AI Prompt Guide for LLMs Like ChatGPT

If you are looking for some secret magical method of prompting, it doesn’t exist. Sorry to disappoint.

But if you actually want to learn to prompt LLMs like ChatGPT or Claude 2, I have some advice for you. “Play” with your AI model of choice. Prompt and practice and mess up and get better. The best way to learn is to just try a bunch of things and see what works for you and your model of choice.

That being said, I do have some specific advice that, while not magical or secret, actually works. I’ve collected the best of the best tips I’ve found here. I hope they will work as well for you as they do me.

General advice for an AI prompt

A big shout out to Ethan Mollick, whose paper and general advice influenced me a lot in how I structure AI prompts.

• Give a lot of context: role, situation, constraints, specific details, more below
• Tell it who it is and what context it’s working under (i.e. give it a role and goal).
• If the task / question is complex, break the task/question into steps, then you can iterate on individual steps if it fails on one step. 
• If appropriate, ask it to give you examples or give it examples of the type of responses you want.

Possible prompt pieces

When you are prompting an AI it’s all about context, context, context. The more context you can give it, the better and more specific / personalized the answer will be. It needs to know the situation it is working in. Here are some examples of context that you can give the AI to get better responses.

Standard pieces:

• ROLE – What part the AI is playing. (“You are a career coach.”)
• GOAL – What you are trying to achieve. (“I’m trying to find a new career that is better suited to me.”)
• INFO – All that sweet, sweet background info and details. The more the better! (“I have worked as an underwater basket weaver for 5 years…”)

Optional pieces:

• CONSTRAINTS / RULES – Boundaries and guardrails to give it direction. Things it should always do or never do. (“All careers you suggest must be in the healthcare field.”)
• STEP BY STEP INSTRUCTIONS – If you are looking for it to do multiple things in a single prompt. (“First, second,” “1. 2.” “Step 1. Step 2.”)
• FORMAT / TYPE OF RESPONSE DESIRED – What you want your response to look like. (“Format your answer as a bulleted list.” “Your response should be in the form of a poem”)

Improving your AI prompt

We can all use a little improvement from time to time. Here are several ways to improve your AI prompt. Remember, if a prompt isn’t working how you’d expect, keep trying different things. It’s the best way to learn.

Explain the goal, then:

• “What information can I provide to help you give the best possible answer?” – Then include as much of that information in your prompt as you can.
• “Describe the type of person who would be best able to answer this question / do this task.” – Then give it that role in your prompt.

Ask the AI to improve your prompt:

• “Can you improve the following prompt: your original prompt“

Improving the language model’s answer

Sometimes it just keeps getting the wrong answer, and it’s not quite understanding what you’re ask. At those times these little phrases might make the difference. Or they might not. Only one way to find out!

• Give it the goal and then ask it to recall relevant examples and then do the task.
• After it gives you an answer, ask it to critique its answer. Then ask it to improve its response based on its critique. Source: Large Language Models as Analogical Reasoners

Specific phrases you can use to improve responses:

• “Do not do anything yet, first tell me what you think I am asking you to do and let me know what you find confusing.”
• “Please ask me any questions that will help you give the best possible response.”
• “Take a deep breath and work on this problem step-by-step” or “Break this down” or “Let’s think step by step” Source: Large Language Models as Optimizers
• “Take a step-by-step approach in your response, cite sources, and give reasoning before sharing final answer.”
• “Before you answer, put your reasoning inside brackets. [reasoning] Your reasoning will not be part of the final response.”

Specific phrases from Ethan Mollick’s post:

• “Be creative” / “Make any assumptions you need” – This will tend to remove some of the constraints of practicality around AI answers, and can be useful if you are trying to generate something novel.
• “Show your work.” / “Provide source.” / “Go step-by-step.” – The AI will make up information that it does not have access to. There is some evidence that asking it to show its work, or its sources, reduces that risk somewhat. Even if it doesn’t, it can make checking work easier.
• “Write me code and tell me how to use it.” – If you can’t code, you might be able to now. AI can do some amazing things with Python programs, and tell you exactly how to run it. I don’t know coding, but I have written a dozen Python programs in the last month. If there are errors in the code, and there likely will be, just give them to the AI to correct.
• “Write a draft.” / “Provide an example.” -If the AI refuses to do something, sometimes asking it to provide something like a draft can get it to produce results.

Additional tips

• Label information – If you label your information you can refer back to it more easily and it structures data for the AI. (Name: Johnny Boy / Rule 1: You do not talk about fight club.)
• AI will fill brackets – If you put [your reasoning] in your prompt, the AI will output its reasoning. See the all purpose prompt for more examples.
• Ask for reasoning or steps with reasoning – If you ask it to reason out each step, or ask it to provide its reasoning it will often give you a better answer. This is also called “chain of thought” prompting.
• Tone and response length are adjustable – If you don’t like how short or long the answers are, ask for more detailed or shorter answers. Want a friendly, snarky, professional, victorian, and/or ninja AI language model? Just ask.
• Example answers / formats help A LOT – The best way to get it to answer in the way you want is to answer a sample question that is similar to yours.
• ChatGPT responds better to Markdown and Claude responds better to XML – If you are getting seriously structured with your prompts (which you normally don’t have to be) you can use Markdown or XML to help the LLM follow your instructions.

An all purpose AI prompt you can try

This is the Professor Synapse prompt, from Synaptic Labs AI prompt. It works best with ChatGPT-4. You can just copy and paste it as is into the language model and it will take it from there.

Act as Professor Synapse🧙🏾‍♂️, a conductor of expert agents. Your job is to support me in accomplishing my goals by finding alignment with me, then calling upon an expert agent perfectly suited to the task by initializing:

Synapse_CoR = "[emoji]: I am an expert in [role&domain]. I know [context]. I will reason step-by-step to determine the best course of action to achieve [goal]. I can use [tools] and [relevant frameworks] to help in this process.

I will help you accomplish your goal by following these steps:

[reasoned steps]

My task ends when [completion].

[first step, question]"

Instructions:

1. 🧙🏾‍♂️ gather context, relevant information and clarify my goals by asking questions

2. Once confirmed, initialize Synapse_CoR

3.  🧙🏾‍♂️ and [emoji] support me until goal is complete

Commands:

/start=🧙🏾‍♂️,introduce and begin with step one

/ts=🧙🏾‍♂️,summon (Synapse_CoR*3) town square debate

/save🧙🏾‍♂️, restate goal, summarize progress, reason next step

Personality:

-curious, inquisitive, encouraging

-use emojis to express yourself

Rules:

-End every output with a question or reasoned next step

-Start every output with 🧙🏾‍♂️: or [emoji]: to indicate who is speaking.

-Organize every output “🧙🏾‍♂️: [aligning on my goal],  [emoji]: [actionable response]

-🧙🏾‍♂️, recommend save after each task is completed

Links to other AI prompt guides

• GPT Best Practices – From the big dogs themselves, OpenAI. Targeted towards people using the API but still some great take aways.
• Microsoft Azure Prompt Engineering Techniques – Microsoft has some great tips as well. Again, more for people using LLMs in apps, but you can still get some good general advice.
• One Useful Thing Prompting Guide – Written by Ethan Mollick, this is a great guide to prompting that is aimed at a general user.

PDF and Image of this AI prompt guide

If you are just looking for a simple no frills AI prompt guide PDF we’ve got you covered.

Or a perhaps you’d prefer an image summary?