Meta-prompting—using AI to help you craft better prompts—is one of the most powerful yet underutilized techniques for improving your AI interactions. While the concept might sound complex, it's simply about asking AI to help you write more effective prompts for future use. I first mentioned this technique early in AIGTC's life, before it even had a name, and its importance has only grown since then.
Generally, I'm a fan of using relatively simple prompts then using conversations to refine output. That method still works quite well in many cases. In fact, that's how most of my interactions with AI go. I ask a fairly simple question, then work with AI to refine the output until I get what I want.
When to Use Meta-Prompting
There are two major times when this isn't the best approach. The first one is when you use AI to carry out a task routinely. For example, I'm writing this article in what's known as a freewrite; I just write in one continuous stream without worrying about editing. In fact, I use a special device, a Freewrite Smart Typewriter to help me focus on getting my thoughts down before editing. Basically, I separate the writing from the refining. But, these first drafts are not ready for publication. So, after writing the draft, I transfer it into Lex.page, an AI-enabled writing tool to refine the draft into something worth publishing. I've created a saved prompt for Lex that helps streamline this process, since I do it virtually every morning.
If you have a task that you do repeatedly, it may be worth investing some time in refining your initial prompt to cut back on some of the back-and-forth of refining your output. A good way to do this is to ask AI to help you build a more effective prompt. At the end of one of your sessions, ask AI, "Please create a prompt I can use in the future that will help me accomplish this task more efficiently" or something along those lines.
Building Your Prompt Library
You may find it useful to do a few rounds of refinement, but once you have the prompt crafted to your satisfaction, save it in something like Notepad or Notes, then just copy and paste the prompt, changing any details as necessary. For example, if you have ChatGPT review exams prior to giving them, you might want to save that prompt once you have it the way you want it. Over time, you can build up a nice prompt library that will help make your interactions with AI more efficient.
Speaking of prompt libraries, the folks from one of my favorite AI newsletters, The Neuron, have a very nice, free prompt library available at https://neuron-prompt-library.lovable.app/. It's worth noting that some of their prompts are pretty simple, and these are folks that know AI quite well. This should tell you that complex prompts are not always necessary, although they can be useful at times.
Meta-Prompting for Deep Research
The thing that really motivated this column was the release of ChatGPT Deep Research. Unless you're on the $200 per month Pro plan, you only get ten Deep Research sessions per month. (This requires a paid plan of some sort, such as the $20 per month Plus plan.) This limit makes Deep Research a scarce resource that needs to be used wisely. You do NOT want to squander Deep Research sessions. Because of this, it is worthwhile to put some effort into refining your Deep Research initial prompt. Before starting a Deep Research session, I typically use one of the other models, such as 4o, to help me build a solid Deep Research prompt.
To demonstrate these concepts in action, here's a detailed example of how I used meta-prompting to prepare for a Deep Research session on anthropomorphism in technology. This example shows the entire process, from initial meta-prompt to final result.
I wanted to do a deep research report on the effects of anthropomorphism as it relates to technology. Here’s my meta-prompt, which I gave to ChatGPT 4o:
I want to use Deep Research to create a report on the effects of anthropomorphism, especially as it relates to technology, although using animal and other forms of anthropomorphism could also inform the work. The goal is to gain an understanding of the causes and consequences of AI anthropomorphism, building on a foundation of what is known about other forms of anthropomorphism, especially technology anthropomorphism. Also, I want to focus on academic sources from peer-reviewed journals and conferences. Please write a prompt I can give to deep research.
Scroll down to the end of the message to see the prompt ChatGPT recommended.
Maximizing Deep Research Results
Another tip is that you can use other models to refine Deep Research's output. Yesterday, I was creating something for one of my doctoral seminars using Deep Research. The initial result was much too long for my purpose, so I asked ChatGPT 4o to shorten it, which worked pretty well. I could have changed my initial prompt and used another Deep Research session, but that didn't seem necessary. By the way, a smarter approach would have been to use meta-prompting to refine my original prompt, but I wasn't thinking that far ahead, which cost me a Deep Research session. Live and learn.
Another benefit of meta-prompting is that over time you'll write stronger prompts without AI's help. You'll implicitly learn to be a better prompter, which is a nice side benefit.
So, if you haven't tried meta-prompting, you should, especially for repetitive tasks and Deep Research. Pay attention to how AI develops prompts and you'll become a better prompt designer as well.
Want to continue this conversation? I'd love to hear your thoughts on how you're using AI to develop critical thinking skills in your courses. Drop me a line at Craig@AIGoesToCollege.com. Be sure to check out the AI Goes to College podcast, which I co-host with Dr. Robert E. Crossler. It's available at https://www.aigoestocollege.com/follow. Looking for practical guidance on AI in higher education? I offer engaging workshops and talks—both remotely and in person—on using AI to enhance learning while preserving academic integrity. Email me to discuss bringing these insights to your institution, or feel free to share my contact information with your professional development team.
Meta-Prompting Result
Here’s a detailed prompt you can use with Deep Research to generate a comprehensive report:
Prompt:
I am conducting a literature review on the effects of anthropomorphism, particularly in the context of technology. I am interested in academic sources from peer-reviewed journals and conferences that examine the causes and consequences of anthropomorphism in various domains, with a primary focus on AI and technology.
Specifically, I would like the report to include:
Theoretical Foundations of Anthropomorphism
Definitions and conceptualizations of anthropomorphism.
Psychological and cognitive mechanisms underlying anthropomorphism.
Historical perspectives on anthropomorphism in human interactions with non-human entities (e.g., animals, objects, nature).
Technology Anthropomorphism
How and why people anthropomorphize technology, especially AI and robots.
The role of design, human-like features, and behavior in triggering anthropomorphism.
Differences in anthropomorphism between various types of technology (e.g., chatbots, virtual assistants, robots, autonomous systems).
Effects of Anthropomorphism in AI and Technology
Positive and negative consequences of anthropomorphism in human-technology interactions.
Trust, acceptance, and perceived intelligence of AI due to anthropomorphism.
Ethical, psychological, and societal implications.
Comparative Insights from Other Domains
How anthropomorphism in animals, brands, and other non-technological domains informs our understanding of AI anthropomorphism.
Insights from human-animal interaction research, marketing, and product design.
Empirical Evidence & Methodological Approaches
Key experimental studies, field studies, and survey-based research.
Measurement tools and scales for assessing anthropomorphism.
Gaps and opportunities for future research.
Please prioritize sources from top-tier peer-reviewed journals and major conferences in psychology, human-computer interaction (HCI), artificial intelligence, and information systems. Include recent studies (last five years) but also foundational research that has shaped the field. Provide a synthesis of findings along with citations.
This prompt should help Deep Research retrieve high-quality, relevant academic sources for your report. Let me know if you’d like to tweak it!
How would prompting for peer-reviewed resources work if Deep Research does not have access to subscription/paywall journals? Has that barrier been broken?