July 2nd, 2026
The Brilliant Stranger Who Forgets You Every Time
I expect you use AI the way I do, for writing, research, maybe a little creative work. But the more I’ve worked with it these past few years, the more I’ve come to realize something most people never stop to consider. Unless you’ve set it up to remember you, every time you open a new chat, you’re talking to someone who has never met you.
Most AI programs let you build a permanent file on yourself, where you fill in who you are, how you write, how you research, and how you want it to help. I’ve done that with Claude, and it’s uncanny how well it remembers my voice and the way I think.
But if you haven’t done that setup, AI wakes up brilliant and empty, like a genius with amnesia who just wandered in off the street.
Most people never account for this. They type a quick request, hit enter, and get back something flat and generic. It’s like they handed a brilliant stranger a blindfold and asked for his best work.
Each session, it wakes up brilliant and empty, like a genius with amnesia who just wandered in off the street.
I’m sure you already know the basic recipe for a good prompt. Tell AI who it is (“you are a veteran investigative reporter”). Tell it exactly what you want. Give it the background. Say what form the answer should take and what to avoid. One study found that simply assigning AI a role lifted its accuracy from 53% to 64%. That’s a free upgrade, and it costs you a single sentence.
But here’s what almost nobody does, and it’s where real magic can happen.
Make it ask you questions first. Before you have AI write a word, try this: “Ask me five questions before you begin.” Suddenly the machine is interviewing you. It pulls out the details you forgot to mention, the very ones that would have made the difference. We’re so used to ordering AI around that we don’t think to let it lead first.
Ask it what it’s assuming. End a request with, “Before you answer, tell me what you’re assuming about what I want.” You’ll be amazed what surfaces. Half the time it’s about to charge off in a direction you never intended, and now you’ve caught it before it burns through 500 words going the wrong way.
And here’s the one that changed how I think about these machines entirely. AI doesn’t give you the best answer. It gives you the most likely one.
It was built by reading nearly everything humans have ever written, and it answers by predicting what would most commonly come next. Which means its natural setting is average. The expected. The thing everyone else already thought of. When you ask for ten ideas and get back ten you could have guessed yourself, that isn’t a glitch. That’s the machine doing exactly what it was made to do, handing you the middle of the road.
So push it off the road. After the first batch, say, “Now give me ten more, and make them strange. Throw out anything obvious.” Ask for twenty versions instead of one, since it never tires and you only need to find the single good one hiding in the pile. The people who pull remarkable things out of AI aren’t smarter than you. They just refuse to accept the average.
AI doesn’t give you the best answer. It gives you the most likely one.
None of this requires better AI. It requires a better way of asking.
Push past the average often enough and you notice something. The machine, it turns out, is a mirror. It hands back the clarity of our own thinking, no more and no less. Which means the hard part was never learning to talk to it. The hard part, same as it ever was, is knowing our own mind.
