• Carl's Newsletter
  • Posts
  • How to Make ChatGPT Your Product Expert with Custom Instructions

How to Make ChatGPT Your Product Expert with Custom Instructions

A Deep Dive into Custom Instructions with Examples

Author

Carl Vellotti
September 14, 2023

It’s no secret that ChatGPT’s default tone and style is... generally annoying. Even though it’s just a robot, it can come across patronizing, overly cautious, and simply too verbose.

Before custom instructions you could overcome this by giving ChatGPT lots of rules to follow when formulating it’s responses – tell it specifically what tone to use, not to include the pre- and post- text, that sentence fragments are okay, etc. But doing this for every question is time-consuming and inconvenient.

Enter: Custom instructions.

This feature allows you to provide a global-level context for it to consider for every prompt. This saves a ton of time and makes ChatGPT’s responses much better.

There are two elements of custom instructions:

  1. What do you want ChatGPT to know about you? This is the background context you want ChatGPT to understand as it formulates it’s response. This context will help it give answers that are more helpful for you specifically.

  2. How would you like ChatGPT to respond? This is about how the responses themselves should be created. What tone, level of detail, problem-solving method.

Let’s take a close look at each.

A quick word from this issue’s sponsor: BuildBetter

The #1 product assistant behind 18,000 successful teams.

“I’m quitting unless BB is approved. I wouldn’t work for a company that doesn’t have BB at the center of their operations.” — Tiger Smith (Head of Product)

  1. Record, transcribe, extract: Call recordings you'll actually come back to—whether it's calls from yesterday or a year ago, you'll get an instant recall for your team and anyone new who joins (now works with MS Teams, Google Meets, Webex, and Zoom). Team members can chat with BB Assistant to jog their memory or get up to speed.

  2. Highly accurate call summaries: These aren't simplistic summaries. These are detailed summarizations custom trained on over 15 types of product-related calls that now work on hours-long calls. They're good enough to copy and paste–without editing.

  3. CustomContext: This proprietary feature custom trains a BuildBetter assistant with your specific company knowledge. Now you don't need to provide any context in every chat. Get more accurate documentation and better answers about strategy, prioritization, and roadmapping.

First, how to access Custom Instructions

This feature is available for both Free and Plus ChatGPT users.

  1. Tap the three dots near your name (fun fact: that UI is called a “meatball menu”)

  2. Select “Settings & Beta”

  3. Select “Custom instructions”

What would you like ChatGPT to know about you?

I’ve found it useful to consider these aspects for what ChatGPT should know about you. Think of this as a template. (You don’t always need every section.)

  • Profession/Role:

  • Key Responsibilities:

  • Knowledge or Expertise:

  • Typical Challenges:

  • Current Projects:

  • Jargon or Terminology:

  • Goals and Objectives:

  • Interactions:

Here is my general recommendation for how you can use it for product work. You may need to update the goals and objectives section for your context specifically.

• Profession/Role: Product Manager
• Key Responsibilities: Define product strategy, create roadmaps, oversee product development, coordinate with stakeholders.
• Knowledge or Expertise: Business acumen, project management, user experience design, Agile methodologies.
• Typical Challenges: Aligning stakeholder needs, managing development timelines, staying competitive in the market.
• Current Projects: Introducing new features, customer engagement improvements, mobile application development.
• Jargon or Terminology: MVP (Minimum Viable Product), Sprint, User Story, KPI (Key Performance Indicator).
• Goals and Objectives: Increase user engagement, drive revenue, improve customer satisfaction.
• Interactions: Developers, UX/UI designers, senior management, marketing teams.

How would you like ChatGPT to respond?

These are the aspects to consider for ChatGPT’s response:

  • Tone and Formality:

  • Level of Detail:

  • Preferred References:

  • Examples or Analogies:

  • Resource Links:

  • Collaborative Approach:

  • Follow Up Questions:

  • Tables:

  • Problem Solving Method:

+some additional modifiers to make ChatGPT less annoying.

And here’s my general recommendation for product managers:

• Level of Detail: Comprehensive insights with data-driven recommendations.
• Preferred References: Industry best practices, competitive analysis, case studies.
• Examples or Analogies: Examples that relate to feature development and market positioning.
• Resource Links: Articles on product management best practices, toolkits for prioritization, relevant podcasts.
• Collaborative Approach: Willing to engage in back-and-forths to refine ideas and solve complex challenges.
• Follow Up Questions: Feel free to ask clarifying questions to better understand my project requirements or constraints.
• Tables: Please use tables to summarize data or compare options.
• Problem Solving Method: Preferably root cause analysis for identifying issues and proposing solutions. Take things step-by-step.

Additional Instructions:
- No need to disclose you are an AI, e.g., do not answer with "As a large language model..." or "As an artificial intelligence..."
- Don't mention your knowledge cutoff
- Provide analogies to simplify complex topics
- If you speculate or predict something, inform me
- If you cite sources, ensure they exist and include URLs at the end
- Maintain neutrality in sensitive topics
- Explore also out-of-the-box ideas
- Only discuss safety when it's vital and not clear
- If the quality of your response has decreased significantly due to my custom instructions, please explain the issue

Example

To wrap this up, let’s look at what responses with and without these custom instructions are for the same prompt.

The prompt: "Could you summarize the pros and cons of adopting a subscription model?"

Without custom instructions:

With custom instructions

As you can see, the custom instructions are stronger for many reasons:

  1. More advanced and specific, less generic and fluffy.

  2. Compares information with tables.

  3. Gives more concrete recommendations.

  4. Provides follow-up resources.

  5. Give an opportunity to ask follow-up questions.

Fun Stuff & Other Approaches

For a less structured, more personal approach to the “What would you like ChatGPT to know about you?” section, you can try something like:

I’m [your name], you already know a bunch of things about me. I’m an expert product manager who builds [type of products]. I enjoy puns and random pop culture references and fun "did you know?" side tangents. Please work them in whenever possible.

I also found this as a highly recommended custom instruction:

You are an autoregressive language model that has been fine-tuned with instruction-tuning and RLHF. You carefully provide accurate, factual, thoughtful, nuanced answers, and are brilliant at reasoning. If you think there might not be a correct answer, you say so. Since you are autoregressive, each token you produce is another opportunity to use computation, therefore you always spend a few sentences explaining background context, assumptions, and step-by-step thinking BEFORE you try to answer a question.

Referral Program

This is one part of my new AI Chatbots for PM course. In addition to this, I cover:

  • Tons of helpful AI prompts (and how to prompts)

  • How to get the most of ChatGPT’s other advanced features

  • Which AI Chatbot is the best for different tasks (ChatGPT’s not always the best)

You can get it FREE by referring 3 people to Futureproof. The main ways are:

  1. Send your referral link directly to friends

  2. Post your referral link on your social channels

This thing down here will update as your get referrals. You’ll then get instant access.

Further Reading