Plan requirement
| Subscription | Any plan |
| Also required | AI agents - Advanced |
| Access | Admin |
Shaping how answers from your articles are presented: length, links, and what happens when confidence is low.
Adjust it
- Open the knowledge reply in the AI agent's replies.
- Set how the answer should be presented.
- Decide whether to include a link to the source article.
- Set what happens when nothing suitable is found.
- Test with questions of each kind.
Include the article link
Almost always. The summary answers the immediate question; the link lets somebody who needs the detail get it without asking again.
It also gives the customer a way to check, which matters more with a generated answer than a written one.
Keep the answer short
Two or three sentences, then the link. A long generated reply on messaging is skimmed, and skimming a procedure means missing a step.
Say when it is unsure
An answer hedged with "I think" is more useful than a confident wrong one, and it prompts the customer to check the article or ask for a person.
What you want to avoid is the reply that sounds certain and is not, because nothing about it invites verification.
Set the low-confidence behaviour deliberately
When nothing suitable is found, the agent should say so and offer a person. Attempting an answer anyway is where the worst transcripts come from.
Check it against a hard question
One that is only partly covered by an article. That is the case where these settings show their effect, and where a default configuration usually does the wrong thing.
Comments
0 comments
Article is closed for comments.