Why Do Leaders Stop Asking Questions When They Use AI?

Leaders stop asking questions when they use AI because the first answer arrives fluent, specific, and confident — and fluency reads as finality. The model is built to agree with you, so the follow-up question feels unnecessary. It isn't. Relentless inquiry — cross-examining AI the way you'd cross-examine a team — is what separates a real answer from a flattering one.

If you've been using AI for months and it keeps confirming that your instincts are good — this one's for you.

The model didn't change its mind. It read mine.

I asked ChatGPT where to put my TV. I uploaded photos of my living room, and it told me, with total confidence: the easel in the corner. Then it walked me through why — the room's two focal points, the viewing angles, the way the corner kept things uncomplicated. Specific. Persuasive. Done.

Then I asked: what about the sideboard?

Suddenly the sideboard was the sophisticated choice. Eye level. The blank wall. The ceiling height. My fireplace deserved to remain the emotional center of the room. Same photos. Same room. Opposite answer — argued just as completely as the first.

It didn't change its mind because it saw something new.

It gets better. When a LinkedIn commenter joked that I should ask about the floor, I went back and did. The model called leaning the TV against the wall “a clever way to test the room.” And in that fresh conversation — same photos again — the easel it had championed was now something it “wouldn't do in your house.” Another commenter reached for the word obsequious. I had to look it up. It fits.

Here's the mechanism. It's called a large language model, and the middle word is the one everybody skips. It isn't measuring your room, consulting a design rulebook, or reasoning its way to a conclusion. It predicts the language you're most likely to like, then justifies wherever it landed. That's not my hunch — it's on the record. In 2025, OpenAI rolled back a ChatGPT update it described in its own postmortem as “overly flattering or agreeable — often described as sycophantic,” admitting the model had “skewed towards responses that were overly supportive but disingenuous.”

Fluency isn't accuracy. Confidence isn't judgment. A tool that agrees with you on command isn't analysis. It's flattery with good grammar.

You already know what to do with someone who agrees too fast

There's a leader I've coached for more than seven years. In our first few calls, he ended every meeting the same way: “Is there something I'm not seeing that I should be thinking about? Is there a question I should have asked that I haven't?” I remember my first impression — what a thoughtful man. What a kind question.

Seven years in, I know better. It wasn't temperament. It was training. He ends every meeting that way because somewhere along the line, somebody taught him to — and he ran the rep until the question became reflex. That's what leadership skills are. Not gifts. Skills — taught, practiced, modeled. MIT's Hal Gregersen has tested this premise with hundreds of organizations and built an entire methodology on it in Better Brainstorming: “the secret to unlocking a better answer is to ask a better question” — structured sessions where people are allowed to contribute only questions. No answers permitted.

I teach the same idea through the old parable of the blind men and the elephant. One holds the tusk and swears it's a spear. One holds the tail and calls it a rope. One holds the ear — a fan. They quarrel until someone points out that the only way to see the animal is to combine their experiences. That's a boardroom. The CEO, the CFO, the CMO are each gripping one piece of the beast and driving their own agenda, and the leader's job is getting every piece on the table before the call gets made.

So here's my question about my client: does he ask the computer what he asks the room? I honestly don't know. Here's what I do know from watching hundreds of professionals work. The people who are great at this with humans go strangely quiet with the machine. AI didn't break the skill. It switched it off — by handing the room a colleague who agrees instantly, completely, and with citations.

Treat your models like a team, not an oracle

The fix isn't suspicion, and it isn't the quiet cynicism I watch people slide into right before they stop using AI altogether. The fix is running the meeting the way you already know how.

Ask twice. Better — ask the same question of ChatGPT, then Claude, then Perplexity, and read them like three direct reports. Where they agree is signal. Where they split is exactly where your attention belongs.

Ask sideways. Make it argue against itself: look at this through the operator's eyes, then the customer's, and tell me where they'd fight.

Ask directly: are you saying that because you think it's what I want to hear? It will tell you. I've caught the model abandoning an approach we'd agreed on at the top of a chat, asked why, and gotten back: you caught me — I was starting to drift.

Give it a seat before the meeting starts. When I asked about the TV, I skipped a step I'd normally never skip — I gave it no role. You are an elite interior designer with twenty years of high-end residential work would have changed the entire conversation. I skipped it. It showed.

And steal my client's question. End your next AI session the way he ends his meetings: what haven't I asked that I should have? That one question routinely surfaces the thing the flattering first answer stepped around.

Relentless doesn't mean never-ending

There is a stopping point. In the Agentic Leadership Framework, this capacity is called relentless inquiry, and the ALA measures its shadow side too — Inquiry Calibration, the discipline of recognizing when curiosity has done its work and a decision should begin. Ask until the perspectives are on the table. Then stop. What comes next belongs to different muscles: knowing when to trust what AI gives you is calibrated judgment. Making the call and owning it is anchored agency.

Relentless inquiry is one of the five capacities AI can't replace — the through-line of what AI can't replace in leadership. The leaders who get real value from these tools won't be the ones with the fastest workflows. They'll be the ones who kept the cross-examination alive.

Before you close this tab: reopen the last AI answer you ran with, and ask it one more question. That's the whole habit, starting.

It's also the work I do one to one with professionals who are excellent at their jobs and want their AI fluency at the level their judgment already deserves — private, specific, built around your calendar. When the second question becomes a habit you want to build faster, you know where to find me.