Variety is a basic human need.
Users love variety. Who doesn’t? When you’re talking to someone, you don’t want them to say the same thing every single time you see them. Or worse, every single time they open their mouth.
Vary skill messaging to encourage retention
For this reason, it’s important to invest in varying the messages and reprompts in your skill. Even 3 different options from which one is randomly picked is better than hearing the same 1 thing over and over again.
Nick Schwab is the successful founder of Invoked Apps, which has released over 40 sounds skills (Ambient Sounds, Sleep Sounds, etc.) for Echo and Google Home. In this voicebot.ai podcast, he says that even for his extremely successful skills, retention is only around 20%. This means that for a new skill, you need to do everything you can to encourage retention.
Differentiate new user experience (NUX)
One way you can do this is by differentiating new user experience from not-new user experience. For example, in my skill Moon Calendar, I say the following to a new user:
"Welcome to Moon Calendar. [a bunch of introductory material] Say help to discover some more things you can ask for."
The introductory material is a chance for me to tell the user how my skill can help them. And the "Say help" message helps the user understand how to navigate the skill.
For non-new users, I say:
"Welcome to Moon Calendar! Say [the few most popular utterances] to get started."
In this way, the experienced user doesn’t have to get annoyed by the long intro, the suggestion that they say “help”, and the fact that I wasted their time.
You can use persistent attributes to do this.
Put yourself in the user's shoes
One way to see whether your skill messaging is varied enough is to use your skill yourself. Use your skill every day for 7 consecutive days, and see whether you yourself get annoyed. If you find yourself pleasantly entertained, then your skill messaging is probably varied enough.
Design the conversation
Don't just type something without thinking about it. Putting the effort into imagining the conversation between your skill (yourself) and the user on your first attempt will reduce the total time it takes you to iterate to a nice result. It’s worth it.