In my script, there is an AI chatbot character - ANNIE - that real characters communicate with via text.

ANNIE has animated reactions and has an artificial speaking voice.

Would I write ANNIE’S dialogue like any other character?

Or would it be more action/description based?

BTW - this isn’t set in the future like the movie HER. So, ANNIE isn’t a super intelligent chatbot.

I want to know how to format/present her dialogue in the script, not how to write it

  • I need some clarification on your question. What would be a more action/description based dialogue? How does it differ from "normal" dialogue?
    – B Altmann
    Dec 15, 2017 at 13:09
  • Why wouldn't you write ANNIE's dialog like every other character's? Dec 15, 2017 at 17:32
  • An example of action based dialouge would be - “Hello. My name is ANNIE. What should I call you?” appears on the screen in the text box. - as if the character was reading a text, book, newspaper etc. Dec 17, 2017 at 23:46
  • 1
    @MarcusMeier Since people kept misinterpreting your question, I added an explanatory line at the bottom. Please feel free to revert it if it is not correct to your intentions. Dec 20, 2017 at 18:27

5 Answers 5


The biggest limitation of most modern chatbots is only replying to the last message stated. They can give great responses, that seem funny or clever, but only to the last sentence said.

So when you write your AI don't look at the conversation as a whole. Look at the last sentence ignoring the conversation and reply to just that. Also don't forget that your replies should know this limitation, and do their best to give generic answers and hide the fact that the AI does not know what you are talking about.

If all else fails, why not try having your dialog with a real chatbot?

  • Up vote for the final suggestion. I think what would be fun is to find a real chat bot, in put the real dialog, and use the output. For lines where ANNIE must deliver a plot critical statement, write that statement (or the jist of it). For everything else, on your first draft, write a filler statement or a (Annie response) notation. For draft two, go to a real chat bot and input the real input chain, and use the result. For important parts, take expected and compare with actual, and if they are close, use actual. Otherwise, use expected.
    – hszmv
    Dec 15, 2017 at 18:52
  • Thanks for the reply! But my real question is: Should I write the AI dialogue in the action - like a book, letter, text - or as a real character? Dec 17, 2017 at 23:53

If it's present day, we're still a long way from anything that would pass a Turing Test, so ANNIE is likely to be defined as much by mistakes and overly generic answers as by anything that would normally be recognised as a character - something more like Siri/Alexa/Cortana than Scarlett Johansson.

If ANNIE is commercially backed, you could also work in spurious and irrelevant advertising.

Even if we're assuming someone could come up with something a lot closer to a character, I would think the responses are as likely to be memorable when they don't do what's expected as they are when they do.


I would write it as dialogue, but avoid (as the narrator and writer) attributing any emotions, humor, or subtlety to the character's text messages. Annie does not say things cautiously, angrily, or allude to things, or skirt the subject. She does not truly understand what is going on, she is a shallow simulation. she cannot be embarrassed, or understand that some topics are embarrassing. She cannot keep a secret unless explicitly told to: She does not understand, like a human, what is a private communication and what is acceptable to reveal to others.


One very useful trick for writing characters who are somewhat dumb or think differently is to change your vocabulary. ANNIE being a bot, I suggest you to opt for more mechanical vocabulary and to limit the words you use to the bare minimum.

Since the story is set in 2017-18, and you established ANNIE is not terribly intelligent, I think it is safe to assume she does not really think. Her reasoning processes can be brought down to simple comparisons, linking between words in search of meanings and simple looping/iterating. Another very important characteristic is that current chatbots use only what the user types in as input. While some of the best chatbots may look up information on the web or other sources, they are unable to see the user's facial expressions or hear their voice.

Here is a small extract illustrating some of those ideas for you:

waiting for the user

-- Err... hello?

reticence. User sounds hesitant. Initiating comforting protocol

-- Hello, user! What is your name?

-- Mark

-- Are you feeling well, Mark?

User is not responding

Are you OK, Mark?

-- I kinda of feel down

checking dictionary for "to feel down". To feel bad, depressed

-- Why do you feel bad, Mark?


I would inspire myself to existing chatbots These are just two examples that came to mind, I hope they help:

  • Emacs psychotherapist a rather old one, asking questions on the previous statement with the purpose of making the user think about their issue
  • A.L.I.C.E. a more recent one, not particularly intelligent in any way, yet more natural than the one above

It seems to me that the common feature is that they lack emotional response. They can be very articulated, evolve a conversation taking into account a long history of sentences, but, in my opinion, they lack empathy.

"I am sorry to hear that" answers anything sad.

"I am happy to hear that" answers anything happy.

A human being could just cringe, or change topic, or modulate their emotional response throughout the conversation. It is entirely possible that a human character would try to place some distance after hearing five consecutive disgraces that occurred to the other person. A chatbot may remain in their static emotion, repeating the same type of response, or cycling through a predefined "optimal" set of answers. It does not have to feel clumsy, just eerily detached.

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.