data. Be clear with your users how you access and process information from the chat service. Explain how the chatbot works within the framework of your website. Use your privacy policy as a roadmap, detailing how user data is handled, and introducing users to your chatbot vendor—the co-pilot on their journey. This allows users to peek into the vendor’s policies and understand the full picture, if they’re so inclined. Be open about how you use and store their information to maintain trust. Make sure your notices are given at or before the point of collection, like a friendly reminder before users embark on their journey with your service. And ideally you’d have users explicitly agree to your policies and the collection of information—like with a cookie banner.

AI chatbots present an enticing customer-service alternative to expensive contractors and the algorithmic bots of yore. But while our virtual assistants may become more and more human-like, let’s not forget that it’s the human touch that makes customer service truly exceptional. And increasingly that human touch is going to involve you monitoring your AI team and making sure they’re playing by the rules.