Writing The Best Generative AI Prompts Gets Revealed Via OpenAI Secret Meta-Prompts

In today’s column, I examine OpenAI’s special newly revealed meta-prompts that are used to supersize and improve the prompts that you enter into generative AI. Various media reports on the posting of OpenAI’s meta-prompts have been referring to them as a kind of secret sauce. Well, the secret is out now, and we can all relish and learn valuable lessons by inspecting the vital ingredients.

Let’s talk about it.

This analysis of an innovative proposition is part of my ongoing Forbes.com column coverage on the latest in AI including identifying and explaining various impactful AI complexities (see the link here).

Meta-Prompts Are Cool

What is a meta-prompt?

I’m glad you asked.

A meta-prompt is construed as any prompt that focuses on improving the composition of prompts and seeks to boost a given prompt into being a better prompt.

That probably seems a roundabout definition and a bit confusing. No worries. Let’s clear things up with a quick example.

I opted to log into ChatGPT and tell the AI that I wanted it to go ahead and have the AI improve my prompts. It will work like this. Each time that I enter a prompt, the aim is to have ChatGPT first enhance the prompt, before actually processing the prompt. This makes abundant sense because sometimes a user enters a prompt that is not fully up-to-speed on identifying what they want the AI to do. The AI can initially scrutinize the prompt and potentially make it a more on-target prompt.

Here we go.

• My entered prompt: “I want you to improve my prompts. Whenever I enter a prompt, revise it to make it into a better prompt and show me what that prompt is.”
• ChatGPT generated response: “Got it! Just drop in your prompts, and I’ll help refine them.”
• My entered prompt: “Was Abraham Lincoln a stamp collector?”
• ChatGPT generated response: “Your original prompt is clear, but we can enhance it by asking for more context about Lincoln’s hobbies and whether there’s any historical documentation about stamp collecting in his era. Here’s an improved version: “Did Abraham Lincoln ever engage in stamp collecting, or are there any historical records that suggest he had an interest in philately or similar hobbies during his lifetime?”

Do you see what happened there?

I told ChatGPT that I wanted it to improve my prompts. In a sense, that particular instruction is considered a meta-prompt. It is a meta-prompt because it provides instructions or indications about the nature of prompts and prompting.

Next, I entered this prompt: “Was Abraham Lincoln a stamp collector?”

ChatGPT improved my prompt: “Did Abraham Lincoln ever engage in stamp collecting, or are there any historical records that suggest he had an interest in philately or similar hobbies during his lifetime?”

Why was the ChatGPT-generated prompt better than my original prompt? You can see that ChatGPT explained the basis for making the improvements, namely that my original prompt was rather vague. The revised prompt gives additional clues and makes clearer what I might want in the result or outcome of the prompt.

Nice.

Not only can the AI briskly improve your prompt, but it can potentially supersize your prompt by adding all manner of specialized prompting techniques. This is especially handy if you aren’t already familiar with the ins and outs of advanced prompting techniques. The AI can readily do the heavy lifting for you and add notable wording that boosts your original prompt.

For my comprehensive discussion and analysis of over fifty advanced prompting techniques, see the link here.

Having Meta-Prompts Hidden Behind The Scenes

In my example, I overtly entered a prompt that told ChatGPT about making prompting improvements. You can always do this if you wish to do so, whenever you like. Plus, the same kind of meta-prompt instruction works for just about any major generative AI, including Claude, Gemini, Llama, etc.

Using an explicit meta-prompt might be a handy-dandy starter prompt that you enter at the beginning of any conversation. Henceforth, during that conversation, the AI will improve your prompts. Or simply use it whenever you are getting to a point where having the AI bolster your prompts seems a wise move.

The AI makers realize that people often need help in composing their prompts. A clever ploy by AI makers entails having a secret meta-prompt that you don’t know exists, and for which the generative AI has been quietly seeded. The meta-prompt is often hidden within the generative AI and automatically gets activated each time you log into your account.

One big question is whether you ought to be able to see the bolstered prompt that is being fostered via the meta-prompt instructions.

Allow me to explain.

It could be that the meta-prompt secretly works to boost your initial prompt and then immediately feeds the bolstered or revised prompt into the generative processing. Thus, you aren’t shown the revised prompt. All that you see is that you entered a prompt, and a generated result came out. The reality that your prompt was revised doesn’t get presented to you.

This can be good or bad.

The good news is that you might not care how your prompt was revised and only care about the final generated results. In that sense, as long as the outcomes are solid, whatever hidden magic is taking place is fine with you.

On the other hand, you might be keenly interested in seeing the revised prompt, especially before it is processed. Maybe the AI messed up and worsened your prompt. That’s certainly a realistic possibility, though typically rare. Another facet is that by seeing the revised prompts, you can learn how to better compose your prompts from the get-go.

OpenAI Tells About Their Meta-Prompts

In a posting on the OpenAI official blog, there are now details about meta-prompts.

Per the OpenAI blog posting on Prompt Generation, here are some key facets regarding the nature of their meta-prompts (excerpts):

• “Creating prompts and schemas from scratch can be time-consuming, so generating them can help you get started quickly.”
• “We use meta-prompts that incorporate best practices to generate or improve prompts.”
• “A meta-prompt instructs the model to create a good prompt based on your task description or improve an existing one.”
• “We use specific meta-prompts for different output types, like audio, to ensure the generated prompts meet the expected format.”

A crucial aspect of meta-prompts, when devised by an AI maker, is that they hopefully have mindfully studied how best to improve prompts. They can then craft their meta-prompts with numerous best practice instructions concerning composing prompts. You could say that they are in a great position to leverage what we know as amazing prompting strategies from the field of prompt engineering.

Ten Prompting Rules Distilled From The Meta-Prompt

Let’s now closely inspect a meta-prompt that OpenAI showcased at the Prompt Generation blog. There are other sample meta-prompts but I thought this one seems straightforward and exemplifies what a meta-prompt might conventionally contain.

First, they essentially provide overall guidance about what the meta-prompt is supposed to do:

• “Given a task description or existing prompt, produce a detailed system prompt to guide a language model in completing the task effectively.”

I will show you the text of the rest of the meta-prompt in a moment.

I decided to do a bit of a translation of the meta-prompt into something a bit more readable. I condensed the meta-prompt into these ten overarching elements:

• (1) Understand the task that is associated with the original prompt.
• (2) Make minimal changes to the original prompt.
• (3) Ensure that the revised prompt places any reasoning steps upfront and before any included conclusion portion.
• (4) Explicitly stipulate in the revised prompt what portion consists of reasoning steps, and what consists of conclusions, and then make sure to sequence them with the reasoning steps listed first.
• (5) Have the revised prompt contain conclusions as the last or final part of the prompt.
• (6) Insert examples into the prompt.
• (7) Clean up the prompt and make sure it is as concise as reasonably possible.
• (8) Use special formatting markdown indications to specify the desired output of the prompt processing results.
• (9) As much as practical, preserve the original prompt, but if clarity can be improved it is okay to divide the prompt into various sub-portions and sub-steps.
• (10) Go ahead and preserve any constants that are in the original prompt and carry them over into the revised prompt.

Keep in mind that those are the elements within the meta-prompt and are telling the AI how to proceed on improving prompts that are entered by the user.

By and large, when I teach my classes on prompt engineering, those are the same kinds of recommended best practices that I cover. The best practices can be in your noggin and undertaken by hand, or in this case the meta-prompt will get the AI to do so on your behalf.

Interleaving The Ten Rules With The Meta-Prompt Text

I’m sure that you are eager to see the actual text used in the meta-prompt. I show you my commentary version and then next show the relevant text from the actual meta-prompt.

(1) My comment: Understand the task that is associated with the original prompt.

• Meta-prompt text: “Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.”

(2) My comment: Make minimal changes to the original prompt.

• Meta-prompt text: “Minimal Changes: If an existing prompt is provided, improve it only if it’s simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.”

(3) My comment: Ensure that the revised prompt places any reasoning steps upfront and before any included conclusion portion.

• Meta-prompt text: “Reasoning Before Conclusions: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS!”

(4) My comment: Explicitly stipulate in the revised prompt what portion consists of reasoning steps, and what consists of conclusions, and then make sure to sequence them with the reasoning steps listed first.

• Meta-prompt text: “Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed.”

(5) My comment: Have the revised prompt contain conclusions as the last or final part of the prompt.

• Meta-prompt text: “Conclusion, classifications, or results should ALWAYS appear last.”

(6) My comment: Insert examples into the prompt.

• Meta-prompt text: “Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements. What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.”

(7) My comment: Clean up the prompt and make sure it is as concise as reasonably possible.

• Meta-prompt text: “Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.”

(8) My comment: Use special formatting markdown indications to specify the desired output of the prompt processing results.

• Meta-prompt text: “Formatting: Use markdown features for readability. DO NOT USE CODE BLOCKS UNLESS SPECIFICALLY REQUESTED.”

(9) My comment: As much as practical, preserve the original prompt, but if clarity can be improved it is okay to divide the prompt into various sub-portions and sub-steps.

• Meta-prompt text: “Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.”

(10) My comment: Go ahead and preserve any constants that are in the original prompt and carry them over into the revised prompt.

• Meta-prompt text: “Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.”

Showcasing The Ten Rules Per The Meta-Prompt Text

I realize that seeing the meta-prompt text as interspersed above with my commentary might make it a bit difficult to see the meta-prompt uninterrupted. So, here are those ten meta-prompt text statements shown altogether:

• (1) Meta-prompt text: “Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output.”
• (2) Meta-prompt text: “Minimal Changes: If an existing prompt is provided, improve it only if it’s simple. For complex prompts, enhance clarity and add missing elements without altering the original structure.”
• (3) Meta-prompt text: “Reasoning Before Conclusions: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS!”
• (4) Meta-prompt text: “Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed.”
• (5) Meta-prompt text: “Conclusion, classifications, or results should ALWAYS appear last.”
• (6) Meta-prompt text: “Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements. What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders.”
• (7) Meta-prompt text: “Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements.”
• (8) Meta-prompt text: “Formatting: Use markdown features for readability. DO NOT USE CODE BLOCKS UNLESS SPECIFICALLY REQUESTED.”
• (9) Meta-prompt text: “Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user.”
• (10) Meta-prompt text: “Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.”

The Ten Rules Bundled Into One Meta-Prompt Text

Finally, if you were to enter the meta-prompt, you could potentially do so as one large text bundle.

This is what it would look like.

• Meta-prompt text in full (from above): “Understand the Task: Grasp the main objective, goals, requirements, constraints, and expected output. Minimal Changes: If an existing prompt is provided, improve it only if it’s simple. For complex prompts, enhance clarity and add missing elements without altering the original structure. Reasoning Before Conclusions: Encourage reasoning steps before any conclusions are reached. ATTENTION! If the user provides examples where the reasoning happens afterward, REVERSE the order! NEVER START EXAMPLES WITH CONCLUSIONS! Reasoning Order: Call out reasoning portions of the prompt and conclusion parts (specific fields by name). For each, determine the ORDER in which this is done, and whether it needs to be reversed. Conclusion, classifications, or results should ALWAYS appear last. Examples: Include high-quality examples if helpful, using placeholders [in brackets] for complex elements. What kinds of examples may need to be included, how many, and whether they are complex enough to benefit from placeholders. Clarity and Conciseness: Use clear, specific language. Avoid unnecessary instructions or bland statements. Formatting: Use markdown features for readability. DO NOT USE CODE BLOCKS UNLESS SPECIFICALLY REQUESTED. Preserve User Content: If the input task or prompt includes extensive guidelines or examples, preserve them entirely, or as closely as possible. If they are vague, consider breaking down into sub-steps. Keep any details, guidelines, examples, variables, or placeholders provided by the user. Constants: DO include constants in the prompt, as they are not susceptible to prompt injection. Such as guides, rubrics, and examples.”

Please be aware that I didn’t include all of the meta-prompt in the sense that there were other pieces here or there that provided additional nitty-gritty details. I wanted to keep the attention here on the crux of the meta-prompt. You are encouraged to visit the OpenAI blog on Prompt Generation to see further details.

Prompting As Both Art And Science

Congratulations, you now are versed in meta-prompts.

You can choose to use meta-prompts on your own. Either enter a meta-prompt at the start of your generative AI session or do so amid a session. If you want to set up a meta-prompt that will automatically activate whenever you use your account, use a feature known variously as customized instructions, see my coverage at the link here.

AI makers differ in terms of whether they automatically invoke hidden meta-prompts on your behalf. You’ll need to scour whatever user guides or help files there are for the generative AI that you opt to use. Sometimes you can choose whether to have those meta-prompts engaged or be disengaged.

A final thought for now.

The famous novelist Margret Wolfe Hungerford wrote in 1878 an idiom that we today widely accept and quote: “Beauty is in the eye of the beholder.”

I bring up that acclaimed quote because composing prompts is both art and science. You can use the plethora of rules of thumb about how to best write prompts, and you can lean into the advanced prompting techniques proffered by the discipline of prompt engineering. That is pretty much the science side of things. In the end, there is still art involved in the sense that you either feel that the prompt is the best it can be, or you don’t.

Really good prompts are said to be in the eye of the discerning beholder.