Custom GPTs Secrets: How to Get Great Results Every Time

Frustrated by inconsistent results from ChatGPT? Wonder how to get lightning-fast, high-quality outputs without using prompts?

In this article, you'll discover how to build and refine custom GPTs that consistently deliver reliable results—saving you time and helping you easily scale your content, marketing, or business workflows.

Why Marketers and Entrepreneurs Should Use Custom GPTs

While Saj Adib, filmmaker and founder of Howfinity, a company that helps creators and businesses learn video creation, uses approximately twenty different AI tools regularly, he uses his custom GPTs every single day.

Why does Saj use Custom GPTs instead of the core version of ChatGPT or other AI tools?

First, Custom GPTs are specialized AI tools. They let you take ChatGPT's broad capabilities and train it to perform a specific task much better than the general version could. 

Second, unlike other specialized AI tools designed to do just one thing well, custom GPTs allow you to create multiple specialized assistants for various tasks.

Third, custom GPTs significantly speed up your workflow. When using regular ChatGPT, you often need to search for the perfect prompt formula, look through prompt directories, or copy and paste templates. With custom GPTs, once set up, you can simply input what you need and get immediate results—no prompting required.

The key to success with custom GPTs is focusing on very specific use cases rather than trying to create one GPT that handles multiple functions. “That's what most people do,” Saj notes. “They're like, ‘I'm going to build this custom GPT that is my social media marketer.' Well, that's like a hundred things. It's just not gonna do a good job.”

Instead, Saj recommends creating separate GPTs for each specific task you perform frequently. This approach ensures that each GPT excels at its designated function.

5 Custom GPTs Saj Adib Uses

Saj has created numerous custom GPTs with tailored instructions for specific tasks he performs regularly. Here are just a few examples:

1. YouTube Title Generator: A specialized custom GPT that creates optimized, clickable YouTube titles based on video content.
2. YouTube Description Creator: Takes a transcript from a YouTube video and transforms it into an SEO-friendly, well-organized description.
3. Script Refresher: Reviews an old video transcript and revises it with updated information. Saj uses these refreshed transcripts to create new versions of previously published videos.
4. LinkedIn Post Creator: Turns YouTube transcripts or search-based content into engaging LinkedIn posts.
5. Marketing Navigator: Primarily used for Facebook ad copy, this GPT is trained with Saj's high-performing ads and marketing strategies to generate effective advertising copy quickly.

How to Create a Custom GPT With System Instructions: Step-by-Step

To create a Custom GPT, you'll need to use the paid version of ChatGPT. 

#1: Create a Custom GPT Draft

When you have set up your account and have identified a specific, repeatable task you perform regularly that would benefit from automation (titles, descriptions, ad copy, etc.), open Chat GPT and click on your profile picture>My GPTs>Create a GPT.

In the next window, open the Configure option.

Give your Custom GPT a Name and a description, then complete the system Instructions.

#2: Give Your Custom GPT System Instructions

Step 1: Give Your Custom GPT a Very Specific Purpose

Start by clearly defining what your GPT should do. Be as specific as possible.

For example, Saj's YouTube title generator begins with: “This GPT generates highly clickable YouTube titles for YouTube creators.”

Step 2: Define the Tone of the Language

Specify exactly what tone you want the output to have.

For the YouTube title generator, Saj uses: “Direct, compelling, and slightly clickbaity, but not misleading.”

Getting the tone right might take some trial and error. Initially, Saj tried simply “make it clickbaity,” but that resulted in sensationalized unrealistic titles. The refined instruction strikes the right balance.

Step 3: Establish Clear Rules

Create a list of specific rules that guide how the GPT should perform its task. For the YouTube title generator, Saj's rules include:

• Keep titles under 60 characters
• Use power words
• Balance human appeal with SEO optimization
• Avoid misleading clickbait
• Provide multiple variations of each title

Step 4: Include Examples of Ideal Output

Provide concise, concrete examples of what you consider successful outcomes. These examples help the GPT understand what output you're looking for but should not be confused with the knowledge base you'll provide later; keep them separate.

System instructions serve a different purpose than the knowledge base, and combining the two creates several problems:

First, the system instructions box has limited space—about 8,000 characters. If you try to include your knowledge base there, you'll quickly run out of room for the instructions that guide how the GPT behaves.

Second, system instructions need to be focused and clear. If you overload the system instructions with too much detail, the GPT gets confused about its primary purpose and doesn't know which parts to prioritize.

Third, custom GPTs typically process information in this order: first checking system instructions, then consulting the knowledge base, and finally using search if you have it enabled. Keeping these separate maintains this efficient processing order.

Fourth, knowledge base documents can be much larger than system instructions allow. Each can be up to 500MB and 2 million tokens (around 1.5 million words), which is far beyond what system instructions can handle.

So, keep your system instructions concise and focused on behavior guidance while using the separate knowledge base feature for reference materials, examples, and domain-specific information. This separation allows both components to function optimally within their intended purposes.

For his YouTube title generator, Saj includes titles that performed well in the past, such as “I Created the Perfect ChatGPT Prompt Formula” and “10 AI Hacks That Will Save You Hours a Week.”

Here's what a complete system instruction might look like for his YouTube title generator:

Use Saj Adib's Custom Instruction Generator to Create System Instructions

Saj has developed a publicly available Custom Instruction Generator GPT to streamline the process of creating custom GPT system instructions.

This GPT helps turn basic instructions into well-structured system prompts following his four-step formula.

Simply provide the context of what you need your custom GPT to do, and Saj's GPT generates system instructions that you can copy and paste.

#3: Upload the Knowledge Base

Although ChatGPT allows you to upload up to 20 documents (each up to 500MB with approximately 1.5 million words), Saj recommends keeping the knowledge files manageable. He starts with five documents and never uses more than ten, keeping each document under three pages.

For his Facebook Ad Copy custom GPT, he included five key documents:

1. Target Audience Breakdown: A full page detailing who the target audience is, including their demographics, pain points, desires, and behaviors.
2. Unique Selling Proposition: A document focusing entirely on the unique way of positioning the product or service and what sets it apart from competitors.
3. Features and Benefits: Details about the product features, but more importantly, how these translate into specific benefits for the user.
4. Best Practices: Guidelines for writing Facebook ad copy based on both general marketing knowledge and Saj's specific experience.
5. High-Performing Ads: Examples of ads that performed well in the past, including their copy, structure, and notable elements that made them successful.

Too much detailed formatting can cause the knowledge base to not work properly. Saj recommends using plain formatting in your knowledge base documents and downloading them in a simple file format like Microsoft Word rather than a PDF with complex layouts, tables, or formatting.

If you don't have the concise documents to work with, you can use ChatGPT to create knowledge base documents that contain only the information the AI needs. Saj often uses ChatGPT itself with this specific workflow:

Tell the AI what you're trying to create: 

I'm trying to make a document to train a custom GPT to [Purpose].

Ask ChatGPT to help you compile the document: 

Ask me as many questions as you need to help me write a document that could be used for this use case. Ask me one question at a time, wait until I reply, and then ask me the next question.

Have ChatGPT compile the information into a document based on your answers.

Based on my answers, create a document I can use in the knowledge base of a custom GPT.

Download the document as a simple Word file and add it to your GPT's knowledge base.

Create Knowledge Base Documents From Visuals and Screenshots

Saj recommends using ChatGPT's vision capabilities to analyze visual examples for your knowledge base. Create a PDF that features screenshots of your top-performing ads and upload it to ChatGPT. Then, ask the AI to analyze the ads. For example, you can direct the AI to find trends in hooks and give you a formula to create similar hooks in future ads. Download the analysis and use it as a knowledge base document.

Create Knowledge Base Documents from Very Large Sources: ChatGPT and NotebookLM

One of the biggest challenges when building effective custom GPTs is handling large volumes of information. While ChatGPT technically allows you to upload massive documents, extensive documentation can overwhelm the GPT with information.

When Saj encounters a large document that contains valuable information for a custom GPT's knowledge base, he uses ChatGPT to distill it down to a 3-page version with its most essential elements.

For situations requiring a more substantial knowledge base or information analysis, Saj turns to Google's NotebookLM because it offers significant advantages:

• The free version supports up to 50 different sources, and the paid version (through Google Workspace) allows up to 300 sources
• It accepts diverse source types including YouTube videos, PDFs, and Google Docs
• It maintains better connections between sources than custom GPTs can handle

#4: Optional: Activate Search for Your Custom GPT

Web browsing is particularly valuable for certain types of GPTs. For example, Saj built a custom GPT that turns search-based content into a LinkedIn post. He enabled search so the custom GPT can find the latest AI news to share in LinkedIn posts.

#5: Create, Test, Refine, and Save Your Custom GPT

After you've completed the first four steps, click on Create.

Now, it's time to test your GPT with real inputs to see if it delivers the expected outputs. Based on the results, you may need to refine your system instructions slightly (this typically takes about 30 minutes of tweaking.)

If your custom GPT doesn’t perform well, copy the system instructions and ask ChatGPT or Claude to analyze and improve them. 

Analyze and improve these system instructions for my custom GPT:

Adjust your system instructions with the suggested improvements and rerun the custom GPT.

When you're happy with the output, save your GPT.

Limitations of Custom GPTs

Despite their usefulness, custom GPTs do have some limitations compared to the full ChatGPT chat experience:

1. Limited Model Selection: At the time of this interview, custom GPTs default to GPT-4, not the newer GPT-4.5 model. You also can't use reasoning models (like O1 and O3 mini) that think through steps before responding.
2. No Deep Research: The advanced Deep Research feature isn't available in custom GPTs.
3. No Advanced Voice Mode: Voice conversations aren't supported in custom GPTs.
4. Static Knowledge Base: Unlike ChatGPT's ability to connect to Google Drive for dynamic updates, custom GPTs have a static knowledge base. You need to update any documents manually when information changes.
5. Limited Functions: Custom GPTs offer just four optional functions: Search, DALL-E for image generation, an analysis tool (which Saj doesn't recommend using), and a writing canvas.

Other Notes From This Episode

• Connect with Michael Stelzner @Stelzner on Facebook and @Mike_Stelzner on X.
• Watch this interview and other exclusive content from Social Media Examiner on YouTube.

Where to subscribe: Apple Podcasts | Spotify | YouTube Music | YouTube | Amazon Music | RSS

✋🏽 If you enjoyed this episode of the AI Explored podcast, please head over to Apple Podcasts, leave a rating, write a review, and subscribe.

---

Stay Up-to-Date: Get New Marketing Articles Delivered to You!

Don't miss out on upcoming social media marketing insights and strategies! Sign up to receive notifications when we publish new articles on Social Media Examiner. Our expertly crafted content will help you stay ahead of the curve and drive results for your business. Click the link below to sign up now and receive our annual report!