iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Keep reusable context in three files: profile.md for who you are and what you need, project.md for the work itself, and instructions.md for how the AI should respond. Add the files to a ChatGPT Project or Claude Project, then put the most important behavior rules in that product’s project-instructions field. This is a practical organization method, not a template required by OpenAI or Anthropic.
Why use three files for persistent context?
Projects give recurring work a home for its chats, reusable material, and instructions. Separating context into three files makes it easier to update one kind of information without rewriting everything: personal background stays distinct from project facts, and both stay distinct from response preferences.
The files do not make the two services remember context in the same way. ChatGPT offers project memory settings that can let chats refer to other conversations in that project. Claude shares context across chats through the project knowledge base; information in one chat is not automatically shared with another unless it is added there. See OpenAI’s Projects guidance and Anthropic’s explanation of Projects.
Recommended Free Tools
What to put in the three files
profile.md: stable background
Include details that help across many tasks but are unlikely to change often. Keep this concise and relevant; avoid sensitive information you do not need the service to use.
#1 Best Overall
- Your role, experience, audience, or recurring responsibilities.
- Long-term goals and constraints that affect recommendations.
- Preferences that are not specific to one project, such as preferred level of detail or accessibility needs.
project.md: the current source of truth
Describe the project and its current state. Favor concrete facts over broad descriptions, and identify authoritative sources so the assistant can distinguish settled information from assumptions.
- Purpose, audience, scope, and success criteria.
- Names, terminology, dates, decisions, and unresolved questions.
- Links to source-of-truth documents or pages, with a note about which information each link controls.
instructions.md: response and task rules
Record the repeatable way you want work handled. Separate mandatory requirements from preferences, and make output formats explicit when they matter.
- How to approach tasks, including when to ask for clarification and how to handle uncertainty.
- Tone, structure, and terminology preferences.
- Required output format, length limits, or checks before finalizing.
Project files can hold detailed reference material, but copy the highest-priority behavior rules into the product’s native project-instructions field. That makes the key directions easy to find while the files provide supporting context.
Free tools Windows power users keep installed
One-click scans. No signup required.
Create the files and add them to a project
- Create three plain-text files named
profile.md,project.md, andinstructions.md. Use short headings and bullets so the content is easy to scan and revise. - Create or open a project in ChatGPT or Claude. Add the files as project reference material or knowledge, and paste the essential behavior rules into the project-instructions field.
- Start a new chat inside the project and test it with a representative task. Correct missing, stale, or conflicting context in the files or instructions rather than adding the same correction repeatedly in chat.
In ChatGPT
Open Projects, create or open the project, and add files in its project file area. Add the core behavior rules in the project instructions. Open the project’s menu or settings to review its memory option; available choices and labels may depend on account and workspace settings. OpenAI’s current Projects guidance says project memory can be default or project-only. With project-only memory, ChatGPT does not refer to personal saved memories or conversations outside the project, though project chats may refer to other conversations in that project. Memory availability also depends on personal and workspace settings. See OpenAI’s Projects guidance.
Rank #3
In Claude
Open Projects, create or open a project, add the files to project knowledge, and enter key behavior rules in project instructions. Anthropic describes uploaded project knowledge as usable across chats in that project. A conversation’s context is not shared with other chats unless the information is added to the project knowledge base. Claude Projects are described by Anthropic as available to paid-plan users; availability can depend on the current plan and workspace. See Anthropic’s project-management guidance.
Markdown format, file limits, and plan details
OpenAI and Anthropic document support for TXT files and project reference material, but the cited upload guidance does not specifically guarantee the .md extension in every interface. If an upload rejects a Markdown file, save a plain-text .txt copy or paste the contents into the project. Check the live help pages and your account’s project interface before depending on a particular extension or limit.
| Service | File and project limits stated in official guidance | Plan or availability note |
|---|---|---|
| ChatGPT Projects | Free: 5 files per project; Go and Plus: 25; Edu, Pro, Business, and Enterprise: 40. A maximum of 10 files can be uploaded at one time. | OpenAI’s Projects guidance; plan entitlements can change. See Projects in ChatGPT and file-upload guidance. |
| Claude project knowledge | 30 MB maximum per file; no stated file-count limit, but total content must fit the context window. | Anthropic’s upload guidance. See supported document types. |
Anthropic says retrieval-augmented generation (RAG) can activate as project knowledge approaches the context-window limit, expanding capacity by up to 10x. That is Anthropic’s product statement, not an independent performance measure. Details can change; confirm current limits and eligibility in the official documentation.
Review memory, sharing, and data-use settings
Check who can access a project and what account controls apply before adding personal or company material. A project file is reusable context, not a confidentiality guarantee.
Quick Recap
- ChatGPT shared projects: OpenAI says these use project-only memory and cannot access a member’s outside context, custom instructions, or memories. This describes project context boundaries, not a general promise about data handling.
- ChatGPT data use: OpenAI says Business, Enterprise, Edu, and Healthcare workspace data is not used to train models by default. Consumer Free, Plus, and Pro project information may be used if “Improve the model for everyone” is enabled. Review the setting and plan that apply to your account. See OpenAI’s Projects guidance.
- Claude sharing: Claude for Work project visibility may be private or organization-wide, depending on workspace settings. Review the controls available to your workspace before sharing. See Anthropic’s project-management guidance.
Keep the context useful over time
- Put durable background in
profile.md, active facts and decisions inproject.md, and repeatable response requirements ininstructions.md. - Date information that may become stale, such as deadlines, product details, or active decisions. Mark whether a statement is confirmed, tentative, or unresolved.
- When a decision changes, update or remove the old version in the source file. Do not leave contradictory instructions for the assistant to reconcile.
- Periodically check whether the files still contain relevant information. Remove details that no longer help, and keep the native instructions focused on the rules that most affect behavior.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
