WritingYour AI is a mirror — it reflects you, good and bad

Your AI is a mirror — it reflects you, good and bad

AI has no standards of its own. It reflects yours. Give it a clear picture and it hands your judgment back at scale; give it a thin or careless one and it reflects that too — or quietly fills the gaps with invention you'll struggle to spot.

It reflects what you put in front of it

AI has no taste, no standards, and no context of its own. Everything it gives back is a reflection of what you put in front of it — your brief, your data, your examples, your intent. That’s the whole game, and it’s why two people get wildly different results from the same tool. It isn’t that one found the magic prompt. It’s that one held up a clear, complete picture for the model to reflect, and the other held up a smudge.

What you put inWhat it reflects back
A clear brief with your standardOutput that meets your standard
Your real examples and voiceWork that sounds like you, not like a chatbot
Rich, relevant contextSpecific, useful answers
A vague or thin briefGeneric, hedged, or plain wrong — the usefulness dies
Gaps left openIt invents to fill them, and the invention bleeds into the good parts — no seam showing
Your biases and shortcutsThe same, reflected back amplified and authoritative-looking

It reflects your best — and your worst

Held up well, this is a superpower. A clear thinker with a good process and real expertise gets that judgment reflected back at scale, in their own voice. But the mirror is honest in both directions. Feed it a vague brief, messy data, or a careless shortcut and it reflects that just as faithfully — then makes it worse, because it hands your sloppiness back in fluent, confident, authoritative-looking prose. Your blind spots come back magnified. Your bad assumptions come back sounding like conclusions. The tool doesn’t fix weak thinking. It multiplies it and dresses it up.

A mirror needs something to reflect

This is where most disappointing AI work actually goes wrong — not the model, the information you gave it. Starve it of context and one of two things happens, and the second is the dangerous one. Either the output dies on the spot: generic, hedged, obviously useless — and at least you can see it’s no good. Or, worse, you gave it just enough to look plausible but left gaps, and the model fills those gaps itself. It doesn’t flag the holes; it papers over them with confident invention, and that invention bleeds into the genuinely useful parts. You end up with a seamless mix of your real information and the model’s guesses, with no line showing where one ends and the other begins. The useful and the made-up arrive looking identical — which is exactly why thin information is more dangerous than none.

So the brief is the work

If the output is a reflection of the input, then the brief isn’t preamble to the work — it is the work. Give it the context, the constraints, examples of what ‘good’ looks like, and your actual standard. Just as important, tell it what you don’t know, and instruct it to ask or to flag uncertainty rather than fill it in — that single move turns silent bleed into a visible question. Mark the boundaries so it doesn’t quietly colour outside them. Thin in, confident nonsense out; considered in, your judgment back at scale.

The input side of the same coin

This is the upstream half of working with AI properly. Briefing it well is how you control what goes in; checking its work like a junior’s is how you catch what comes out. Together they bracket the tool — a clear picture going in, a human eye on what comes back. That is most of the difference between AI that embarrasses you and AI that holds.

So, what should you do?

If your AI results have been underwhelming, the fix is rarely a cleverer model — it’s what you’re putting in front of it. Working out what the mirror needs to see for the work you actually do, and how to stop it filling the gaps you didn’t, is the conversation.

It’s the first thing I check on any build: what the model is being shown, and where it’s being left to guess — before the first wrong output reaches a customer, not after.

Talk to me →