The five details most prompts leave out

After scoring a lot of prompts, the same gaps show up again and again. They are not exotic. They are the things you know so well that it does not occur to you to write them down.

1. Who the output is for

The single most common omission. "Explain our pricing change" produces completely different text for existing customers, for the sales team, and for the board. Without an audience the model writes for a generic reader, which is why so much AI output has that weightless, could-be-anyone quality.

Better: "for existing customers on the old plan, who will see their bill go up".

2. What the reader already knows

Audience alone is not enough. A developer audience that has never used your API needs a different piece from one migrating off a competitor. Stating the starting point prevents both the patronizing explanation and the one that assumes too much.

3. What to do when something is unknown

This is the detail that separates output you can trust from output you have to fact-check line by line. Models fill gaps confidently unless told not to. One sentence changes that: "If a detail is not in the material I gave you, write [unknown] rather than guessing."

4. The shape of the answer

Table, bullets, JSON, five sections, under 200 words. Prompts that leave the shape open get whatever shape the model defaults to, which is usually three paragraphs of prose. If you are going to paste the result somewhere specific, say what it needs to look like.

This matters even more when the output feeds something else — a spreadsheet, a script, a document template. "Return only a JSON array with keys name, reason, priority" is a short sentence that saves a lot of cleanup.

5. What good looks like

An example. One piece of your own writing, one row of the table you want, one sentence in the voice you use. Describing a voice — "friendly but professional" — is far weaker than showing two sentences of it. This is the most underused move in everyday prompting, and it is nearly free.

It works in the negative too. One line of "do not write like this" removes a whole category of output you would otherwise reject one draft at a time. If every AI draft opens with "In today's fast-paced world", the fix is not a better adjective — it is a sentence naming that opening as forbidden.

Why these five and not fifty

There are longer checklists. Most of their items are variations on these five, or techniques for specific situations: chain-of-thought for reasoning problems, few-shot examples for classification, personas for tone. Useful occasionally. But for ordinary work — an email, a summary, a plan, a review — the gap between a bad answer and a good one is almost always one of the five above.

You can test that cheaply. Take a prompt that disappointed you last week and add only the missing items, changing nothing else. In our own scoring that edit typically moves a one-line prompt from the twenties into the seventies, and the difference in the answer is usually obvious on the first try.

Write them down once

If you find yourself typing the same context into chat every time — who the audience is, what tone you use, what to never invent — that context belongs in a saved prompt rather than in your memory. A prompt that carries its own context gives the same result next month, and gives a colleague the same result it gives you.

Putting it together

None of these are techniques. They are just the information you already have, written down. A prompt that answers all five is usually the difference between a first draft you edit and a first draft you delete.

1. Who is this for?
2. What do they already know?
3. What should happen when something is unknown?
4. What shape should the answer take?
5. What does a good version look like?

If answering those questions every time sounds tedious, that is exactly what Smart Clarify does: it checks which of them your prompt is missing and asks two to four questions before optimizing — and skips the questions entirely when the prompt already covers them.