Write for someone
Say who will read the output and what they will do with it, so the level, emphasis and tone follow.
このレッスンはまだお使いの言語に対応していないため、別の言語（English）で表示しています。
Why it matters
The same facts make a different text for every reader. A summary of quarterly results for the board, for new staff and for a local newspaper shares almost nothing but the numbers. When the prompt is silent about the reader, the model writes for an average person who does not exist, and the result is pitched too high for some and too low for others.
Two questions do most of the work: who is this for, and what will they do with it. The first sets the level and the vocabulary; the second sets the emphasis. A resident sorting rubbish needs “which bin”, not “why the policy changed”. Anthropic's guidance makes a related point: explaining the purpose behind a request helps the model give a more targeted response.
Who the output serves is one of the requirements the ROPE researchers taught people to state. It is part of what you need, not a matter of style.
What good looks like
- 0 — No reader or purpose can be inferred. “Explain photosynthesis.”
- 1 — Reader implied, never stated. “Explain photosynthesis simply.” Simple for whom?
- 2 — Reader named, use unstated. “Explain photosynthesis to a 12-year-old.”
- 3 — Reader and purpose both stated. “…to a 12-year-old preparing for a science quiz next week.”
- 4 — Reader, purpose and what they already know. “…who has covered cells but not chemistry; she must answer short written questions, so favour precise terms over analogies.” Level, emphasis and register now follow from the description.
Some tasks have no human reader at all, such as a pure code transform or a data extraction. There the criterion is marked not applicable.
Common mistakes
- Using a register word (“simply”, “professionally”) instead of describing the reader.
- Naming the reader but not what they will do with the output.
- Forgetting what the reader already knows, so the model explains too much or too little.
- Writing one text for two audiences with different needs instead of asking for two sections.
- Describing yourself when someone else will read the result.
Try it on your own prompt
Paste a prompt into the workspace and add one sentence: “This is for …, who will use it to ….” If you can, add what that reader already knows. Then see how the audience score changes, and read the output to check whether the level now fits the person you had in mind.
解説付きの例
Recycling leaflet
Explain the new recycling rules.
<rules>
{{paste the council's new recycling rules here}}
</rules>
Explain these rules for a leaflet going to residents of a sheltered-housing block. Most readers are over 75, several have poor eyesight, and none attended the consultation meetings.
They will use the leaflet on collection day to decide which bin each item goes in, so they need to know what goes where, not why the rules changed. Use short sentences and everyday items as examples (a yoghurt pot, a pizza box, a wine bottle) rather than the policy terms in the rules.
The strong version names the reader, what they already know and what they will do with the leaflet, so the model can choose the level, the emphasis and the examples.
Release notes
Write release notes for version 3.2.
<changes>
{{paste the list of changes in version 3.2 here}}
</changes>
Write release notes for version 3.2 of our accounting app for two groups of readers.
1. Small-business owners who use the app weekly. They know the invoicing screen well but not technical terms, and will skim the notes to see whether anything changes their routine. Start with what they will notice, in plain words.
2. Our support team, who know the product in depth and will use the notes to answer customer questions this week. Give them a short section with the technical detail and any known issues.
One set of notes rarely suits everyone. Naming both readers, what each knows and what each will do lets the model write two sections at the right level instead of one muddled one.
演習
Rewrite this prompt so the model knows who is learning, what they already know and what they will do next.
Rewrite this prompt for a specific reader with a specific use for the summary.
Rewrite this prompt so the model knows who will read the review and which decision it feeds.