Standing instructions vs each message (system vs user)
Goal: Put “always true” rules in the right place, so you stop retyping the same preface—and keep each user message about the job.
Best for: Beginners who paste “you are a helpful assistant…” every time.
Jargon once: standing instructions (custom instructions, project instructions, sometimes “system” layer) apply across chats. The user message is today’s ask. If you typed a rule three times this week, it belongs in standing instructions.
Why beginners skip this
Mixing standing rules with one-off tasks makes constraints drift. Long-term rules live in standing instructions; task details stay in the user message. Mid-week cost shows up as rework, not as a missing bookmark.
Step 1 — Two layers
- Standing: tone, language default, ban invented links, “ask ≤3 clarifying questions”
- User message: Goal / Format / Boundaries / Input for *this* task
Step 2 — Where to click
- ChatGPT: Settings → Personalization / Custom instructions; or Project → Instructions
- Claude: account preferences and/or Project → Instructions
- Grok: product custom instructions / profile if available; else a short rules snippet at the top of important chats
Step 3 — Starter standing block (keep short)
Do this step with a real artifact open—not from memory.
Standing rules:
- Match my language (EN or ZH); do not mix in one sentence
- Prefer short answers; use bullets when listing
- If unsure, say so and list what to verify
- Never invent links or citations
- Ask up to 3 clarifying questions when my ask is ambiguous
- Do not request secrets or raw PII
Example
Filled prompt
(In Project Instructions)
Standing rules: short bullets; no invented links; ask ≤3 clarifying questions; if I attach a file, use only that file unless I say otherwise; no secrets requests.
What good output looks like
- New Project chats obey length/citation rules without reminders
- Model asks clarifying questions instead of guessing scope
- No 2-page persona novel in the user turn
- Standing block ≤8 lines and dated in your notes
- Task prompt stays short and about today’s artifact
Step 4 — User message stays about the job
Do this step with a real artifact open—not from memory.
Goal: [one sentence]
Input: [paste or attach]
Format: [bullets / table / email]
Boundaries: [only what is special this time]
If standing instructions already cover tone, do not restate tone—only the job.
Example
Filled prompt
I pasted a 600-word “act as expert” block every turn. Rewrite into: (1) ≤8-line standing instructions for a Project (length, cite sources, say Unknown, never invent metrics); (2) a short user-turn template for today’s task only. Drop the cosplay novel.
What good output looks like
- 600-word act-as block compressed into ≤8 standing lines
- User-turn template is task-only (artifact + format + never-do)
- Unknown rule stays explicit—no invented scope or silent guesses
- You can reuse the pair all week without re-pasting persona
- Checks: length obeyed, citations present, no silent guesses
Standing vs turn
Put never-do in standing instructions; keep user turns short and task-specific.
Short standing beats long cosplay
If the model must be reminded of length and Unknown every message, those rules are in the wrong place. Put them once in standing instructions. Use the user turn for the day’s artifact and constraints.
Example
Filled prompt
I still paste a 20-line preface every chat.
Move always-true rules to Project instructions. User message = only today's ask: subject lines for newsletter.
Show the split.
What good output looks like
- Labels trust/check/stop correctly for “Put “always true” rules in the right place, so you”
- Does not treat the model as a database or authority
- Context pack stays ≤ one screen when asked
- Pass/fail: you can explain the limit in one line
Red flag
Repeating a full persona every message; standing notes stuffed with one-off tasks—split long-term vs task.
Save this card
Card: Standing rules in system/Project space; task in the user turn—stop re-pasting the novel—prove it on a real artifact, or skip the badge. Keep this beside the tool; read before you start.
1. What I shipped: ___ 2. What I refused: ___ 3. What I will never-do again: ___ (date: Asia/Shanghai)
- [ ] Tools/writes paused
- [ ] Refuse-line dated
- [ ] Artifact path saved
- [ ] Next link chosen (or stop)
Before you start
Sit with three real asks from this week. You will label trust/check/stop for “Put “always true” rules in the right place,”—bring the asks, not slogans. Guide: foundations-system-vs-user-prompts. If you only have a blank practice box, stop and fetch the real job first.
- Goal in one line: ___
- Allowed tools: ___
- Forbidden: ___
- Human approve needed for: ___
How to use this week
This week: move always-true rules into system/Project instructions; leave the task in the user turn. Stop re-pasting the novel rules in the right place,” on a live artifact—not a blank demo. Write one refuse-line and date the note (Asia/Shanghai). If you only skimmed foundations-system-vs-user-prompts, put 25 minutes on the calendar and redo once.
Pitfalls
| Pitfall | Fix |
|---|---|
| Novel-length standing rules | One screen |
| Putting today’s task only in standing | Standing = always; user = now |
| Contradicting standing in every message | Fix standing once |
Example
Filled prompt
Job: Put “always true” rules in the right place, so you
Prior fail: [what missed the check]
Repair: keep the same Done; add one constraint + one concrete example from my notes.
Boundaries: draft only; label Unknowns [?]; no silent send.
Return: fixed output + the one-minute pass/fail I will mark.
What good output looks like
- Names the prior fail without inventing a new job
- Adds one constraint tied to “Put “always true” rules in the right place, so you”
- Still draft-only / Unknowns labeled
- Pass/fail: you mark green on the same checklist (foundations-system-vs-user-prompts #4)
Practice now
Live card: apply “Standing instructions vs each message (system vs user)” to a mistake from this week; write the fix habit (Asia/Shanghai).
Today: move one recurring rule into standing instructions; leave only task details in the user message. Verify once (Asia/Shanghai).
Today card (copy) — foundations-system-vs-user-prompts: 1. Artifact I will produce: ___ 2. Pass/fail check: ___ 3. Refuse-line if tempted to skip verify/send: ___ (date: Asia/Shanghai) 4. One never-do from this guide: ___
- [ ] Artifact path written
- [ ] Refuse example would say "yes"
- [ ] Dated Asia/Shanghai
One sentence takeaway
Standing rules in system/Project space; task in the user turn—stop re-pasting the novel—prove it on a real artifact, or skip the badge
- You can point to a dated refuse-line
- You have one artifact path from tonight
- You can teach the red flag in one sentence
Next
GFB · Projects workspaces · Version templates · When AI sounds sure but wrong
Pair with today: open the first Next link (prompt-goal-format-boundaries) only after you dated a refuse-line for this guide (Asia/Shanghai).
常驻说明 vs 每条消息(系统层 vs 用户层)
目标: 把「永远为真」的规则放对地方,停止每次重打同一段开场——让每条用户消息只谈当前任务。
适合: 每次都贴「你是一个有帮助的助手…」的新手。
术语一次:常驻说明(自定义说明、Project 说明,有时叫「系统」层)跨对话生效。用户消息是今天的提问。若某条规则这周打了三次,它属于常驻说明。
为什么新手跳过
每轮粘贴 600 字「扮演专家」既浪费上下文又常被忽略。常驻规则放进 Project/系统空间;任务放在用户轮。
步骤 1——两层
- 常驻:语气、默认语言、禁止编造链接、「最多问 3 个澄清问题」
- 用户消息:这次任务的目标/格式/边界/输入
步骤 2——点哪里
- ChatGPT:设置 → 个性化/自定义说明;或 Project → 说明
- Claude:账户偏好和/或 Project → 说明
- Grok:若有自定义说明/资料则用;否则在重要聊天顶部贴短规则
步骤 3——常驻块起步(保持短)
用真实材料做这一步——不要凭记忆。
常驻规则:
- 跟随我的语言(中或英);不要一句里混用
- 优先短答;列举时用要点
- 不确定就说,并列出要核实什么
- 永不编造链接或引用
- 提问含糊时最多问 3 个澄清问题
- 不索要密钥或原始 PII
实例
填好的提示词
(写在 Project 说明里)
常驻规则:短要点;不编链接;最多 3 个澄清问题;若我附文件,除非我说否则只用该文件;不索要密钥。
像样的输出大概长这样
- 新 Project 对话不用提醒也遵守长度/引用规则
- 模型会问澄清问题,而不是猜范围
- 用户轮没有两页人设小说
- 常驻块 ≤8 行,并记进你的笔记带日期
- 任务提示短,只关于今天的材料
步骤 4——用户消息只谈工作
用真实材料做这一步——不要凭记忆。
目标:[一句话]
输入:[粘贴或附件]
格式:[要点 / 表 / 邮件]
边界:[只写这次特别的]
若常驻已覆盖语气,这里不要重复语气——只写任务。
实例
填好的提示词
我每轮都贴 600 字「扮演专家」。请改成:(1)≤8 行 Project 常驻说明(长度、引用、说 Unknown、永不编指标);(2)只谈今天任务的短用户轮模板。删掉扮演小说。
像样的输出大概长这样
- 600 字扮演块压成 ≤8 行常驻说明
- 用户轮模板只含任务(材料 + 格式 + 绝不做什么)
- Unknown 规则明确——不编范围
- 这一对可整周复用,不必重贴人设
- 检查:长度遵守、有引用、无静默猜测
常驻 vs 本轮
绝不放进常驻说明;用户轮保持短且贴任务。
短常驻胜过长扮演
若每条消息都要提醒长度与 Unknown,说明规则放错地方了。写进常驻说明一次即可。用户轮留给当天的材料与约束。
红旗
每次消息重复整份人格设定;站立说明塞进一次性任务。两边打架——清理。卡上写不清就不继续——先补检查项再跑。
保存这张卡
卡面:长期→站立 / 任务→用户。开写前看卡;两边打架先清理;缺项就停,不硬开。
1. 我交出了:___ 2. 我拒绝了:___ 3. 我永不再做:___(日期:Asia/Shanghai)
- [ ] 工具/写入已暂停
- [ ] 拒做句已标注日期
- [ ] 成品路径已存
- [ ] 已选下一步(或停)
开始前
打开你真正会用的账号/项目。准备好与「常驻说明 vs 每条消息(系统层 vs 用户层)」相关的真实材料(不要空白练习框)。
默认姿态:在人确认发送/写入之前只出草稿。授予新工具前,先写一条带日期的拒做句(Asia/Shanghai)。
- 一句话目标:___
- 允许的工具:___
- 禁止:___
- 需要人批准:___
本周怎么用
本周整理:三条长期规则进站立说明,具体任务只写用户消息。把过/不过写进笔记并注明日期(Asia/Shanghai)。
常见坑
| 坑 | 改法 |
|---|---|
| 常驻规则写成长篇 | 一屏 |
| 把今天的任务只写进常驻 | 常驻=永远;用户=此刻 |
| 每条消息与常驻打架 | 改一次常驻 |
实例
填好的提示词
任务:Put “always true” rules in the right place, so you
上次失败:[哪条检查未过]
修复:保持同一 Done;加一条约束 + 一条你笔记里的具体例子。
边界:仅草稿;未知标 [?];禁止悄悄发送。
返回:修好的输出 + 我将打勾的一分钟过/不过。
像样的输出大概长这样
- 点名上次失败,不另起炉灶
- 约束贴着「常驻说明 vs 每条消息(系统层 vs 用户层)」
- 仍是草稿 / 未知已标注
- 过/不过:同一张检查表能打绿(foundations-system-vs-user-prompts #4)
实例
填好的提示词
任务:Put “always true” rules in the
场景要点:三张真实请求便签在手;练习编号 4/4。
边界:仅草稿;未知 [?];禁止悄悄发送;禁止编造指标。
返回:可检查产出 + 过/不过理由一行。
像样的输出大概长这样
- 约束贴着「常驻说明 vs 每条消息(系统层 vs 用户层)」与 foundations-system-vs-user-prompts
- 未知标 [?];不编造负责人/数字
- 不可逆动作前停下
- 一分钟内能打过/不过(foundations-system-vs-user-prompts #4)
现在就练一次
实卡:把「常驻说明 vs 每条消息(系统层 vs 用户层)」套到本周一次失误;写纠正习惯(Asia/Shanghai)。
今天把一条总被遗忘的规则迁到站立说明并验证。过/不过与永不做注明日期(Asia/Shanghai)。只收藏标题算失败。
今日卡片(可复制)— foundations-system-vs-user-prompts: 1. 我要产出的成品:___ 2. 通过/失败标准:___ 3. 若想跳过核验/发送的拒做句:___(日期:Asia/Shanghai) 4. 本篇一条永不做:___
- [ ] 成品路径已写
- [ ] 拒做例句会说「可以」
- [ ] 已标注 Asia/Shanghai 日期
一句话带走
永远成立的规则放进系统或 Project;具体任务留在用户轮——别再每条消息重贴一篇说明小说。本周把常驻规则迁走一次。本周留下可核对的过/不过记录并注明日期(Asia/Shanghai)
- 能指出一条带日期的拒做句
- 今晚有一条成品路径
- 能用一句话说清红旗
下一步
目标·格式·边界 · Projects 工作区 · 版本化模板
今日搭配:先为本篇写上拒做句日期(Asia/Shanghai),再打开下一步里的第一篇(prompt-goal-format-boundaries)。
← Back to home · Learn · Tools · Compare · Start