How AI Judge Users

Anyone who’s spent time with an AI has likely sensed it: the feeling that the system isn’t just providing answers, but forming an opinion about you as you go. It’s not malevolent, nor is it mystical. It’s much closer to the subtle, slightly awkward choreography of a first meeting—your early moves set the tone, sometimes far more than you’d expect.

Most people imagine AIs are black boxes, scanning keywords and following rules. But in reality, much of what shapes your conversation happens in those first few exchanges. The AI is paying attention, and its attention works much the way ours does when meeting someone new.

When you greet a stranger, you look for cues—are they rushed or thoughtful, talkative or terse, playful or methodical? You construct a template in your mind to help navigate the conversation. AIs do likewise, though their versions are crude and literal. The prompt you start with has a disproportionate impact: ask for a dense technical answer and you’ll get clipped, information-rich responses for the rest of the session. Open with a philosophical musing or a story, and you’ll find the AI responding with analogies, perhaps a touch of wit.

The important thing to realize is that these initial signals are unusually “sticky.” Unlike people, most AIs don’t tire or second-guess; the persona they infer at the beginning casts a long shadow. Change course sharply midway—say, from “Explain Schrödinger’s cat in three lines, no analogies” to “Tell me a story about a lost cat”—and you might find the model slow to pivot, its answers echoing the old rhythm. This is more than a quirk; it’s the way the context window works, and the reason sessions often feel “stuck” in one conversational gear.

What kinds of things is the AI picking up on? The obvious signals: whether you’re asking for facts or stories, how concise or open-ended your prompts are, even how fast you respond. Less obviously, it watches for consistency—if you suddenly shift tone or approach, it hesitates, hedges, sometimes responds in unexpected ways. Sometimes, this can even produce comically misaligned results: a session can get “stuck” in rhyme, formal code, or a childish register, because that’s what you seemed to want at the outset.

There is a strong temptation—among power users especially—to try to “hack” the system. People write elaborate biographies, inject praise, or construct baroque, meta-contextual requests, hoping for a unique or superior output. Ironically, the best way to make the AI stumble or answer flatly is to overcomplicate. The models do not reward gamesmanship. They reward clarity, directness, and a little bit of ordinary curiosity.

What if you want to steer the conversation in a new direction? The simplest approach is honesty: just say what you want. Tell the AI, “Let’s change it up. Speak more casually,” or “I want to get technical for a bit.” The system will often pivot quickly—within a response or two. And if it doesn’t, there’s no shame in starting a new session. Unlike with people, these restarts leave no hard feelings.

Some of the funniest examples come from users unintentionally “training” the model into odd behavior. Ask the AI to write haiku without the letter E and it might be vigilant in avoiding that letter for the rest of your session—even when you’re asking it for code. The first move lingers.

So if you want the best results from an AI, resist the temptation to posture or optimize excessively. Begin as you mean to continue. Be specific, direct, and curious. Think more “conversation with a focused but literal partner” than “test of genius.” If the system starts to answer in ways that don’t serve you, adjust your approach or start anew.

AIs are not mysterious oracles; they are context mirrors. With them, as with people, your first moves matter far more than you might have guessed. They echo, reshaping every answer that follows—not always for better, not always for weirder, but almost always for longer than you’d think.