It Cannot Infer What You Did Not Write

The new hire test

Imagine handing a task to a capable person who started this morning.

They are intelligent and quick. They know nothing about your company, your audience, your history with this client, or what "the usual format" means.

If you said "make this better", they would ask five questions before starting.

The model does not ask. It guesses, plausibly and invisibly, and hands you something built on assumptions you never saw.

The five things

Almost every disappointing output traces to one of these being unstated.

Goal. What is this for? Persuade, inform, record, decide?

Audience. Who reads it, and what do they already know?

Format. Email, bullets, table, memo, script?

Length. "Short" means nothing. "150 words" means something.

Constraints. What must be included, avoided, or kept from a previous version?

You were holding all five in your head. That is why the request felt complete when you typed it.

Before and after

Summarise this report.

Against:

Summarise this report in 150 words for our finance director, who has not read the original. Focus on cost implications. Three bullets, no jargon, and flag anything requiring a decision this quarter.

The second is not more elaborate for the sake of it. Every clause removes a guess the model would otherwise have made silently.

The why does the most work

Of the five, the highest-value addition is the sixth thing: the situation.

This is going to a client who is already unhappy about the delay.

You have not specified tone, structure, or which points to lead with. But now the model can work all of that out, because it understands what the writing is for.

This is the difference between a specification and a brief. A specification lists requirements. A brief conveys purpose, and purpose lets someone make good decisions about the parts you forgot to mention.

Longer is not the point

A caution, because people over-correct.

Adding words is not the goal. Removing ambiguity is. A 30-word prompt that names audience and format beats a 200-word one that rambles about background.

The test is not length. It is: could a new colleague produce what I want from this?