ChatGPT for conversation designers: editing dialog

ChatGPT is taking the internet by storm, but for some, it hasn’t gained much traction in the workplace. Given the hype, it can feel anticlimactic. If you’re still puzzled on how to put ChatGPT to good use, I’m documenting my ChatGPT journey and sharing use cases that I particularly enjoy as a conversation designer.

In this article — and a few more to come — I’ll recap my explorations so that other conversation designers might find inspiration on how to use ChatGPT in their own work.

Many of the conversation designers I’ve spoken admit that they’ve struggled to come up with ways to use ChatGPT productively, let alone envision a world where it strips us of our corporate badges. That’s because it simply can’t do what we do — design with thoughtfulness, intention, accuracy, reliability, and adept problem-solving skills.

ChatGPT is a tool, not a replacement

ChatGPT is a generative AI. It parses through text, outputs conversational dialog, and is only racist sometimes. As humans, we’re generally better at all of those things and more. We’re just painfully slow in comparison.

But it’s still a pretty nifty design tool.

If you’re still feeling sus, remember: ChatGPT is only as creative as you are. And as a designer, you’re outstandingly creative. So don’t feel bummed if it’s not working for you yet — remember the first time you looked at a JSON file? Or was that demoralizing for just me? Anyway, once you’ve familiarized yourself with what it can do, you’ll find creative ways to tailor those capabilities to your specific needs.

For now, I wanted to focus on one specific use case — using ChatGPT to edit dialog. As always, depending on what kind of design you do and for whom, your mileage may vary.

Challenges of editing conversational dialog

One of the funner parts of our job as conversation designers is creating our AI’s actual responses. Writing sample dialogs, seamlessly connecting sequential dialogs within a workflow, crafting precise responses in hyper-specific contexts. This can also be a tricky and mentally exhausting task when there are so many factors to consider, like:

• Maintaining a voice and tone that align with your AI’s persona
• Using verbiage that’s consistent with what’s in your densely populated, hard-to-navigate CMS
• Not sounding like an incompetent, helpless buffoon in error scenarios
• Not writing too much
• Ensuring whatever we’re writing sounds good across modalities, and not just in our heads
• Challenging biases that affect the language we use and the users who receive it
• Maintaining the good graces of your Legal & Compliance Overlords, complying with whatever policies they unsheathe from their black, velvety cloaks that prevent you from using certain terminologies, or force you to use others, depending on risk and business contexts. (Yeah, I used to work at a bank.)
• Frustrating your boss because it’s taken so long and you still haven’t pushed your work through content review yet. They probably aren’t feeling that way, but not being able to tell is the worst.

Now, you can provide ChatGPT with your draft dialog and ask it to help with any of the above points — except for the last one, that’s a you-problem solvable by using ChatGPT more efficiently and taking a chill pill.

Editing conversational dialog with ChatGPT

Here’s a sample prompt you can try out right now:

Rewrite this dialog to make it sound friendly, helpful, and polite, and keep it under 300 characters:

The Food and Nutrition Board (FNB) defines the Tolerable Upper Intake Level (UL) as the highest level of daily nutrient intake that is likely to pose no risk of adverse health effects to almost all individuals in the general population. This level is different for each nutrient. To view the UL for Vitamins and Elements (also referred to as minerals or electrolytes), please refer to the Tolerable Upper Intake Levels table from the FNB.

I pasted that into ChatGPT– here’s the peppy response I got in return:

Hi there! The Tolerable Upper Intake Level (UL) is the highest daily nutrient intake that is unlikely to cause any adverse health effects. This varies for each nutrient, and you can find the UL for Vitamins and Elements (minerals or electrolytes) in the FNB’s Tolerable Upper Intake Levels table. Check it out to make sure you’re consuming safe amounts of these nutrients!

Overall, I’d say it did an OK job at rephrasing my original text. Realistically, I’d prompt a few more times until I got something I was comfortable working into my Voiceflow or Figma boards. (Reminder: you only get out of ChatGPT what you put into it.)

Refine, reprompt

Try spicing your prompt with one of these:

• Provide other sample dialogs that you’d want your new response to sound like
• Set a character maximum and/or minimum for your new response
• Identify a specific goal you want the response to accomplish
• For an FAQ answer, list the info or steps you’d like to see condensed
• Ask for variations ✨

Here’s another example:

Create a new dialog under 400 characters in length that describes how to sign up for an online cooking class. Use the following paragraph as an inspiration for the tone of your dialog:

Hey! I’m Trish, your nutrition assistant. I’m here to tell you all about nutritional foods and facts, share healthy recipes, and motivate you to take care of your body — starting with what you put in it! You can ask me questions like, “what allergens are in Caesar salad dressing?” or “How much protein is in bowl of lentil pasta?” Or, pick a topic below to kick things off.

And here’s what I got back from ChatGPT:

Hi there! I’m Chef Sarah, and I’m here to help you sign up for an online cooking class. Follow these simple steps:

1. Browse our selection of classes and choose the one that interests you.
2. Click the “Sign Up” button and create an account.
3. Choose your class date and time.
4. Enter your payment information to complete your registration.

Once you’re signed up, you’ll receive an email with all the details and instructions for the class. Can’t wait to see you in the kitchen! Let me know if you have any questions.

Not 100% what I was expecting, but I can extract the parts like about it and ditch the rest. It’s important to cherry pick and do the fine-tuning yourself, otherwise you’ll get called out. 👀

Tip: avoid ChatGPT changing the meaning of what you’re writing. So stay focused, keep your original copy or design requirements at hand to make sure you’ve hit the mark instead of missed it.

Another cool way to iterate response dialog is to play around with personas. In my next article, I’ll talk about plugging personas into ChatGPT and using them to create sample dialogs.