[Essential Guide] The Secrets of Instruction Symbols in Generative AI
Clearly structured prompts make it easier for AI to understand your intent and obtain consistent output. Therefore, I have summarized the specification symbols that generative AI, especially ChatGPT, can easily understand, along with how to use them.
Even with prompts in Japanese, ** and - can be used as they are (because they are Markdown compatible). However, in ChatGPT, the clarity of instructions and the consistency of context are more important than emphasis.
General Prompt Specification Rules
Emphasis Expressions
**text**
: Emphasis (sometimes recognized as bold, semantic emphasis)
*bullet point item* → Sometimes treated as part of a list.
Lists and Hierarchical Structures
-: Bullet points, enumeration of elements
*: Bullet points, enumeration of elements part 2
・: Frequently used for bullet points in Japanese
# : Title (main heading)
## : Subheading
### : Minor subheading
→ The more symbols there are, the lower the heading level.
Weighting Specification (Image Generation Systems)
(word:1.2): Assign a weight (priority) to a specific word (usually 1 is the standard)
Exclusion Specification
--no: Exclusion in Midjourney-style prompts
Example: --no text means "do not include text"
Composition Description (Image Composition)
left, right, center, background, foreground: Positional relationships
in the style of: Specification of artistic style
Example: in the style of Studio Ghibli
Language Structure Rules (for ChatGPT)
Q: and A:: Explicit question-and-answer format
[Command content]: Specification of command statements or output formats, e.g., [Instruction] or [Output Format]
##: Used as section headings (Markdown-style notation)
`code`: Inline display of code
"""comment (newlines allowed)""": String literal in Python code, comment
※Note
Lines starting with # are ignored as comments.
Strings enclosed in """triple quotes (triple-quoted string) are ignored (unless assigned to a variable)
# 今日のおやつリスト
- りんご
- ばなな
- *チョコレート*
# 注意点
- たべすぎないこと！
- *アレルギーがある人は注意！*
"""
バナナはおやつに入らないかという
ありふれた論争に
ここではお弁当箱に入っていれば
おやつに入らないのとする
"""ChatGPT and Markdown-compatible tools treat "headings," "lists," and "emphasis" as having meaning, and the AI also reads the structure to improve output accuracy.
Variables and Placeholders (Temporary Input Values)
{}: Specification for variables or parts to be changed later
Usage example:
Please replace {name} with your own.
→ {name} is the part intended to be replaced with a name later.
Question: {User's question}
→ A marker for where the user should input something.
How it is used in ChatGPT prompts:
When providing a template to a user:
以下の形式で入力してください：
{ジャンル}の{テーマ}について、{文体}で書いてください。Options (optional), lists/arrays, or explanations/notes
[] : Represents options (optional), lists/arrays, or explanations/notes
Usage example:
[Optional] Enter your nickname.
→ You may or may not enter it.
[Instruction] Use casual tone.
→ Used for commands or explanatory notes.
[1, 2, 3]
→ Used as list (array) syntax.
How it is used in ChatGPT prompts:
Used to explicitly create instruction blocks:
[命令] 以下の文章をですます調に変えてくださいSpecifying output format:
[Output format] Q&A形式で答えてくださいConditional expressions:
[女性向けの場合のみ] やわらかい言葉づかいにしてくださいTag-like expressions or systematic markup symbols
<> : Used as tag-like expressions or systematic markup symbols.
Usage example:
<name> or <topic>: HTML-like tags or placeholders.
<system>: Sometimes represents meta-instructions for the AI within a prompt.
How it is used in ChatGPT prompts:
Explicitly defining specific sections or meta-structures (often seen in Advanced Prompting):
<system>
あなたは厳格な編集者です。以下の文を校正してください。
</system>As delimiters for people, sections, or structures:
<user>: こんにちは
<assistant>: こんにちは。今日は何をお手伝いしましょうか？Framework for input formats:
Please fill in: <Title>, <Theme>, <Style>In fact, ChatGPT does not understand these symbols as meanings themselves, but rather infers their intent from context and recurring usage patterns.
For example:
If {} is frequently used as template syntax, it will interpret it accordingly.
If [] is frequently used in contexts indicating notes or commands, it will treat them as commands.
If <...> appears as a mark for structure or roles, the AI will treat it as a delimiter.
In short, please recognize them as symbols that can be given meaning depending on how they are used. Furthermore, using them in generally accepted ways makes them easier for both AI and humans to understand.