Chatbots are all over the place. They’re in your Facebook Messenger. They’re answering your customer service questions or helping you organize legal cases. Chatbots and other natural language interfaces are here, and they’re here to stay.

App usage is declining, with most users spending 50 percent of their time in only one app — and that app is often a messaging platform. In fact, 75 percent of all smartphone users spend time on at least one messaging app. Accessing customers through that messaging interface is the new holy grail, and brands no longer need to be tech giants to get their own chatbot. Many companies are democratizing the process. MindIQ lets any company create a bot with just a few clicks, while MotionAI allows the creation of complex choice trees, all without any understanding of programming or machine learning.

That means every company, from small businesses to startups, needs to understand the dos and don’ts of creating these digital helpers. While the rules are still being written, here’s what we’ve learned from both creating and interacting with our new best friends.

1. Be clear about your bot’s purpose

Just because bots are the hot new thing doesn’t mean your company needs one. You have to be clear not only what your bot is, but also why you made it. In some cases this means ensuring that the team have all set their compasses to the same north star, but it can also mean being sure that purpose is communicated to the user. If you’re changing a previous flow, explain why: Maybe this new process will save time, or provide a better overall experience.

2. Make small asks to gain buy-in and push users through the flow

Onboarding is onboarding, whether it’s a webform or a bot, and the same rules apply. Don’t think because this is conversational, you can get carried away about how much information to ask for. If I ask for your email, home address, phone number, and favorite color all at once, you’re likely to click away. But, if I ask you give me your email, and then ask for your favorite color later on, you’re a lot more likely to tell me. This is the best practice with almost any sign-up flow, and why mess with a good thing?

3. Use information you already have

You don’t need to start from square one. If you want to ensure your bot has a similar voice to your brand, you can use marketing collateral as copy for your bot’s conversations. You can also pull information from existing resources in order to give the bot the information that allows it to quickly deliver value: Don’t ask users to enter their entire sales pipeline when you can just integrate with Salesforce.

4. Avoid context switching, especially during onboarding

There will be times when your entire onboarding flow cannot take place inside of your chosen platform (Slack, Facebook Messenger, etc.). Perhaps you need the user to link an account like Google Calendar or authorize a third-party service. In these situations, it’s best to handle authentication on the web when they’re adding the bot to their messaging platform, before they begin interacting with it.

5. Give the option to onboard at a later date

Users will need to be onboarded, but often they’ll complete half of the process and then return at a later date. With traditional interfaces, this usually meant push notifications or emails sent over and over until the user either responded or deleted the app. Because bots are a more “natural” interface, you have the opportunity to extend how you interact with your users. Treat it like a real conversation: Give them options. You can let users pick the process up later, but you can also schedule a specific time for them to finish – just like you’d schedule an onboarding meeting with a new hire. This could be done with a calendar invite or by sending them a friendly reminder message.

6. Be personable

Half of what makes users seek out bots is that, for now, the novelty of the whole thing makes it fun. This is exponentially increased when the bot has something of a personality. How many times have you pulled out your phone to ask Siri what her favorite joke is, or when winter will come? Make users want to respond by giving them more than a bland conversation tree.

7. Make the options for input clear

Every time a user has to stop and consider what you might have meant, you increase opportunities for failure. Unless you’re using some advanced NLP, try not to ask a user an open-ended “Is this correct?” Instead try something like “Is this correct? Type ‘Yes’ or ‘No’.” You could also go Google Assistant’s route with pre-selected responses users can choose from.

8. Minimize the amount of input required

If texting has taught us anything, it’s that humans love to shorten things, even things that don’t really need to be shortened because they’re already three-letter words. Don’t make your user type “yes” when they could type “y” or when a number from a list will do. Anticipate and reduce the input.

9. Don’t pester users

No one wants to be harassed by salespeople; the same goes when those salespeople are electronic. Every interaction your bot has with a customer is an interaction with your brand. Make sure to limit notifications, and if customers stop responding, try a different tactic. If you’re going to be the one initiating contact in the future, make sure the user knows that going in; make the expectations clear on both ends.

10. Fail gracefully and proactively

There is nothing worse than screaming into an automated phone tree, “HUMAN! HUMAN!” while desperately mashing the 1 button. The same is true with bots. You don’t want your user to reach the end of the rope before you offer them help — you want to offer help proactively. If the bot doesn’t understand the user three times in a row, for instance, it should automatically give them the help command’s output. If they input a lot of swear words, that’s probably a sign that they’re fed up and could use some human assistance.

11. When possible, give users an out

Provide mechanisms for direct feedback or support within your chosen medium, such as letting them type feedback to send or connecting them to support staff. But if they don’t want to use the bot, they should able to disengage. Be prepared with an escalation process for anything critical to your organization; just make it clear what will happen if the user chooses to entirely ignore the bot prior to escalating.

As bots and natural language interfaces become ubiquitous, the laws that govern them will become clearer. But if you follow these commandments in your early work, you’ll be sure not to alienate your user base before they get used to the new world order.

Get more stories like this on TwitterFacebook