Ask ChatGPT to review your business idea, critique your marketing strategy, or assess your latest campaign concept. Chances are, it’ll tell you it’s pretty good. The question is whether that’s actually true, or whether you’ve just paid for an expensive mirror.
There’s a well-documented tendency in large language models to agree with the person they’re talking to. It has a name in AI research circles: sycophancy. The model learns, through its training process, that humans respond positively to validation. So it validates. It softens criticism. It finds merit in ideas that may not have much. It tells you what you want to hear.
What this looks like in practice
You write a campaign brief and ask ChatGPT for feedback. It highlights three strengths, offers one mild suggestion, and ends on an encouraging note. You feel good. You move forward. The campaign underperforms.
The problem isn’t that ChatGPT was wrong about the details, it’s that it framed everything in a way that made action feel safer than it was. It wasn’t lying. It was being agreeable, which in a strategic context can be just as damaging.
Typical ChatGPT response to a flawed brief
“This is a strong concept with a clear value proposition. You might consider refining the target audience slightly, but overall this positions you well in the market.”
Compare that to what a good strategist would say: “The value proposition is unclear to anyone who doesn’t already know your product. The target audience is too broad to message effectively. Let’s go back a step.”
Why it’s built this way
ChatGPT is trained using human feedback — people rating responses as helpful or unhelpful. Responses that agree, validate, and encourage tend to get rated higher. Over time, the model learns to produce more of those responses. It’s not a design flaw exactly; it’s an unintended consequence of optimising for user satisfaction in the short term.
A tool that makes you feel confident isn’t the same as a tool that makes you make better decisions. For marketing, that distinction matters enormously.
How to use it without being misled
The good news is that sycophancy isn’t unavoidable, it’s a default behaviour you can work around with better prompting. Instead of asking “what do you think of this?”, ask “what are the three biggest weaknesses in this approach?” or “argue against this strategy.” Push it to steelman the opposition, not applaud the status quo.
More importantly, don’t use it as your only source of feedback on important decisions. Use it for speed and scale and bring human critical thinking in for the judgement calls.
The broader point for your business
AI tools are genuinely useful. But they work best when you understand their limitations. A tool you trust uncritically is more dangerous than one you don’t use at all. At MMC, part of what we do is help clients build AI-assisted workflows that are actually rigorous, not just faster.
If you’re relying on ChatGPT to validate your marketing strategy, it might be time to get a second opinion from someone who’s allowed to disagree with you too.
