Personalized Picks in Minutes – Hotel AI, Marketing, Tech and Loyalty

But I wanted to take things a step further this time. I wanted something I could use when I’m anywhere—on the couch, on the move, or even up in the air, deciding what to watch next. That’s when I decided to build a Custom GPT—my very own AI Movie Muse. This GPT isn’t just any recommendation tool; it’s fine-tuned to know exactly what I like based on my IMDb ratings, my mood, and even my quirks as a movie lover.

Here’s how I built it (and how you can, too!), plus what worked well and where I hit some limitations along the way.

Step 1: Getting Started – Why A Custom GPT?

You might be wondering, why bother creating a Custom GPT when you can just use standard tools? Well, I wanted something that wasn’t generic. I wanted an AI that could give me personalized, spot-on movie recommendations no matter where I am. That’s where a Custom GPT comes in.

Creating Custom GPTs with ChatGPT is pretty easy, and the platform walks you through the process. Whether you want your AI to recommend movies, help you pick out music, or suggest books, the setup is simple and super accessible. The great thing is you can also make these GPTs available to the public…in this case since I wanted to personalize it, I decided to keep it private.

Step 2: Naming Your GPT and Defining Its Purpose

This is where the fun begins! I wanted my GPT to feel personal, like a friend who knows my taste inside out. I named mine “JJ’s Movie Muse”, and gave it a simple mission: recommend movies and TV shows based on my unique preferences, moods, and personal IMDb ratings.

Naming your GPT and giving it a clear purpose is important because it sets the stage for how it will interact with you. Do you want it to be casual and fun, or more formal and professional? I went for something in between—an AI that’s both insightful and easy to chat with.

Step 3: Fine-Tuning with Custom Instructions

Now, here’s where I dialed in the fine-tuning. I wanted JJ’s Movie Muse to act like a movie critic who knows my tastes intimately, while also keeping things quick and snappy. To get the right balance, I gave the GPT custom instructions that included:

• Using short, bullet-pointed responses (because no one likes walls of text).
• Being casual yet professional, blending expert movie knowledge with a personal touch.
• Always providing reasons for recommendations based on my past ratings and preferences. For instance, it knows I love sci-fi, dark humor, and thought-provoking plots, but that I’m not really into overly sentimental dramas.

This part was essential for performance. The better you fine-tune the instructions, the more useful the GPT becomes. I made sure to cover different input types—whether I’m searching by mood, specific movie title, or even images of movie posters.

For Beginners: If you’re not super comfortable providing the detailed instructions, you can also leverage the “Create” mode and have ChatGPT do the work for you based on what you tell it.

Pro Tip: If you’d like fine-tuned output and behaviour from the custom GPT, use the “Configure” tab and provide all the details. To really dial up the quality of the input, I used ChatGPT (in a different chat session) to create the custom instructions after having it analyze my personal IMDB ratings and getting a better understanding of my preferences. I also asked it to craft instructions based on different scenarios, e.g. what I want when I ask about a specific title vs what to do when given a picture of various movies, etc.

Step 4: Uploading My Ratings for Super-Personalized Results

Here’s where things got really interesting: I uploaded my personal IMDb ratings—a file with over 700 movies I’ve rated over the years (see screenshot below…you have an “Export” option available). This allowed the AI to reference actual data about my movie preferences and predict what I’d like (or hate) with much better accuracy.

Pro tip: Uploading your own data adds a layer of personalization that can’t be matched by off-the-shelf recommendation engines. Whether it’s movie ratings, book lists, or something else, giving your AI specific data to work from makes it your AI, not just another generic assistant.

Step 5: Testing It Out and Customizing for Real-World Use

With everything set up, I started testing the GPT with real-world scenarios. Here’s a rundown of the three input methods I fine-tuned:

1. Movie Title Search: When I enter a movie title, JJ’s Movie Muse does a web search to pull IMDb ratings, reviews, and even user opinions. It then compares the movie’s vibe to my taste profile and lets me know—snappily—whether it’s worth my time.
2. Mood-Based Recommendations: If I’m feeling a certain way (e.g., contemplative, adventurous, or in need of a good laugh), I just tell the AI my mood. It searches for trending films that match that feeling and my tastes, offering me three recommendations with reasons why I’ll enjoy them.
3. Poster/Image Analysis: I can even upload an image of movie posters, and JJ’s Movie Muse ranks them based on how much I’ll like each one. It’s a fun feature to use when I’m deciding what to watch next on a long flight.

Here’s a quick demo of how it works…

What Worked Really Well:

• Accuracy: By uploading my IMDb ratings and having ChatGPT analyze my preferences, the GPT became pretty accurate in predicting my preferences. I’ve tried other recommendation engines, but this one nails it because it’s built just for me.
• Flexibility: Whether I’m searching by title, mood, or even posters, the AI adapts. It’s like having a personal movie critic I can take anywhere. Using this on the ChatGPT mobile app while on the move makes it particularly convenient.
• Fun Factor: Custom GPTs make AI feel less like a tool and more like a personal assistant. There’s something satisfying about getting recommendations that feel like they’re coming from someone who really knows you.

Where I Hit Limitations:

• Data Limitations: The GPT can only go as far as the data it’s fed. While uploading my IMDb ratings worked great, the GPT sometimes struggles with less-known movies or niche genres that it doesn’t have as much data on.
• Mood-Based Recommendations: While the GPT does a great job matching mood and preferences, it’s limited by what’s trending. Sometimes, I wish it could dig deeper into lesser-known titles that fit my niche tastes.
• Hallucinations and Laziness: I’ve noticed that ChatGPT doesn’t do a particularly good job of fetching IMDB and Rotten Tomatoes scores reliably from web searches…sometimes it just makes them up or says they’re not available, even when they are.
• Not great at excluding already watched titles: Perhaps this is due to limitations on how ChatGPT uses uploaded data, but despite giving it repeated instructions about NOT recommending titles already in the personal ratings file, it fails to follow through on this ask.

That said, these are relatively minor hiccups compared to the overall experience and I hope the experience gets better as the tools and models improve. The flexibility and personalization offered is worth it.

Conclusion: Why You Should Build Your Own Movie Muse

Building my own AI Movie Muse has been a fun and rewarding follow-up to my previous experiment last year, where I built a personal movie database. This time, it was driven by convenience and my desire for deeper personalization.

Clearly, there are things that work really well—like the personalization, convenience and flexibility—but there are also areas that could be improved, especially around retrieval of information from the web and uploaded files. Still, the ease of creating a custom GPT makes this a worthwhile effort for anyone looking for more tailored experiences, whether it’s movie recs, book suggestions, or even help with everyday tasks.

It’s simple, intuitive, and genuinely useful. If you’ve been thinking about making your own Custom GPT, I’d highly recommend giving it a try. You’ll be surprised by how much it can improve your day-to-day—and how much fun you’ll have along the way.