Suggested questions
The starter prompts a user sees on an empty chat screen. Configured on the Suggested questions tab of an assistant.
Why they matter more than they look
A blank chat box is the single biggest obstacle to adoption. People who have never used an AI assistant often do not know what it is reasonable to ask, so they ask nothing, or they ask something the assistant cannot help with and conclude it is useless.
Three or four good suggested questions solve this. They teach the scope of the assistant in the time it takes to read them.
Adding questions
Each question has two parts:
Head
The short title on the card. This is what people scan.
Detail
A supporting line explaining what the question covers.
Choose Add question to add an empty row.
Fill in Head and Detail.
Repeat for each question.
Choose Save questions.
Remove a question with the red bin button next to its row.
Generating a set automatically
Generate questions asks the platform to write a set based on the assistant's instructions and purpose.
Generated questions are a decent starting point, not a finished product. Read them and edit. They reflect what the assistant was told it does, which is not always what users actually want to ask.
Writing good ones
Use your users' words, not internal terminology. "How much leave do I have left?" beats "Query leave balance entitlement".
Cover the range. Pick questions from different parts of what the assistant does, so users infer the breadth of it.
Be specific. "Can I carry unused leave into next year?" teaches far more about the assistant than "Ask about leave".
Three to five is plenty. More becomes a wall of text people skip.
Revisit them. Once the assistant has been live for a few weeks, look at the chat history to see what people actually ask, and make the real questions the suggested ones.
Last updated
Was this helpful?