What prompts help executives think more clearly with AI?

The prompts that help are the ones that make AI ask instead of answer. An MIT study published in April 2026 found AI help raised accuracy 21%, but unassisted judgment fell 15.3% by week four. John Whitmore's GROW model gives the fix: have the assistant state your goal, split facts from interpretations, and ask three questions only you can answer, before any advice.

The default way leaders use an assistant for a hard call is to describe the situation and ask what to do. It feels efficient. The research published this year suggests it quietly costs you the thing you were trying to improve.

In a four-week MIT study published in April 2026 and reported by The Guardian on June 19, 67 participants judged whether news headlines and images were real, with and without an AI assistant. With help, they were 21% more likely to make the right call. But their unassisted performance on new items was 15.3% worse by the fourth week, and about a quarter believed their skills were improving when they were not. The authors note that participants "go along with the system because it sounds knowledgeable," and that probing, guided questioning, rather than prescriptive answers, is what protects judgment. A separate study of 173 Georgia Tech students, published in July 2026, points the same way: an AI tool that kept asking students to explain their reasoning reduced their score decline on harder material by about 3.3 points of an 11.8-point drop.

None of this would surprise John Whitmore. In Coaching for Performance he argues that telling creates dependence while questions build the two things performance actually rests on: awareness (seeing the situation as it is) and responsibility (genuinely choosing what to do). His GROW sequence (Goal, Reality, Options, Will) exists to stop people jumping to options before they have looked at reality. He also has a rule about questions: open ones starting with What, When, How, and not Why, which tends to sound like criticism and produces defence instead of thought. An assistant that hands an executive a recommendation skips Goal and Reality entirely. So the useful prompt forbids advice and makes the AI do Whitmore's first two steps on your own words.

I wrote that prompt, with a realistic founder dilemma as the test note, and ran it once:

What it did well: the Reality split is the most useful part. It caught that "replace her" assumes hiring a VP means her exit, which opens an option the note had closed without noticing. It flagged "I think" in the account-loss prediction and questioned a deadline that may be self-imposed. The goal conflict (a credible board story versus loyalty) is stated plainly. What it did badly: the "question you're avoiding" came back as a yes-or-no question. The prompt only demanded open questions in section 4, but a closed question invites a quick yes and ends the thinking; Whitmore would phrase it as "What would she need to show you by the board meeting?" One of its three written questions, not shown above, lists possible causes (targets, pricing, product, hiring), which drifts toward options. And it returned Markdown symbols, which is cosmetic but tells you it treated the format loosely.

How to use it: write your note yourself, in five minutes, before you open the assistant; the note is where the thinking starts, and the model can only examine what you put in it. Answer the three questions in writing. Only then ask for options, ideally in a fresh conversation, so the advice is built on your answers rather than on the model's first impression. Keep personnel details out of consumer plans with training switched on.

The limit is the one Whitmore would name first: a prompt is not a coach. It cannot notice what you left out of the note, hold you to what you said you would do next week, or hear the hesitation in your voice. It is a good way to get from "tell me what to do" to "here is what I am actually deciding," and that step is the one most leaders skip.

Sources: Over-reliance on chatbots can diminish critical-thinking skills, study finds, The Guardian, June 19, 2026; MIT Media Lab summary of the same study, MIT Media Lab, 2026; Socratic Mind: impact on student learning and higher-order thinking, Georgia Tech and UC San Diego researchers, July 8, 2026; John Whitmore, Coaching for Performance (5th edition); prompt-test record, October 9, 2026.

Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · Why Exploration Is Important for Success