Thinking with AI

How to use AI as a thinking partner, not a ghostwriter

An AI will happily do your thinking for you. Set it up the other way round and it becomes something more useful: a second mind that questions, challenges and stretches ideas that are still yours.

By the Offo teamUpdated 9 min read

The short answer

To use AI as a thinking partner, have it respond to your thinking instead of replacing it. Start with your own rough idea, ask for questions, objections and other routes rather than finished answers, and keep the pen: you decide what stays, and you write the next version in your own words.

  • Think first, then ask. Bring a rough idea, not a blank page.
  • Ask for questions and pushback, not finished work.
  • Keep the exchange short, so you stay in charge of the thinking.
  • Write the next version yourself.

What is an AI thinking partner?

An AI thinking partner is an AI you use to develop your own ideas rather than to produce finished work. You bring the idea, however rough. It replies the way a sharp, curious friend would: it asks what you meant, spots the assumption you skipped, offers a route you hadn't considered, and hands the problem back to you.

The opposite is AI as a ghostwriter. You describe what you want, and it writes the essay, the plan or the list of ideas for you. That's often useful. It just isn't thinking, or at least, it isn't yours.

Programmers have known the first idea for decades. In The Pragmatic Programmer (1999), Andrew Hunt and David Thomas describe "rubber ducking": explaining your code, line by line, to a rubber duck on your desk. The duck says nothing, yet the bug often shows itself halfway through the explanation, because explaining forces you to be precise. A good AI thinking partner is a rubber duck that can ask questions back.

Why the difference matters

How you use AI changes how much thinking you do. Two recent studies are worth knowing about, along with their limits.

  • Confidence shifts the work. In a survey of 319 knowledge workers describing 936 real tasks, researchers at Microsoft Research and Carnegie Mellon University found that the more people trusted generative AI with a task, the less critical thinking they reported doing. The more they trusted themselves, the more they did (Lee et al., 2025). The data is self-reported, but the pattern is clear: confidence in the tool can quietly stand in for effort.
  • Outsourced essays don't stick. In an MIT Media Lab study, 54 people wrote essays with ChatGPT, with a search engine, or with nothing but their own heads, while their brain activity was recorded. The ChatGPT group showed the weakest brain connectivity, and in the first session 83% of them couldn't quote a single line from the essay they had just written (Kosmyna et al., 2025). It's a small, early study, published as a preprint, but it matches what many people notice: words you didn't wrestle with don't stay with you.

Neither study says AI is bad for thinking. Psychologists call it cognitive offloading when we hand mental work to the world around us, from shopping lists to calculators (Risko & Gilbert, 2016), and offloading is often the smart move. The question is what you offload. Hand over the typing, the formatting and the fact-finding, and you free up attention. Hand over the framing, the judgement and the first draft of your idea, and you lose the part that made it yours.

There's an upside hiding in the research too. People remember what they generate far better than what they read, a finding so reliable it has a name: the generation effect (Slamecka & Graf, 1978). A thinking partner that makes you produce the next thought puts that effect to work for you instead of against you.

Five things to ask a thinking partner

Most of the value comes from asking for the right kind of reply. These five cover almost everything:

  1. “What question am I avoiding?” Every idea has one. A good partner finds it faster than you will on your own.
  2. “What's the weakest assumption here, and how could I test it cheaply?” This is pushback with a purpose. Not “this won't work”, but “this is the bit to check first”.
  3. “Give me a genuinely different way to reach the same goal.” Not a variation on your idea: a different route. It stops your first idea becoming your only one.
  4. “What's the smallest next step I could take this week?” Thinking that never turns into doing isn't much use. Small and concrete beats ambitious and vague.
  5. “Tell me what you think I mean.” If the summary surprises you, your idea is fuzzier than you thought. Better to find out now than when someone else reads it.

Your pageNewsletter idea:
a weekly roundup of every gig in town?

Offo writes backUseful, but “every gig” is the part I'd worry about. Listings go stale by Wednesday, and you'd spend every Sunday chasing venues. What if it were three gigs worth leaving the house for, picked by you, every Friday? Less to chase, and more of your taste in it. Who's the first person you'd send it to?

An illustrative example of a note and the kind of reply Offo writes back: short, specific, and ending with a question.

Better questions, side by side

The easiest way to change how you use AI is to change the shape of the question. Ask for output and you get output. Ask for leverage on your own thinking and you get better thinking.

Instead of asking…Try asking…
“Write me a business plan for a board game café.”“What would a sceptical landlord ask about my board game café?”
“Give me ten names for my newsletter.”“What do the three names I like have in common, and what's missing from all of them?”
“Rewrite this paragraph.”“Which sentence in this paragraph is doing the least work?”
“Should I take this job offer?”“What am I not saying about this job offer?”
“Plan my lesson on fractions.”“Where will my students get stuck in this lesson on fractions?”

Habits that keep the thinking yours

Think for five minutes before you ask

Asking too early anchors you. Once you've read a confident answer, it's hard to think past it. Spend five minutes on your own first: write the idea down, sketch it, list what you know and what you don't. Then ask. You'll get a better reply, because there's more of you in the question.

Keep replies short

A long answer invites you to read instead of think. Ask for a short reply, a question or a single objection, and you stay in the conversation. A few lines you can argue with are worth more than a page you'll skim.

Write the next version yourself

When a reply sparks something, don't copy it. Write the next version of your idea in your own words, even if it ends up close to what you read. That's where the generation effect earns its keep, and where the idea becomes yours again.

Disagree on purpose

A thinking partner is only useful if you're willing to argue with it. When a reply feels wrong, say why. Half the time, explaining why it's wrong shows you what you actually believe.

Keep the judgement calls

Taste, values and final decisions stay with you. An AI can lay out the trade-offs of a decision; only you know which ones you can live with.

Try this

Next time you're about to ask an AI for an answer, write your own answer first, in three lines. Then ask it for the weakest point in your version. Notice how different that conversation feels.

When a ghostwriter is fine

None of this means you should never let AI write for you. It's fine, often sensible, when the output matters more than your understanding of it: a polite reply to a routine email, tidying a document's formatting, a first pass at meeting notes you'll check anyway.

The skill isn't refusing help. It's knowing, each time, whether you want the thing done or want to get better at doing it. When the work is your idea, your argument or your decision, keep the pen.

Why a page works better than a chat window

Chat windows are built for questions and answers. You type a prompt, you get a wall of text back, and the conversation scrolls away from the thing you were thinking about. That shape nudges you towards the ghostwriter habit.

A page is different. Your idea stays in the middle, where you put it. You can sketch as well as write, draw an arrow between two thoughts, and come back tomorrow to see the whole thing at once. Writing by hand slows you down just enough to choose your words. (Our guide to thinking on paper goes into why that helps.)

That's the idea behind Offo, a notebook for iPad and Apple Pencil that writes back. You sketch or draft an idea and send it, and Offo replies beside it on the page, in handwriting, usually with a question. You can ask it to push back, to find another way or to suggest what to do next, and then the pen is yours again. It won't do your thinking for you, but you won't have to think alone.

Questions people ask

Does using AI make you worse at thinking?

Not by itself. Research suggests the risk comes from how it's used: people who lean on AI with high confidence report doing less critical thinking, and people who have AI write for them remember less of the result. Using AI to question and challenge your own ideas, rather than to produce them, keeps you doing the hard part.

What's the difference between an AI thinking partner and a chatbot?

A chatbot is built to answer: you ask, it produces. A thinking partner is built to respond to your thinking: it asks questions, pushes back, suggests other routes and hands the next step back to you. The same AI model can do either. The difference is in what you ask for and how the tool is designed.

What should I ask an AI to help me think?

Ask for leverage, not output: the question you're avoiding, the weakest assumption and a cheap way to test it, a genuinely different route to the same goal, the smallest next step, or a summary of what it thinks you mean.

Can AI help me brainstorm without taking over?

Yes, if you generate first and ask second. Write your own ideas down before you ask for anything, then use the AI to question them, combine them or find the gaps, rather than to produce a list for you. Our guide to brainstorming on your own has techniques to try.

Is Offo a chatbot?

No. Offo is a notebook for iPad that writes back on the page. You sketch or write an idea, and Offo replies beside it in handwriting, usually with a question, so you keep thinking rather than hand the thinking over. Offo's replies are written by Claude, an AI model made by Anthropic.

Sources

  1. Hunt, A. & Thomas, D. (1999). The Pragmatic Programmer: From Journeyman to Master. Addison-Wesley.
  2. Lee, H.-P. et al. (2025). The impact of generative AI on critical thinking: self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. CHI '25. doi.org/10.1145/3706598.3713778
  3. Kosmyna, N. et al. (2025). Your brain on ChatGPT: accumulation of cognitive debt when using an AI assistant for essay writing task. Preprint. arxiv.org/abs/2506.08872
  4. Risko, E. F. & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. doi.org/10.1016/j.tics.2016.07.002
  5. Slamecka, N. J. & Graf, P. (1978). The generation effect: delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604. doi.org/10.1037/0278-7393.4.6.592

Try it on your next idea

Offo is a notebook for iPad that writes back. Sketch or draft an idea, send it, and Offo replies on the page with a question, a push back or a next step.