How Building a Custom GPT Saved Me 20 Hours a Week

For me, it was sometime last year when AI went from plaything to serious work assistant. Up to that point, I was mostly using ChatGPT to do silly stuff. You know, write a limerick about bagel toppings and that sort of thing. 

But then I took the plunge and started using it for anything I could. My skepticism evaporated and I became an AI evangelist. 

I now believe that we can use AI to solve almost any business problem in front of us, assuming we come to it with the right mindset.

For me, I knew I had a problem. Well, not a problem exactly, but a very time-consuming aspect of my job that was eating up bigger and bigger chunks of my calendar each week.

So, I set out to use AI to solve it. Below, I’ll walk you through exactly what I did, what I learned, and how the results turned out. 

My Problem: Content review was eating up too much time

At IMPACT, we provide marketing training to our clients. We often work with successful SMBs that want to build trust with their audiences but are suspicious of hiring an agency to speak for them.

We teach them to produce content for their ideal buyers. 

As a content trainer, I oversee a lot of this work, and I review a ton of content. So, when a company writes a new blog article or shoots a new video, they send it over to me so I can make sure it’s checking all the boxes. 

Is it following SEO best practices?

Is it formatted properly?

Are there any spelling or grammar errors?

As a former teacher, this work is right up my alley. It’s like I’m grading essays all over again.

This work is time-consuming, though. On average, it takes me about 20 minutes to review a single piece of content. And let’s be honest, 20 minutes is never really 20 minutes because the communication on either side tends to add up. 

At any given time, I’m working with 15 or more clients. We advise our clients to produce three new pieces of written content each week. You can start to see the problem.

What’s more, we promise to review each piece of content and return it within 48 hours. 

This all adds up to a bottleneck. Lots of content, high demand, and a quick turnaround, all the while wanting to maintain the high standards that IMPACT is known and valued for. 

My Solution: An AI-powered content review bot

I had played around with some APIs in Chat GPT, but I found them clunky and difficult to implement. But then OpenAI announced the ability to build custom GPTs a few months ago. My thought was this: Could I build my own GPT bot to do the initial article review for my clients? Could I generate something that was accessible through a single link? Could this possibly be the answer?

I didn’t want to take myself entirely out of the loop, but if the bot could do some of the heavy lifting, I could tame my workload and get feedback to my clients instantaneously. 

Building my custom GPT

First off, you can only create custom GPTs if you are working in ChatGPT-4, which costs $20 per month. (The free version is ChatGPT-3.5, which is less advanced and, frankly, less useful.)

From the home screen, sign in as you normally would.

openai-homepage

Then, go to the upper left to “Explore”.

explore-openai

 

From there, you’ll get into a screen that lets you configure a custom GPT.

Click the button to create a new GPT.

From there, you’ll get to a screen that allows you to configure the back-end of your new bot. It looks like this:

custom-gpt-page

The builder program will walk you through the process, asking specific questions about the purpose of what you’re looking to accomplish. 

This is pretty easy and pretty intuitive, but remember, the clearer you are, the more information you give, the better the output.

In my case, I explained exactly what I wanted the bot to do, and what the process would look like. Specifically, I said that a user would be pasting in the text from a document, and I wanted it to look for certain things in every draft.

But, as the goal of this bot is teaching, not just fixing, I instructed it to ask questions in a Socratic method instead of just making corrections. That way, the users could interact with the bot and improve their skills — rather than relying on the bot to do the work for them.

Then, in the “Knowledge” section I added a bunch of examples of feedback I’d given on drafts in the past. I included comments, corrections, and questions so the bot could mimic my tone and style when giving feedback.

At this point, my bot was ready for testing. 

I gave clear instructions and then I shared it with a few clients who already knew me and my style really well.

The Results: Streamlined feedback and more time for human connection

And with that, I released my GPT into the wild. Well, not really. Just like with a Google Doc, you can make your GPT private, public, or somewhere in between. I chose somewhere in between. I made it so anyone with the link could use it, but I was judicious about who got the link.

At first, I shared it with some colleagues and clients. And I kept stress-testing it myself. 

The people who used it gave me feedback — and the bot got smarter each time it was used.

Using a beta testing model with this GPT was crucial to its eventual success, and I would recommend this process to anyone who is building a custom GPT. By exposing the bot to various communication styles, writing samples, questions, and clarifications — the bot was able to better define what was needed for a varied audience. 

I took all of the feedback from that initial testing group, further refined the back-end instructions, and then turned the bot out to the rest of our trainers here at Impact to repeat that process.

Over the next few weeks we ran into multiple issues, which we expected:

  • The bot occasionally gave confusing or contradicting directions. 
  • It sometimes gave clients the answer instead of helping them learn it themselves.
  • Once or twice it even bugged out and wouldn’t work at all. 

But we kept refining the instructions and improving the performance. 

Remember, AI doesn’t replace humans, it enhances them

When I start working with a new client, I want to build a relationship based on trust. I need my clients to know that I’m there to guide them. Therefore, I wouldn’t throw my GPT at them right after kickoff. 

In those early meetings, the human touch is vital.

But as we settle into a cadence, the GPT can be another ‘teacher’ they interact with. The bot provides instant feedback, and they can go back and forth with questions. If the bot says the article’s introduction needs to pose a stronger question, the writer can try a few out and get feedback in real time. 

Then, they can come to me with higher-level questions or if they need further guidance.

So far, reviews have been positive. Collectively, our entire species is figuring out how to interact with AI, and I think people are willing to jump in, try something out, and be open-minded. 

What used to take me 20-25 hours per week, has now gone down closer to 5 or 10 hours. Some weeks I’ve skated by with even less — though not all the time. It’s still very important to me that our clients have that human interaction. 

Now it’s your turn

In a recent article, my colleague Chris Duprey shared four principles he uses to guide his AI exploration. He adapted these from Wharton Professor Ethan Mollick:

  1. Use AI for everything you can. Tools are proliferating. New ones come out every day. We should always be on the lookout for novel ways to utilize AI’s power.
  2. Be the human in the loop. AI has astonishing abilities — and sobering limitations. It is not a human with judgment and character. You need to keep humanity in whatever you’re doing.
  3. Treat the AI like a human and tell it who it is. AI can be whoever you want it to be, but only if you’re clear. The better the input, the better the output.
  4. Remember, the AI you’re using today will be the most rudimentary AI you ever use. Things are advancing so quickly that what’s new today will be outdated in a few months. New capabilities are always emerging

I kept these in mind when I was experimenting, and I advise you to do the same. What current challenge can you use AI to solve? What new tools can you explore?

If you’re a business owner, how can you build a healthy AI culture at your organization that rewards exploration and experimentation?  

If you’re concerned that you’re sitting on the sidelines of the AI revolution, talk to the team at IMPACT. We sift through the hype and the fearmongering to empower teams to come together and use technology in a productive and ethical way.

To learn more, check out our AI Mastery page.

By