You typed 200 words. The read found the 12 that matter.

You wrote the whole thing out. Four paragraphs. Every angle you could think of, because you wanted to be fair, because you wanted the answer to be right. And somewhere in there you're bracing for the AI to just grab the loudest part and hand it back to you dressed up as wisdom.

That fear is correct about most tools. It is not correct about what a read is for.

The dilemma (a composite)

This is a composite of the kind we see in this category — stitched from many, belonging to no one.

Rahul manages a six-person data team. A senior role opened up. On the inside: Nisha, four years with him, the one who stayed late through the migration nobody else understood, reliable to the point of invisible. On the outside: a candidate from a competitor who walked into the loop and made the whole panel sit up.

Rahul's message ran about two hundred words. It covered a lot. The outside candidate has broader exposure. The team might read an internal promotion as the default and get complacent. Nisha is strong but hasn't led a project end to end. The market is competitive and a flashy hire signals ambition to leadership. He owes it to the team to pick the best person, not the most familiar one. He doesn't want to be sentimental. He's thought about this carefully. He really has.

Then, near the end, almost as an aside, one clause: "and honestly the interview with the outside candidate just went so much better, Nisha's was fine but a little flat."

He kept writing after that. Two more sentences about team morale. The aside got buried.

The read

Here is the thing length hides: most of those two hundred words are justification, and justification reads differently from fact. A read doesn't reward volume. It sorts the sentences by what they're doing — carrying a fact, or carrying a feeling that needs a fact-shaped costume to get past you.

Watch which words survive.

Discernment — the lens that separates what's doing work from what's doing decoration (one of the eight). It sees that "broader exposure" and "ambition to leadership" and "the team might get complacent" are all defensible, all true-ish, and all arriving after the conclusion. They branch. One sentence does not branch: the interview went better. Everything else is built to support a verdict the interview already delivered. Discernment pulls hard toward naming that clause as the hinge — and noticing it was written to sound incidental.

Duty — what the role actually requires of the person filling it. It asks a flat question: does the senior role reward interviewing well, or doing the work? If the job is leading migrations under pressure, Nisha has four years of evidence and the outsider has ninety minutes of performance. Duty sees that "hasn't led a project end to end" is the one real factual gap in the whole message — and that it's a developable gap, not a verdict. Duty pulls toward the person with the track record, unless the track record is genuinely absent.

Now the tension, which is the whole point. Honesty enters and disagrees with how Rahul framed his own case. Honesty isn't about whether he lied to the panel — it's about the sentence doing emotional work while wearing a factual face. "Her interview was a little flat" is a feeling about a ninety-minute window. Rahul presented it as data. Honesty and discernment together land on the uncomfortable read: the two hundred words are a brief written in defense of being impressed by a stranger. The embarrassment — I don't want to admit the outsider dazzled me and I'm reverse-engineering reasons — is the hinge he spent 180 words talking around.

So which way does the read go? Not "promote Nisha, loyalty wins." That would just be swapping one sentiment for another. The read goes to the missing fact: strip the interview impression out entirely, because an interview tests interviewing, and ask what remains. What remains is a four-year record versus a résumé.

The verdict: this decision isn't ready — not because it's hard, but because it's currently resting on a ninety-minute impression dressed as strategy. Name the one real gap (Nisha hasn't led end to end), test it directly with a scoped project or a structured work sample for both candidates, and let the evidence decide what the interview was never qualified to.

Why this isn't a chatbot agreeing with you

A general AI chatbot weights whatever you emphasize. Write two hundred words leaning toward the exciting outsider and it will help you hire the outsider, fluently, with your own justifications polished and handed back. Lead with loyalty instead and it pivots without friction — same tool, opposite answer, because agreeing with your emphasis is a well-known tendency in these systems, not a bug you can prompt away. A read does the opposite thing on purpose: it treats your word count as noise and your buried clause as signal, isolates the one sentence doing emotional work, and makes a single committed call that won't renegotiate itself when you push back. The mechanics aren't magic and aren't unbiased — they're a discipline: ask for the fact that would change the answer first, then decide once.

Do this tonight — the two-pass test

Free, no account, works on anything you've over-written.

1. First pass — underline the facts. Read your own message and underline only sentences that would be true whether or not you wanted them to be. "She has four years here." "The role leads migrations." Those are facts. "The team might get complacent" is a prediction wearing a fact's clothes — leave it un-underlined.
2. Second pass — circle the one that's warm. Find the single sentence that carries heat: embarrassment, excitement, the sting of being impressed, the wish not to look sentimental. There's almost always exactly one, and it's almost always short, and it's almost always near the end where you hoped it would pass unnoticed.
3. Then ask the swap question: "If I deleted the warm sentence, would my conclusion still stand on the underlined ones alone?" If yes, you have a decision. If no, the warm sentence is the decision, and the rest was the alibi.

Length is not signal. You can write two hundred careful words and have the whole thing hinge on twelve.

Rahul didn't write a bad message. He wrote an honest one that hid its own honesty under justification. The fix was never to write less — it was to find the sentence he'd buried and read it out loud. You can bring your own long, winding one and watch which words survive. See worked reads if you want the shape first, or more on how the read sorts signal from volume.

So: in the dilemma you've been drafting, which sentence did you put last because you hoped it would slip by?