The AI Doesn’t Need a Better Prompt. It Needs a Better Partner.

Why do some people get so much better results from AI than others?

Not usually because they know a secret prompt formula. The real difference, most consistently, is whether AI gets treated as a partner in the work or as a replacement for the thinking part of it entirely. Handing over a task along with all the judgment calls that go with it — what matters here, what the real constraints are, what a good answer actually looks like — tends to produce output that technically answers the question and still misses the point. The people who consistently get strong results are the ones still doing the thinking; the AI is doing the execution.

That’s a real distinction, not a motivational line. It shows up concretely in how a prompt gets written, what happens when the AI is uncertain, and what happens when the person asking doesn’t fully understand the topic themselves yet.

Does the first prompt in a conversation matter more than the rest?

Yes, disproportionately so — especially for anything complex, or where a specific outcome actually matters. The first prompt sets the frame for everything that follows in that conversation; a vague opening produces a conversation built on a shaky foundation, and every follow-up message ends up patching a foundation rather than building on a solid one.

This isn’t just a hunch. It lines up with what’s called the anchoring effect in how language models process a conversation — a well-documented pattern, not a folk theory, where models are disproportionately influenced by whatever information arrives first and adjust less than they should from there. Multiple recent studies confirm this pattern holds for modern models, not just older ones. Practically, that means the highest-leverage moment to be detailed isn’t “somewhere in the conversation” — it’s the opening message, before anything else has been said.

Do you still need to tell the AI what role to play?

Less than it used to seem. Framing like “you are an experienced SEO specialist” used to feel like a necessary setup step. In practice now, what actually matters is describing the real inputs and the real desired output clearly — the persona framing turns out to add less than it seems like it should.

That’s not just a personal impression, either. A large-scale academic study testing 162 different personas across major LLM families on over 2,400 factual questions found that adding a persona to a prompt had no meaningful positive effect on performance compared to giving no persona at all — and in some cases, a small negative one. Skipping the role-play setup and going straight to a precise description of the task tends to be time better spent than crafting the perfect persona.

What should you do when you don’t understand the topic yourself?

Ask the AI to explain it before asking it to do the task — not the other way around. It’s tempting to hand off something unfamiliar entirely, but a prompt written from genuine understanding is sharper and more specific than one written from a guess at what the task even involves, and sharper prompts are what produce better output in the first place. Understanding the topic first isn’t a detour before the real work — it’s part of what makes the next prompt good enough to be worth sending.

This also means a useful move on a complex task is explicitly telling the AI to ask before assuming. Adding a line like “if anything here is unclear, ask before proceeding” changes how a complex request gets handled — instead of the AI filling gaps with its own assumptions (which then need to be caught and corrected after the fact), it surfaces what’s actually ambiguous up front. That’s real time saved, not a nice-to-have — catching a wrong assumption before output gets generated is faster than catching it after.

What’s the most common mistake people make?

Two, and they’re related. The first is writing short, vague requests and then spending the rest of the conversation correcting what came back — which usually takes longer overall than writing one detailed opening message would have. The second, more consequential mistake is outsourcing the whole job, thinking included: letting the AI make the calls that actually needed a human’s judgment, then accepting whatever came out because it sounded confident.

The fix for both is the same underlying habit: think, learn, and work with AI — not hand it the entire task and hope. That’s not a caveat on using AI well. It’s the actual definition of using it well.

What’s next

None of this replaces understanding the actual problem you’re solving — it’s what makes the difference between an AI that executes your thinking well and one that’s quietly making decisions you never meant to hand over.