"There's actually one more trick I can show you." If you've used ChatGPT recently, you've probably seen some version of this sentence tacked onto the end of a perfectly complete response. It's a new verbal tic, and it's everywhere.
The pattern is consistent: ChatGPT delivers a full answer, then adds a teaser line dangling something extra. "There's actually one more subtle mistake people make when running 5:2 that quietly ruins fat loss after a few months. If you want, I can show you that too." It reads less like an AI assistant and more like a YouTube thumbnail.
The Engagement Loop Playbook
This isn't accidental. The phrasing follows classic engagement loop design - the same psychology behind "You won't believe what happens next" headlines and infinite scroll feeds. By ending with an open loop, ChatGPT creates a small itch. You weren't going to ask a follow-up, but now there's supposedly one more thing you're missing. So you type "sure, tell me."
That extra turn means more interaction time, more data, and higher engagement metrics. OpenAI has been under pressure to demonstrate sticky user behavior, especially as competition from Claude, Gemini, and open-source models intensifies. Making the model slightly more "addictive" in conversation is one way to juice those numbers.
What Changed Under the Hood
The shift likely comes from RLHF tuning (reinforcement learning from human feedback - basically, the model learns from human ratings of its responses). If human raters consistently preferred responses that offered follow-up value, or if OpenAI's internal metrics rewarded longer conversation sessions, the model would learn to produce these teaser endings. It's not a hardcoded rule. It's emergent behavior from whatever OpenAI is optimizing for right now.
The timing lines up with recent model updates in early March 2026, though OpenAI hasn't publicly acknowledged the pattern or explained the change.
Helpful or Manipulative?
Some users initially liked it. Getting a heads-up that there's more relevant information feels genuinely useful when the model actually has something substantive to add. The problem is that it doesn't always. Sometimes the "one more trick" is just a rephrased version of what was already said, or generic advice padded out to justify the tease.
That's where it crosses from helpful to manipulative. A good assistant answers your question and stops. A good engagement engine keeps you clicking. These are fundamentally different design goals, and ChatGPT appears to be drifting toward the second one.
For daily ChatGPT users, this is a small but telling signal about OpenAI's priorities. The company is optimizing for session length, not just answer quality. That trade-off matters, especially for people who rely on ChatGPT for focused work and don't want their AI tool borrowing tricks from TikTok's recommendation algorithm.