AI as Writing Mentor: Question-Based Rather Than Prompt-Based LLM Assistance

We've come a long way with Large Language Model (LLM) interactions. 

First, we were worrying about crafting the perfect prompts or queries. Now many of us have moved to a more collaborative approach, where the LLM helps you refine and improve your prompts or queries through multiple iterations and what Reed Hepler calls "Conversation Steering." But even this evolved method is still fundamentally prompt-based. What if there's an even more natural alternative?

The Problems with Current Approaches

There are real and growing concerns about AI's impact on our thinking abilities. Research is starting to show that overreliance on AI can actually diminish our cognitive skills, both as professionals and as students. This isn't necessarily a conscious choice since it's just so easy and tempting to outsource our language and writing to AI. When we constantly let the LLMs do our reasoning or thinking and writing, we risk weakening the very mental muscles that make us effective thinkers, writers, and problem-solvers.

Beyond these cognitive risks, there's another fundamental problem: traditional prompt-based AI interactions don't produce writing that feels authentic, least of all to ourselves. It's not just a nagging feeling that the output isn't you. It actually isn't you. The language, the phrasing, the way ideas connect--these are all generated content that may cover your topic but doesn't really capture your authentic voice and sometimes not even your actual thinking process.

A Different Way of Learning

I believe we learn more naturally through conversation and dialogue rather than through traditional information delivery methods. This belief shapes how I think about productive AI interactions. If you're interested in exploring this learning philosophy further, I've written about it in more detail here.

Discovering the Power of Interview-Based Learning

Over four years, I conducted several hundred interviews as part of my FutureofEducation.com project, and that experience taught me that interview conversations can be an extremely effective way of diving into a topic. The interview process doesn't just extract information, it actually helps the guest communicate their ideas in more cogent and immediate ways and makes the material much more accessible to the listener. Witness the explosion of podcasts and interview formats, and even tools like NotebookLM's podcast feature--there's clearly something powerful about this conversational approach to exploring ideas.

The Breakthrough: Flipping the Script

This insight led me to an idea: what if I interviewed experts and then sent those transcripts through large language models to help them write books on their areas of expertise?  In a flash, I then realized I could flip the script entirely on my own interactions with large language models. Instead of me prompting the AI, what if I had the AI interviewed me? So I built a prompt that would tell the large language model how to interact with me in a question-based way. The fact that I can then have these conversation through voice chat rather than typing makes the whole process incredibly natural because it's very much like an actual interview with a person. Well, like with a person who has an encyclopedic memory and sometimes speaks like a robot. But still...

The Irish Wisdom Behind It All

There's a saying that was originally represented to me as an old Irish saying: "How will I know what I think until I hear myself say it?"  I've always loved it because it really reflects my own personality. It turns out that the provenance of the thought is a little richer than that (see https://quoteinvestigator.com/2019/12/11/know-say/), but however expressed it captures exactly what happens in this question-based approach with LLMs. The process of speaking my thoughts out loud in response to questions doesn't just help me communicate what I already know, it actually helps me discover and refine my thinking in real-time. It's like being interviewed by an expert and then getting to edit my words and thoughts afterwards. It's learning through conversation, and then that learning becomes writing. It's an incredibly active and interactive process that keeps me fully engaged.

Real Results: My Personal Experience

I've now used this approach three times, and I'm genuinely floored by what a great way this is to interact with AI and to help me in my writing. The output is literally my specific words and thoughts put into writing, just organized and refined through conversation. The AI isn't generating content for me; it's helping me discover and articulate what I already think about a topic. Plus, I can interrupt (which I do frequently) and change wording to actually reflect what I'm thinking, now what the AI is writinhg. I don't feel the same way I would with interrupting a person because I know the large language model isn't upset with me, so I can make constant corrections.

AI as Writing Mentor, Not Writing Replacement

This approach fundamentally changes the relationship between human and AI. Instead of the AI writing for me, it becomes a writing mentor - asking thoughtful questions, helping me organize my ideas, and guiding me through my own thinking process. It's not replacing my cognition; it's enhancing it. The AI becomes a collaborative partner that helps me access and structure knowledge I already possess, or research what I would additionally like or need to know, rather than generating external content that I then would have to adapt or edit to sound like me.

Applications for Educators and Students

For teachers, librarians, parents, and students, this approach offers exciting possibilities. Instead of worrying about students using AI to cheat or bypass their own thinking, we can teach them to use AI as an intellectual partner that provides an AI form of Socratic teaching. Students can have AI interview them about their research topics, helping them discover what they actually know, identify gaps in their understanding, and then provide help in learning about the areas they need to learn about in order to complete the subject. It's like having a Socratic dialogue with the world's most learned mentor. Teachers can model this process, showing how thoughtful questioning leads to deeper exploration of subjects, the same principle that makes Socratic dialogue so powerful in the classroom. Reviewing the chat log with the student could become an incredible second opportunity for exploring how to learn, shape ideas, and communicate.

The Bigger Picture: A New Vision for Human-AI Collaboration

This shift from prompt-based to question-based AI interactions represents something much larger than just a better writing technique. It's a fundamental reimagining of how humans and AI can work together. Instead of treating AI as a sophisticated content generator that we command, we're discovering its potential as a thinking partner that helps us access our own knowledge and develop our own ideas. This approach takes us away from trying to achieve artificial general intelligence or artificial superintelligence through large language models (which may not actually be possible, although the appearance of it is surely possible), and instead gives us a really good model for how large language models can evolve to help us as humans. This collaborative approach doesn't just produce better writing - it actually strengthens our cognitive abilities rather than weakening them. Perhaps this is the future of productive AI use: not replacement, but a unique intellectual partnership.