How to Adapt ChatGPT Custom Instructions for Claude
Summary
- Start by extracting the intent of your ChatGPT Custom Instructions (role, audience, tone, constraints, and “do/don’t” rules) rather than copying them verbatim into Claude.
- Rewrite instructions as a short “working agreement” Claude can follow: priorities, format requirements, and a clear escalation path when info is missing.
- Convert ChatGPT-specific references (Custom Instructions, Memory, Projects, Custom GPTs) into model-agnostic behaviors you can paste into any Claude chat or project-like workspace.
- Use a small translation checklist and a test prompt to verify Claude follows your style, structure, and safety constraints before real work.
- Store your instruction sets as reusable snippets so you can quickly reuse them across Claude, ChatGPT, Gemini, docs, and tickets without rewriting each time.
If you built a strong set of ChatGPT Custom Instructions, you already did the hard part: you defined how you want an AI to behave. The challenge is that Claude does not use ChatGPT’s “Custom Instructions” feature in the same way, and even when both models understand the same words, they can interpret priorities and formatting differently.
This guide shows how to adapt your ChatGPT Custom Instructions for Claude in a practical, repeatable way: extract what matters, rewrite it in Claude-friendly language, and package it so you can reuse it across tools and teams.
What you are really “porting” from ChatGPT to Claude
ChatGPT Custom Instructions usually contain a mix of:
- Identity and role: “Act as an SEO strategist…”
- Audience and context: “I work with B2B SaaS clients; assume…”
- Style and tone: “Concise, direct, no fluff…”
- Output formats: “Use tables, bullets, include examples…”
- Constraints: “Don’t invent sources; ask clarifying questions…”
- Workflow preferences: “Start with a summary, then steps…”
Claude can follow all of these, but you will get better results if you rewrite them as a priority-ordered instruction block that is easy to paste into a new Claude conversation (or whatever persistent workspace you use in Claude).
A simple 4-step method to adapt Custom Instructions for Claude
Step 1: Split your instructions into “always” vs “task-specific”
ChatGPT Custom Instructions often become a dumping ground for everything. Claude performs better when your “always-on” rules are short and stable, and task-specific details are provided per request.
- Always-on: tone, formatting defaults, how to handle uncertainty, citation rules, how to ask questions.
- Task-specific: the client, the product, the campaign, the dataset, the current quarter goals, the exact deliverable.
Practical tip: If a line would be wrong for your next project/client, it is probably task-specific and should not live in the base instruction set.
Step 2: Remove ChatGPT-native feature references and rewrite as behaviors
Some instruction lines assume ChatGPT features or UI concepts. Instead of naming features, describe the behavior you want.
- Instead of: “Use my Custom Instructions and Memory.”
- Write: “Use the context I provide in this chat. If something is missing, ask questions before assuming.”
- Instead of: “Follow the project brief in my Project.”
- Write: “Treat the following brief as the source of truth. If there is a conflict, call it out and propose a resolution.”
This makes your instruction set portable across Claude, ChatGPT, Gemini, and even non-AI contexts (like a support macro or a content brief template).
Step 3: Add priority and conflict-resolution rules
When instructions get long, models can “average” them. Claude responds well when you explicitly rank priorities and define what to do when rules conflict.
Example priority block you can add:
- Priority 1: Follow user constraints (no invented sources, comply with formatting requirements).
- Priority 2: Be accurate; if unsure, ask clarifying questions or label assumptions.
- Priority 3: Be useful and actionable (steps, examples, checklists).
- Priority 4: Keep it concise unless I ask for depth.
Step 4: Create a “verification prompt” to test the port
Before you rely on the adapted instructions, run a short test prompt that forces the model to demonstrate compliance.
Verification prompt example (paste after your instruction block):
- “Summarize my instruction priorities in 5 bullets.”
- “Now draft a 150-word answer to: ‘What is a positioning statement?’ Use my formatting rules.”
- “List any missing context you would need to tailor this for a B2B SaaS client.”
If Claude misses a rule (tone, structure, or “don’t invent”), tighten the wording and re-test.
Translation patterns: ChatGPT Custom Instructions to Claude-ready instructions
Use these patterns to rewrite common instruction types so they are clearer and more portable.
| Instruction type | ChatGPT-style wording (example) | Claude-ready rewrite (example) |
|---|---|---|
| Role | “You are my senior SEO editor.” | “Act as a senior SEO editor. Optimize for clarity, search intent, and practical steps.” |
| Audience | “Assume I’m a marketer.” | “Write for marketers and cross-functional teams. Avoid jargon; define terms briefly when needed.” |
| Tone | “Be friendly and concise.” | “Use direct, professional language. Keep paragraphs short. Avoid filler.” |
| Formatting | “Use bullet points and tables.” | “Default to: short intro, then headings, bullets, and a compact table when it helps decisions.” |
| Uncertainty | “Don’t hallucinate.” | “If you are not sure, say what is unknown, ask up to 3 clarifying questions, and offer safe assumptions labeled as assumptions.” |
| Constraints | “No made-up stats or citations.” | “Do not invent sources, quotes, or numbers. If a claim needs a source, ask me to provide one or omit it.” |
| Process | “Think step by step.” | “Show your work as a checklist of steps and decision points (no hidden reasoning required).” |
Ready-to-paste templates (adapt for your role)
Below are model-agnostic instruction blocks you can paste into Claude. Keep them short, then add task context per request.
Template A: Consultant / strategist
- Role: Act as a consultant. Provide options, tradeoffs, and a recommended path.
- Output: Start with a brief summary, then steps, then risks/assumptions.
- Questions: If key info is missing, ask up to 3 clarifying questions before finalizing.
- Accuracy: Do not invent sources or numbers. Label assumptions clearly.
- Tone: Direct and practical. Avoid fluff.
Template B: Marketing / content team
- Role: Act as a content editor and strategist.
- Style: Clear, skimmable, and specific. Use headings and bullets.
- SEO: Match search intent; include examples and decision criteria when relevant.
- Constraints: No invented citations, quotes, or stats.
- Deliverables: Provide 2-3 variants when it helps (e.g., headline options).
Template C: Recruiter / talent
- Role: Act as a recruiter. Optimize for clarity, inclusivity, and role fit.
- Output: Provide structured outputs (requirements, nice-to-haves, screening questions).
- Constraints: Avoid sensitive inferences. Ask clarifying questions if the role scope is unclear.
- Tone: Professional and human.
Template D: Support / success
- Role: Act as a support specialist.
- Output: Start with the likely fix, then step-by-step instructions, then “If this fails…” escalation.
- Constraints: If you do not know the product detail, ask for version/environment rather than guessing.
- Tone: Calm, empathetic, and concise.
How to keep one instruction set consistent across ChatGPT, Claude, and Gemini
If you use multiple models, the biggest time sink is rewriting the same “how to work with me” block in different places. A practical approach is to maintain:
- A base instruction block (stable, short, portable)
- Role-specific add-ons (consulting, SEO, recruiting, support)
- Client/project briefs (task-specific, swapped in and out)
Then, for each new conversation, you paste:
- Base block
- Role add-on (if needed)
- Task brief + deliverable request
This structure also makes it easier to update your “rules” once and reuse them everywhere.
Where CopyCharm fits: saving, finding, and reusing your instruction blocks
If you are adapting ChatGPT Custom Instructions for Claude, you are really building reusable context. CopyCharm can help you keep that context close at hand while you work across chats, docs, and tickets.
A concrete workflow (save -> find -> reuse)
- Save: Store your base instruction block and each role-specific add-on as Saved Prompts in CopyCharm (separate from favoriting copied text clips).
- Find: When you need them, search in CopyCharm to pull up the exact instruction set (for example, “support escalation” or “SEO formatting rules”).
- Reuse: Copy/paste the saved prompt into Claude (or Gemini, email, docs, a ticketing system) to start a new conversation with consistent expectations.
If you also use ChatGPT, CopyCharm offers an authenticated ChatGPT connector backed by optional AI Access sync. After you sign in with an eligible active CopyCharm purchase, authorize the CopyCharm Desktop connection, enable and complete AI Access sync, and authorize the connector, ChatGPT can search and retrieve only supported Synced Data (Favorite Clips, Saved Prompts, and optional Other Clips within the time range you select). ChatGPT cannot access unsynced local CopyCharm data, and connector retrieval is user-directed.
For Claude, Gemini, Cursor, and other apps, the workflow remains manual: you search or retrieve the content in CopyCharm and then copy/paste it into the destination.
Common adaptation mistakes (and how to fix them)
Mistake 1: One giant instruction wall
Fix: Keep the base block short. Move client details into a separate brief you paste only when needed.
Mistake 2: Vague rules like “be helpful”
Fix: Replace with observable behaviors: “Ask up to 3 clarifying questions,” “Provide a checklist,” “Include 2 examples.”
Mistake 3: Conflicting constraints
Example: “Be concise” + “Be comprehensive” + “Ask questions first.”
Fix: Add priority rules and a default: “Be concise by default; go deep only when asked.”
Mistake 4: Hidden assumptions about tools and access
Fix: State boundaries: “Use only the info in this chat; do not claim you checked external systems unless I provide the data.”
Frequently Asked Questions
FAQ 1: Can I copy my ChatGPT Custom Instructions into Claude verbatim?
Answer: You can paste them, but you will usually get better results by rewriting them as a short, priority-ordered “working agreement” (role, output format, constraints, and how to handle missing info). Remove references to ChatGPT-only UI concepts and describe the behavior you want instead.
Takeaway: Port the intent, not the exact wording.
FAQ 2: What is the fastest way to convert ChatGPT instructions into Claude-friendly instructions?
Answer: Use a 4-step pass: (1) split “always-on” vs task-specific, (2) rewrite tool-specific lines into model-agnostic behaviors, (3) add priority rules (what matters most), and (4) run a short verification prompt to confirm tone and formatting are followed.
Takeaway: A quick rewrite plus a test prompt beats trial-and-error on real work.
FAQ 3: How do I handle ChatGPT-specific features like Memory, Projects, or Custom GPTs when moving to Claude?
Answer: Replace feature names with explicit instructions about what the model should treat as authoritative (your pasted brief), what it should do when context is missing (ask questions), and what it must not do (invent sources or claim access to external systems). This keeps your instructions usable even when platform features differ or change.
Takeaway: Convert “use feature X” into “follow this context and these rules.”
FAQ 4: How long should my Claude instruction block be?
Answer: Aim for the shortest block that still enforces your non-negotiables: role, default structure, constraints (like no invented sources), and how to handle uncertainty. Put client/project details in a separate brief you paste only when needed.
Takeaway: Keep the base block short; swap in task briefs separately.
FAQ 5: How can teams keep consistent instructions across consultants, marketers, recruiters, and support?
Answer: Maintain one shared base instruction block (tone, formatting defaults, uncertainty rules), then create role add-ons (SEO editor, recruiter, support specialist) and project briefs (client-specific). Team members paste the base + the relevant add-on + the brief into each new conversation.
Takeaway: Standardize the base, modularize the rest.
FAQ 6: What should I do if Claude ignores my formatting or tone rules?
Answer: Make the rule measurable and put it near the top: “Output must be: Summary (5 bullets) + Steps + Risks.” Then run a verification prompt and tighten wording based on what fails. If needed, restate the format in the specific request as well (not only in the base block).
Takeaway: Make rules explicit, test them, and restate format at request time.
FAQ 7: How do I adapt instructions for SEO and content workflows without inventing sources?
Answer: Add a hard constraint: “Do not invent citations, quotes, or statistics.” Then add a fallback behavior: “If a claim needs a source, ask me to provide one or omit the claim.” For SEO outputs, specify what you want instead of stats (search intent match, examples, decision criteria, and clear structure).
Takeaway: Replace “sound authoritative” with “be verifiable and practical.”
FAQ 8: Can CopyCharm help me reuse the same instruction blocks in Claude and ChatGPT?
Answer: Yes. You can save your base instructions and role add-ons as Saved Prompts in CopyCharm, search them when needed, and copy/paste them into Claude. If you use ChatGPT, CopyCharm also has an authenticated ChatGPT connector: after eligible authorization and AI Access sync, ChatGPT can search and retrieve supported Synced Data (and cannot access unsynced local CopyCharm data).
Takeaway: Store instruction blocks once, then reuse them across tools with clear boundaries.