Prompts & tools · 10 min read
Prompt engineering basics on AI Plaza
Practical prompt patterns for AI Plaza: roles, constraints, language rules, Auto routing, and how scenario tools encode expert prompts.
· · AI Plaza
Quick steps
Step 1. State the role and deliverable
Tell the model who it is and what artifact to produce (email, table, script).
Step 2. Add constraints
Length, tone, language, must-include facts, and must-avoid claims.
Step 3. Provide raw material
Paste notes, diffs, product facts, or upload files on AI Plaza Chat.
Step 4. Iterate with a tight ask
Ask for a revision of one section instead of regenerating everything.
Prompt engineering basics on AI Plaza
Clear roles, constraints, and raw materials beat clever wording—whether you use General chat or a scenario tool.
The four-part prompt
- Role — “You are a UX writer…”
- Task — “Write three push notification variants…”
- Constraints — length, tone, language, format, banned phrases
- Inputs — paste facts, links, diffs, or upload files
Scenario tools already supply (1)–(3). You focus on (4).
Cross-language work
UI locale and deliverable language are independent:
- Chat in Traditional Chinese → ask for a Japanese email body.
- Paste Chinglish → use Native English Polish.
- Localize marketing with Marketing Localization for Taiwan, Japan, US, etc.
Always name the target language in one short line.
Patterns that work well
Rewrite / polish
Keep meaning. Improve clarity and tone for senior executives. Output the polished version first, then 3 bullet notes on changes.
Structured extraction
From the attached PDF, extract: parties, dates, fees, termination clauses. Use a markdown table. Flag unknowns as “not found”.
Diff-aware coding
Paste this git diff and write a Conventional Commit subject + body. Offer one alternative subject.
Creator hooks
Give 10 YouTube titles and 3 thumbnail text overlays. No false promises. Front-load keywords.
Using Auto vs locking a model
Auto (Smart) inspects length, keywords, and tool context to pick a model that is good enough and cheaper when possible. Chinese-heavy tasks may prefer Qwen / DeepSeek when available.
Lock a model when you are A/B testing quality, need a named model for compliance, or Enterprise routing caps expensive classes.
Stretch your usage %
- Draft on economy / Auto, polish once on a stronger model
- Prefer tools over reinventing long system prompts
- Avoid pasting huge irrelevant context
Apply it
Open Chat with Auto on—or start from Tools. Next: How to use prompt tools · Models, usage, and Auto.
Frequently asked questions
- Should I write prompts in English on AI Plaza?
- No. AI Plaza tools are language-agnostic. You can chat in Chinese and request an English deliverable (or the reverse). State the target language explicitly when it matters.
- When should I override Auto?
- Override Auto when you need a specific model’s style, a flagship for deep reasoning, or a cheaper lock for bulk drafting. Otherwise Auto on AI Plaza is the best default.