At a recent Japanese Design meetup in San Francisco on the subject of chatbot and conversational UI, the topic of humour (or humor to localize the term) came up. We're all familiar with the "Alexa, tell me a joke" chit-chat that surrounds the intent, utterances and prompts of a chatbot conversation, but can do we determine which humour works for which audience?

For example, I asked if humour was acceptable in a chatbot in the Japanese financial sector, and the response was, "Yes, if it's professional humour..."...:)

Any guidelines, or is down to good old Wizard of Oz research and a talented voice or script artist with a background in global comedy improv?

MyPoV: Avoid humour altogether...

3 Answers 3


Unless your chatbot is purely for entertainment, or low-pressure functions, I'd be very cautious.

Humor (unless explicitly requested by a "tell me a joke" command) is an incredibly subtle thing when you're dealing with

  1. text (how many times have you misinterpreted the intent of a text message or email from someone—even if you know them quite well—because text just can't communicate emotions and the full range of expression that a human voice, facial expression, or body language can)
  2. a computer (as a user of this system, I may not expect or even think possible that a computer would have a sense of humor, thus I may be predisposed to taking everything it says literally)

To counteract this, you could try to give your chatbot a very casual personality by choosing to use colloquial vocabulary, some slang, or even have it poke fun at itself ("Sorry, I don't know what you mean because the Broca's area in my frontal lobe is still developing. Would you mind rephrasing the question?"). These might signal to the user that the computer has a "personality" and possibly also a sense of humor.


You could try something like this (modified from Microsoft documentation web page):

Response box from Microsoft documentation page

  • 1
    Surely jokes are funny, not helpful...
    – Midas
    Commented Oct 27, 2017 at 18:12
  • @Midas Why put something in to a computer system that is not helpful?
    – user67695
    Commented Oct 28, 2017 at 1:42

Yes you could measure it by letting your audience rate different jokes (for example if someone responds "haha" or "lol" in the chatbot you could use this instead of doing a classic survey regarding the joke)

The artificial intelligence would need to adjust accordingly to the target group which of course is really hard (technically).

When it comes to stuff like this i would always look at Amazon/Apple/Windows/Google.. you will see that even their solutions right now don't really offer something like this. Of course you can ask them for a random joke but they will not learn how you liked this joke.

Right now i would avoid using humor in sectors like financial, or at least keep it to a minimum.

  • Is it true that cannibals won't eat clowns because they taste funny?
    – user67695
    Commented Nov 29, 2017 at 17:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.