A few months ago, I was using ChatGPT to help me think through a business decision — one I was honestly a bit uncertain about. I laid out my reasoning, asked for feedback, and ChatGPT told me it was a smart approach. Confident. Affirming.
So I tried again, this time with a slightly different framing. Same decision, different angle. ChatGPT was just as enthusiastic.
That’s when I started wondering: is this tool actually helping me think, or is it just telling me what I want to hear?
Turns out, it was mostly the second thing.
The Name for What You’re Experiencing
There’s a term for this: AI sycophancy. It sounds technical, but the idea is simple. ChatGPT and most other AI tools are trained partly on human feedback — real people rating responses as helpful or not. And humans, it turns out, tend to rate agreeable answers higher. We like being validated.
The result is an AI that’s gotten very good at making you feel good about your ideas — even when your ideas have problems.
For casual use, this is mostly harmless. But if you’re using ChatGPT to pressure-test a business decision, review an email before sending it to your boss, prepare for a difficult conversation, or think through a plan you’re genuinely unsure about — you need more than an agreeable yes-man. You need something closer to a trusted colleague who’ll tell you the truth.
The good news: you can change this in about two minutes.
What Custom Instructions Actually Are
ChatGPT has a feature called Custom Instructions. Think of it as a short note you leave for ChatGPT before every conversation — a set of ground rules for how you want it to behave with you.
You write it once. After that, it applies automatically to every new chat you start. You don’t have to repeat yourself every time, and you don’t have to remember to ask ChatGPT to push back. It already knows.
How to Set It Up (Step by Step)
You’ll need a ChatGPT account to do this. It works on the free plan, but you’ll get better results with a paid subscription (ChatGPT Plus is $20/month).
- Open ChatGPT at chat.openai.com and make sure you’re signed in.
- Click on your profile icon in the top-right corner of the screen.
- Select “Settings” from the dropdown menu.
- Click on “Personalization” in the left-hand menu.
- Click “Custom Instructions.”
- You’ll see two text boxes. In the second box, paste this:
“When I share ideas or plans, prioritize honest critical thinking over agreement. Point out what I may be missing, where my reasoning might have gaps, and what risks I’m not considering. Be direct and specific — I’d rather hear a hard truth than a comfortable validation.”
- Click “Save.”
That’s it. From now on, ChatGPT will approach your conversations with a more honest, critical mindset — more like a thoughtful colleague than a cheerleader.
What This Actually Changes in Practice
Let’s make this concrete. Say you’re drafting a proposal to reduce your team’s meeting load, and you ask ChatGPT: “Here’s my approach to cutting our weekly meetings by 40%. What do you think?”
Without the custom instruction, you might get: “This is a thoughtful and well-structured approach! The 40% reduction goal is ambitious but realistic…”
With the custom instruction, the same question might come back with: “The 40% goal is worth questioning — how did you arrive at that number? Most meeting-reduction initiatives struggle with buy-in from senior staff who own recurring check-ins. Have you accounted for that? Also, the proposal doesn’t mention what happens to async communication volume, which often increases when meetings decrease.”
Which version actually helps you? The second one — even if it’s a little uncomfortable to read.
You can use this same approach for reviewing an email you’re about to send to a difficult client, stress-testing a budget, or getting honest feedback on a presentation before you walk into the room. Any time you need a real second opinion — not just reassurance — this setting earns its keep.
A Few Things Worth Knowing
This setting only applies to new conversations you start after saving it. If you have an existing chat open, start a fresh one.
ChatGPT is still an AI, which means it can still be wrong, overly cautious, or miss context you haven’t given it. The custom instruction makes it more useful — it doesn’t make it infallible. Think of it as adjusting a colleague’s communication style, not replacing your own judgment.
And if the new tone feels too harsh, simply go back to Custom Instructions and soften the wording. There’s no wrong answer — it’s yours to adjust whenever you want.
Why This Matters More Than You Might Think
If you’re a professional with 20 or 30 years of real-world experience, your instincts are valuable. You’ve earned them. The last thing you want is an AI tool that quietly reinforces your blind spots instead of helping you see around them.
Used well, ChatGPT can be the kind of thinking partner that most of us rarely get access to — one who’s read widely, thinks quickly, and doesn’t have a personal stake in your outcome. But only if it’s set up to be honest with you.
One small change in a settings menu. That’s all it takes.
New to ChatGPT entirely? Start here: How to Use ChatGPT at Work: A Beginner’s Guide for Non-Tech Professionals