Reusable Context for ChatGPT and Claude With Three Files

Updated 10 November 2026 · Standard formula, shown below

A three-file context system gives ChatGPT and Claude a compact, reusable reference for your goals, preferences, and ongoing work. Create one file for stable details about you, one for how you want the assistant to respond, and one for the current project; then provide those files in each tool’s available project or conversation instructions.

What You’ll Need

A text editor or word processor that can save plain-text files.
A ChatGPT or Claude account, and permission to create a project, add reference files, or attach files where you plan to use the context.
A few accurate notes about yourself, your preferred responses, and the project you want help with. Do not include passwords, secret keys, or sensitive information you would not want processed.

Step-by-Step

Choose a project or recurring use case. Start with one kind of work, such as drafting reports, studying a subject, or planning a product. Reusable context works best when it describes a clear purpose rather than trying to cover every possible conversation. Decide what the assistant should remember and what you can explain in a prompt when needed.
Create the first file: profile.md. Record stable facts that change your answers, such as your role, audience, region, relevant experience, and long-term goals. Keep each fact short and useful. For example: “I manage communications for a small nonprofit” or “My readers are new to personal finance.” Avoid a life story; add a detail only if it helps the assistant produce better work.
Create the second file: preferences.md. Describe how you want responses written and structured. Be specific: “Use plain language,” “Put the recommendation first,” “Use headings for long answers,” or “Flag assumptions instead of presenting them as facts.” Include preferences that apply consistently. Do not put instructions here that are relevant only to one assignment.
Create the third file: project.md. Explain the current project’s purpose, audience, status, constraints, terminology, and next steps. Include relevant facts the assistant should not have to rediscover. Add dates or a “Last updated” line for information that may change. Keep this file separate from your profile so you can replace it when you move to another project.
Make the files easy to scan. Use clear headings and short bullet points. State important requirements directly: “Do not invent statistics; ask for a source if one is missing.” Resolve contradictions before uploading. Prefer a few high-value facts over long background notes, and make clear whether a statement is a firm rule, a preference, or a working assumption.
Save and review the files. Save them as profile.md, preferences.md, and project.md. Markdown is convenient because headings and lists stay readable, but plain text is also suitable. Read each file once from the assistant’s perspective: Is the meaning clear without extra explanation? Remove stale details, duplicated instructions, and anything private or unnecessary.
Add the files to ChatGPT. Create or open a project if that feature is available, then look for its instructions or reference-file area and add the three files. Interface labels and available features can change, so use the current controls shown in your account. If you are not using a project, attach the files to a conversation when file attachments are available, and tell ChatGPT to use them as reference.
Add the files to Claude. Create or open a project if available, and add the files in its project knowledge or instructions area. If you are working outside a project, attach the files to the conversation when supported. Do not assume that a file added in one tool is available in the other; upload the files separately to each service.
Give a short instruction with your task. Context files guide the assistant, but they do not replace the request. Try: “Use the attached profile, preferences, and project notes. Draft a one-page update for the project audience. Ask me if a required fact is missing.” Name the files if useful, and specify the deliverable, audience, and any task-specific constraints.
Test and refine. Ask for a small, representative task. Check whether the result follows your preferences and uses project facts accurately. If it misses something, determine whether the file is unclear, unavailable, or in conflict with your prompt. Correct the source file rather than repeatedly patching the same issue in individual chats.
Maintain the project file. Update project status, decisions, and next steps when they change. Keep a local copy as the source of truth, then replace or update the uploaded copy in each service. Start a fresh project file for unrelated work instead of letting old assumptions carry over.

Common Mistakes

Putting everything in one oversized document: Long, mixed notes are harder to maintain and may obscure the important details. The three files have distinct jobs: profile for stable background, preferences for response style, and project for current work.

Writing vague preferences: “Make it good” does not explain what good means. Describe observable choices, such as length, tone, reading level, format, or how uncertainty should be handled.

Leaving contradictions in place: A preference for short answers can conflict with a project request for a detailed report. Say which instruction takes priority, or clarify that project-specific requirements override general preferences.

Assuming uploads are permanent memory: File access depends on the tool, conversation, project, and current feature settings. Verify the files are present in the place where you are working, especially in a new chat.

Including confidential material: Do not upload passwords, access tokens, private client information, or other sensitive data unless you have confirmed that sharing it is appropriate and understand the service’s data controls.

Tips

Put only durable details in profile.md; move temporary deadlines and decisions to project.md.
Use “must,” “prefer,” and “ask before” deliberately so firm requirements are distinguishable from suggestions.
Add examples when a style preference is difficult to explain, but keep them brief and label them as examples.
For fast-changing projects, include an update date and remove superseded decisions rather than appending an endless history.
Keep the same three source files locally and upload their latest versions to both tools. There is no need to make the wording identical to a particular tool’s interface.

Troubleshooting

The assistant ignores a file: Confirm the file is attached to the active conversation or added to the project you opened. Ask the assistant to summarize the relevant instructions; if it cannot, reattach the file or paste the essential portion. Avoid assuming that a file in another project or chat carries over.

The response follows one instruction but misses another: Look for conflicting wording between the files and your prompt. Clarify priority, then update the appropriate file. Put a one-time requirement in the prompt rather than making it a permanent preference.

The answer uses outdated project information: Check the update date and replace the old project file. If the tool retains older reference material, remove the outdated copy where possible and confirm which version is current in the prompt.

The answer is still generic: Add concrete audience details, examples, constraints, or decision criteria to the relevant file. Then test again with a specific task. More text alone may not help; prioritize the missing information that would change the answer.

A file will not upload: Check the service’s current file and project requirements, including accepted formats and size limits. Save a simpler Markdown or plain-text copy, shorten it if necessary, and try again. If uploads are unavailable, paste the relevant notes into the conversation.

Frequently Asked Questions

Do I need exactly three files?
Three is a useful starting structure, not a technical requirement. You can combine them for a small task, but separating stable background, response preferences, and project details makes updates clearer.

Can I use these files with both ChatGPT and Claude?
Yes. Keep a local copy and add each file separately to the relevant project or conversation in each service. Features and file-handling behavior may differ.

Will the files apply to every new chat automatically?
Not necessarily. They generally apply only where you have added them, such as a particular project or conversation. Check the active workspace before relying on them.

Should I include personal information?
Only include details that are useful for the task and appropriate to share. Avoid secrets and unnecessary sensitive information, and review each service’s current privacy and data controls.

How often should I update the files?
Update stable profile or preference details when they change. Review the project file whenever its status, decisions, audience, or constraints change, and keep a date to show when it was last checked.

What if the assistant misreads a note?
Rewrite it in direct, unambiguous language and add a short example if helpful. Then ask the assistant to restate the relevant requirement before continuing with important work.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.