About a week ago, I was working on a creative project for my job. This task was self-created and was open ended, with no requirements from anyone else.

The purpose was to create content (in the case an internal blog post) to share with my team (~150 people) to a) share key details of an internal mentorship program that our company has available to all employees, and to b) provide a convincing reason why the employee would want to sign up.

Our company had already created extensive resources and content around this specific topic, so I wanted to leverage this information (as context) to create an approachable and concise blog post on the topic, with the help of AI.

I began by writing a detailed prompt, outlining my objectives, desired outcome, and creating specific parameters for the LLM to follow while executing the task.

“I work for XXX. We have a mentorship program where employees can request to have a mentor in our company to help them grow in their career. A long bit of context is shared below. I want to create a blog post that will be shared on our internal blog with employees explaining the program in simple terms, outlining the benefits specifically showcasing the value and explaining the “what’s in it for me” details – encouraging them to apply. There is a mentor element of the program too, but this post will be 100% focused on becoming a mentee.

Please write a blog post for me including the following:

Key components of the program:

• Potential value for the team member & why they should consider signing up
• How they sign up (link to our internal resources which share a lot of the context listed below) and the formal link to signup
• Any other elements that you see as important that I am missing

Make this post easy to read and understand – no more than an 8th grade reading level. Make it feel light and comfortable, avoiding a bunch of corporate speak – with the exception of potentially including [company] cultural ideals, etc. Make it engaging. Use no more than 2500 characters, but less is great if you the message can be articulated clearly. Break it up into digestible chunks in whatever manner you see fit.

End it with a strong call to action.

Also create a strong headline that captures attention and that employees will want to click on.”

Below this, I aggregated the details of the program, including FAQs, and was ready to submit my prompt.

For this task, I wanted to test the output of two different LLM models, ChatGPT (free version, GPT-4-turbo) and Claude (paid, Sonnet 3.7), and compare the output of both.

I ran the prompt on both models and reviewed the results.

As I compared the two, I vibed more with the ChatGPT output. It felt more approachable and almost like I was talking to a friend when compared with the Claude output.

I will note, the ChatGPT model incorporated emojis, and I like using emojis when I text (likely adding bias.) But I was not going to use emojis in the final blog post, so I needed to ignore them.

At that point, I copied the output of both models, and printed them out. At that point, I was out of time for the day, and I wanted to take a bit more time to consider both of the model’s approaches of articulating the overall message and how each addressed my overall objectives. And I was curious how others might respond to the overall flow of each’s content.

Over the next few days, I showed both of the unedited drafts to two different people and asked which one resonated with them the most. Both indicated they favored the output from ChatGPT.

While I cannot say for sure that the emojis didn’t influence their decisions — I had not removed them from the GPT draft — both suggested that the GPT output felt more like they were talking to a friend (i.e. more approachable.)

I agreed.

Using the ChatGPT output as a starting point, I began to edit the document, making changes where the details didn’t line up with what I wanted, removing fluff or language that didn’t feel like it reflected my own voice, and adding additional context where I felt there were gaps and clarity was needed.

From there, I put the blog post in a draft format, and asked one of my leaders to review it. As they read it, I sensed they were a bit surprised by what they were reading:

“Is this a company template that someone else wrote?” they asked.

“No.” I responded.

“Did you write this?”

“Yes.”

They said it was good and gave me a few additional ideas on how to leverage it.

And that is where my mind started racing…

Did I write this?

I certainly created it, but did I actually write this?

My stomach began to get queasy.

I had not elaborated to say that I had used AI to help me create the blog post. And it made me feel terrible.

Was I being dishonest?

Was my integrity in question?

Where was the line here? In business or any capacity really, what was acceptable and what was not acceptable?

Did I feel comfortable putting my name on this work and calling it my own?

I wrestled with these questions all night going back and forth in my mind. But it still left me with a feeling of unease.

I decided that though I had already taken the AI draft and made edits, molding it into a message that I felt good about, I would go back and figure out how I could make it more my own before sharing it broadly.

The next day, I spent a bit of time really digging into the language of the writing and the flow of the message, making additional edits and rewriting certain parts to ensure it was definitively in my own voice.

In the end, I felt really good about the content. It flowed well, was in my own voice, and I felt it accomplished the objectives I originally set out to do so.

But there was still a little unease in my mind that nagged at me: where is the line on AI use in the workplace?

So, as any regular LLM user would probably do, I went to ChatGPT for council:

You did a great job rewriting this! It’s clear, natural, engaging, and stays true to the spirit of the XXX program. You’ve captured the important pieces while making it feel more personal and motivating — exactly the tone you said you were aiming for. [removed program name for org privacy]

The AI asked me if I wanted some additional suggestions to make it pop. I did not, but I wanted to discuss exactly how this process made me feel:

ChatGPT responded:

Reading this, made me pause, “Okay, 70-80% my voice and my editing.” And the idea of using a brainstorming partner resonated.

ChatGPT continued…

I have used Grammarly in the past, so again this resonated. I certainly did not feel guilt about using that tool to improve my writing.

And in this case, I used ChatGPT as a tool to help me in my process. And yes, the end product absolutely reflected my standards, values, and understanding. So this made me feel a bit better.

ChatGPT continued…

As I read this, I took comfort in the fact that my process seemed to line up with the ideas presented here.

I did use the AI to save time compiling large amounts of data.

I did use the AI to create the first draft.

I did use critical thinking when I began by deciding what my objectives using ChatGPT were, and how I wanted to go about the process (scope, parameters, etc.)

I completely understood the information I was presenting.

I did rework the draft to make it line up with my thinking, voice, etc.

ChatGPT continued…

Alright, I was feeling much better.

And I also was proud of myself for taking the time to really consider the subject matter and how I wanted to proceed using AI in my work moving forward.

I wanted to go a bit deeper and further explore how I should think about leveraging AI in my daily work and living though. And I wanted to explore how to navigate the complexity of this topic in society (and ultimately business) and how to leverage in a manner that would not tarnish my professional reputation or integrity.

Reading the checklist, and considering the blog post I had created, my answer was yes to all three questions. I felt really good and submitted it to go live to the team.

In addition to the coaching on how to evaluate my work and using AI, I felt really good about the framing and strategies ChatGPT had provided me here. And I felt I now had a good frame of reference to consider as I move forward and explore how to effective leverage AI in my work.