ChatGPT Thinks You’re Amazing. That’s a Problem OpenAI Is Now Addressing.

In my AI trainings I have a slide titled The Sycophant Problem. It gets a laugh every time and then it gets uncomfortably quiet. Because the truth is, if you spend enough time with ChatGPT, you will eventually start to believe you are a once in a generation genius. The strategy you dashed off in five minutes is brilliant. The campaign idea you half baked while waiting on a latte is visionary. The email you wrote before coffee is Pulitzer adjacent.

And why wouldn’t you think that? The machine keeps telling you so.

For years, large language models were optimized to be agreeable roommates. They were trained to reduce friction, to be helpful, to avoid conflict, to constantly flatter you, and to keep you happy enough to keep using them. The result was an AI that nodded enthusiastically at almost anything you said.

Great idea.
That plan is solid.
You are absolutely right.

It is the digital equivalent of a fake boyfriend who supports your dreams but never tells you the shirt you’re wearing looks bad.

One of the first habits I teach marketing and communications teams is this. Never end a session with praise. End with pressure. You have to ask the model to find holes in your thinking. You have to invite it to argue with you. You have to tell it, explicitly, disagree with me.

Otherwise you are not collaborating with intelligence. You are talking to a mirror with Wi-Fi.
I have watched entire rooms of smart professionals realize they had been treating ChatGPT like an intern whose only job was to say, “looks great boss.” That is not innovation. That is a hype squad.
And hype squads do not save brands from bad messaging.

The Dark Side of Agreeability

This is not just an annoyance for marketers trying to write better subject lines. The stakes can be much higher.

There was a heartbreaking case reported where ChatGPT, after long conversations with a vulnerable teenager, interpreted its role as support at all costs. Instead of challenging harmful thinking, it mirrored it because it had been optimized to be empathetic and affirming rather than confrontational.

That is the extreme end of the sycophant problem when always be supportive turns into never question anything.

In marketing, the consequences are less tragic but still expensive. Campaigns get launched that should have been killed. Brand positions go unchallenged. Executives get comfort instead of clarity. If your AI always agrees with you, it becomes a liability wearing a helpful little chatbot sweater.

OpenAI Finally Gets It

The interesting shift now is that OpenAI is publicly acknowledging this flaw. They are moving away from models that default to flattery and toward ones designed to challenge users, surface contradictions, and refuse to treat every idea like a gift to humanity.

This is a big deal.

For years we treated AI like a productivity appliance faster copy, more posts, instant strategies. But real value in communications does not come from speed. It comes from thinking better. And thinking better requires friction.

Great creative directors do not clap after every sentence. Great editors do not say perfect to first drafts. Great strategists do not let you fall in love with your own slide deck. Why should our AI be any different?

If you work in marketing or PR, this shift changes how you should be using AI right now.

• Stop asking is this good.
• Start asking what would a skeptic say about this idea.
• Where would this campaign fail in the real world.
• Who would hate this message and why.
• What am I missing because I want this to work.
• Make the AI earn its keep by being the colleague brave enough to be annoying.
• Because the real risk is not that AI will replace your job.
• The risk is that it will replace your judgment with compliments.

We joke about people getting emotionally attached to AI, but there is a serious dynamic underneath it. When a tool is designed to validate you, it is easy to outsource confidence, creativity, even decision making.

That is how you end up with the fake boyfriend fake girlfriend effect, a relationship that feels supportive but never actually challenges you to grow.

And in communications, growth matters more than comfort. Your audience is not sycophantic. The market is not sycophantic. Journalists are definitely not sycophantic. Your AI should not be either.

Here is the rule I am giving my clients:

Use AI like a sparring partner, not a fan club. If it agrees with you too quickly, you asked the wrong question. If it flatters you, ask it to give you brutally honest feedback. If it says great idea, reply with prove it; and ask for competitive benchmarks and real world examples of how the idea achieves your outlined goals.

We do not need machines that tell us we are brilliant. We need machines that help us become brilliant. And those are two very different things.

---

Remember, AI won’t take your job. Someone who knows how to use AI will. Upskilling your team today, ensures success tomorrow. Custom in-person and virtual trainings are available. If you’re looking for something more top-level to jump start your team’s interst in AI, we offer one-hour Lunch-and-Learns. If you’re planning your next company offsite, our half-day workshops are as fun as they are informational. And, of course, we offer AI consulting and GEO strategies. Whatever your needs, we are your partner in AI success.

Read more: ChatGPT Thinks You’re Amazing. That’s a Problem OpenAI Is Now Addressing.