Article Quick Links:
Every time a chatter asks a question, the bot must decide which Answer to provide. The bot compares the chatter's question to the training questions across all Answers to find the best match. There are three possible outcomes:
- The bot is confident that there is a clear winner amongst the Answers. The bot will respond by providing that particular Answer.
- The bot is not completely confident that there is a clear winner amongst a few possible Answers. The bot will respond with Needs Clarification.
- The bot isn’t equipped with an Answer (or Training) related to the question asked and is not confident that it knows how to respond. The bot will respond with Not Understood.
Every bot has a built-in Answer called Needs Clarification. It is a Locked Answer, meaning it is customizable, but cannot be deleted. The bot will answer with Needs Clarification when the chatter asks a confusing question, such as:
- A vague question or statement (e.g. “Help!” or “I have a question”)
- A double-barreled question with more than one intent that the bot is trained to answer (e.g. “How do I pay my bill on the website or by phone?”)
The most likely Answers are determined automatically and then surfaced as Clarifications. These Clarifications are presented as buttons titled with the Answer Label of the corresponding Answer.
Note: If you do not provide an Answer Label in the Answer Settings, the button text will appear as the Answer Title by default.
An Answer will not be surfaced as Clarifications if:
- It doesn’t have any Training Questions
- It’s a Locked Answer
- It’s toggled ‘OFF’
- The Answer Setting ‘Include in Clarifications and Suggestions’ is toggled off
How does the bot decide what do display?
- The bot will determine which 3 Answers are most likely to be correct.
- If any of those Answers meet the exclusion criteria above, they will not be shown.
- If there is only 1 Answer remaining, the bot displays it as a singular option (i.e. as a Clarification button, not the full Answer content).
- If 0 viable options remain, the bot responds with Not Understood.
Every bot has a built-in Answer called Not Understood. This Answer is triggered when the bot isn’t confident that it can respond accurately to a chatter's question. Not Understood Answer is a Locked Answer. This means that it is customizable, but cannot be deleted. Not Understood should ask the chatter to rephrase their question and give them the option to contact support.
Common examples when the bot is not confident and responds with Not Understood:
- The chatter sends gibberish to the bot
- The question is out of scope (e.g. "Why is the sky blue?")
- The question is relevant, but the content hasn't been created yet (content gap)
- The content exists but the training questions are weak/lacking
|Scenarios that trigger “Not Understood”||Suggested fix|
|The chatter’s question doesn’t align with any Training in the bot, but there’s an existing Answer that could be triggered.||Add the chatter’s question as a TQ to the appropriate Answer in the bot.|
|The chatter’s question doesn’t align with any Training in the bot, despite it being something that the bot should be able to Answer.||A new Answer should be created and trained to address this gap in content.|
|The chatter has asked something that the bot is not expected to Answer, so there is no content (or Training).||No action required - the bot shouldn’t be trained to answer questions that are out of scope.|
For tips on how to train the bot, check out our Answer Training article.
Have any questions? Contact your Ada team—or email us at email@example.com.