So you’ve decided to add chat bots to your website in order to improve your customer service, reduce costs, and drive sales more effectively. Using a platform like Formilla it’s easy to set up an AI-based Helper Bot to perform basic customer interaction tasks for you without worrying about any coding. But chat bots are only as effective as what they’ve been programmed to say. In order to get the most out of your new chat bot you need to carefully consider both what the bots will say and how they will say it.
In this article we’ll discuss best practices for chat bot response scripting. We’ll also offer some beginner tips and advanced guidelines in order to make your chat bots as effective as possible.
If you’re starting from a totally blank slate with a chat bot, it might be tricky to figure out what responses you need to write first. How can you possibly cover all the needs your site visitors and customers might have without writing hundreds of different responses for your new bots?
In reality, it’s likely that a lot of the work you’ll need to do has already been done for your business, in one form or another. Many companies have a “frequently asked questions” document, either live on their website for customers to view or for internal customer service training purposes. These are exactly the questions you should start with when writing for your chat bot. Even if you don’t have this FAQ document to reference, anyone in your company with experience in customer-facing roles can likely write down a few broad categories of common questions and issues they encounter.
If necessary, gather your customer service experts together for a meeting, and try to come away with a dozen questions your chat bot should be able to answer right from the start. We’ve covered suggested basic questions in detail before, if you need some examples to help you get started.
Once equipped to answer the most common questions it will encounter, your chat bot can start helping your business right away. It might only be able to field 25 to 30% of customer interactions without assistance at first – but that still represents a substantial savings of time and money for your customer service team. And once the bot is live and interacting with customers, you’ll get valuable chat transcripts that you can use to guide future responses you write for your chat bot.
Chat Bot Scripting Examples: The Basics
What is the image that you want to portray with your business messaging? Is your brand casual and jokey, or polished and professional? Do you want to be seen as a leading authority in your specialized field, using industry jargon and terminology? Or do you want to be a brand that makes your industry accessible to a wider audience, using simple and universal language?
For many companies, Brand and Marketing leaders will have already considered and answered all of these questions. It’s important to view chat bots as an extension of these overall company messaging strategies, not an exception to them. Because they are interactive and highly visible, the way your chat bots communicate will be a lot more noticeable to customers than a static advertisement might be. So you’ll want to make sure the way your chat bots communicate with new and returning customers is consistent with your overall company image, to avoid a potentially confusing and brand-damaging disconnect.
For example, if you run a hip coffee shop catering to a young crowd, and you use emojis and internet slang in your marketing materials and on social media, there’s nothing wrong with continuing that same style of communication via live chat or a chat bot on your site. In fact, your fans will likely expect it. However, if you run an investment bank that projects authority and stability with your branding, the use of a laughing emoji in an on-site chat with a client would seem very strange, and could undercut the brand image you’ve worked so hard to cultivate.
Chat Bot Scripting Examples: Brand Voice
The longer and more complicated your sentences are, the more chances there are for your meaning to get lost or misinterpreted. And this is especially true via a chat interaction, when part of your site visitor’s mind is focused on their specific issue or question. In the case of regular online content, like this article you are reading right now, you can expect a certain degree of focus from your readers. But a chat is collaborative, and your bots will be interacting with humans who are spending part of the time thinking of how to phrase their next message. In these cases, short and direct sentences are the best way to make a conversation flow smoothly.
Wharton Magazine identifies short and clear sentences as one of their six tips for effective business communication, and their general rule is to “watch out for sentences that go on for three or more lines.” That’s a solid rule of thumb, though it doesn’t apply cleanly to the cramped confines of a chat widget. When drafting responses for your chat bots, become familiar with the relationship between the number of characters in a response and how many lines a response takes up in chat. For a typical chat widget, a single sentence probably shouldn’t go much further than four lines.
If writing short chat bot messages is proving difficult for you, look for ways to break your long sentences into simpler ideas. If you find yourself using multiple commas in a response, there’s a good chance it can be split into separate messages that are easier to read. And some live chat software will support bullet points, list formatting, and other methods to communicate certain types of information more clearly in chat.
Chat Bot Scripting Examples: Short and Clear Sentences
One of the reasons site visitors can sometimes be annoyed with chat bots is that the bots don’t seem to take “No” for an answer. If someone on your site has decided they aren’t interested in interacting with your bot, no amount of prodding on your part is going to change their minds. In fact, you’re much more likely to irritate a visitor with an insistent chat bot than you are to convince them that they should give it a chance.
No, it doesn’t matter if your bot is funny or cute or actually very helpful. We know, you worked very hard on your bot, and you just want to be able to assist your site visitors. The truth is, no matter how good your bot may be, some people just aren’t going to want to interact with it – and forcing the issue can only have bad results.
Don’t worry too much about particular individuals choosing not to engage with your chat bots. But if you notice that your bots have consistently low engagement, then you’ll need to diagnose the reason for that. Are people ignoring it, or actively closing it? Could a notification sound effect help? Or maybe you need to revise the opening message the bot delivers to make the benefits of interacting with the bot more clear. Whatever the issue is, your focus should be on offering straightforward help with your bots when it’s welcome, and getting bots out of the way of site visitors when they have indicated they aren’t interested.
No matter how well-trained your AI-based chat bot is, it’s not always going to be perfect. There will be questions it isn’t able to answer, even after repeated attempts – and that’s why you need to make sure your chat bot knows when to admit its limitations. With Formilla you can tweak this setting, establishing how many “I’m sorry, I didn’t understand you” replies your bot will deliver before it gives up and tells the site visitor that it can’t help. At that point in the conversation it’s standard practice to have the chat bot inform the visitor that it will attempt to connect them with a human live chat agent, so that your customer (or potential customer) isn’t left hanging and frustrated.
These “failures” are actually an essential part of the chat bot training process. By reviewing chat transcripts regularly, you’ll get insight into the kinds of questions your customers have that your chat bots aren’t yet able to answer. Sometimes the questions will be so specific or so complicated that only human interaction will be able to resolve them, but in many cases these failures are just indications of responses that you still need to write. This is how AI-based chat bots grow “smarter” and more effective over time, as you can specifically expand their abilities exactly where you have identified needs.
It’s natural to want to give your chat bot a personality, and in most cases that’s a good thing. However, one of the biggest potential pitfalls of a chat bot comes when its responses clash with the feelings of its human conversation partner.
Think about your own interactions with customer service representatives in the past. How often in those situations have you been angry, or frustrated, or confused? Probably pretty often, right? Now just imagine how you would have felt in those situations if you had to deal with a chat bot that was making jokes, and seemed totally oblivious to how you were feeling.
Chat bots that respond with no consideration for emotions give the whole chat bot industry a bad name. There’s nothing inherently uncaring or cold about chat bots – they just need to be properly designed and deployed. Think of bots like greeting cards. A well written card given at the appropriate occasion can be helpful and touching. But giving a funny birthday card when a condolence card would have been more appropriate feels much worse than not giving anything at all.
Chat Bot Scripting Examples: Emotional Responses
A/B testing is the practice of showing multiple versions of something to an audience and seeing which one is preferred. It’s a common practice in marketing, as A/B testing advertisements or landing pages can show you which version has higher engagement or conversion rates. And while it’s possible to do live A/B testing with a chat bot on your site, it’s easier and cleaner to develop an internal “focus group” to help you judge the proper way to phrase responses for your chat bot before you release those responses into the wild to be presented to your customers.
Chat bot responses are (usually) purely text, so all readers will have to go on is the words you use and the order in which you present those words. But written language can be a subtle thing, and it’s all too easy for something that one person writes that makes sense to them to confuse a particular reader, or to be misinterpreted. Don’t fall into the trap of assuming your audience will automatically understand your intentions, especially since you’re likely to encounter customers from different language backgrounds, people who are multi-tasking, and new site visitors who have no understanding of your business or the jargon associated with it.
When evaluating the different ways that a written response can be interpreted by people, there’s no substitute for actually showing that response to other human beings. You can gather A/B test data by coming up with at least two possible ways of writing each new chat bot response, and putting those in an email or a survey form you send out to your internal focus group. Ask your team to choose their favorites, and to describe the difference in meaning, if any, between the differently phrased responses. You might be surprised at the answers you get, and the data will help you get better at writing scripted responses for your chat bots in the future.
If you’re ready to get started with customizable chat bots on your own site, sign up for a free trial of Formilla today. You can use the Formilla platform to add chat bots and live chat capabilities to your existing website via a plugin or app.
AI-based chat bots can help you with customer service, sales, or marketing tasks. For more on what chat bots can do and how to get the most out of them, check out our full chat bot content category.