AI Projects: Persistent Context

The game-changer most people don't use — persistent context that carries across conversations.

INFO

• Time: ~25 minutes
• Difficulty: Beginner
• What you'll learn: How to set up AI assistants that know your work

---

This Page Covers

• What Are AI Projects? - Persistent context that carries across conversations
• Setting Up a Project - Step-by-step for Claude and ChatGPT
• ChatGPT Memory - A different kind of persistence that works across all conversations
• Custom Instructions - Templates for different use cases (business analyst, writing assistant, researcher)
• What to Upload - Documents that make AI more useful
• Privacy Considerations - What to think about before uploading

---

What Are AI Projects?

Both Claude and ChatGPT offer "Projects" - a way to create specialized AI assistants with persistent context. Instead of re-explaining your situation every conversation, you set it up once and it remembers.

What you can do:

• Upload documents - AI can reference them in every conversation
• Set custom instructions - Rules that apply to every response
• Create specialists - Different projects for different types of work

Think of it like: Creating a new employee who already knows your company, your style, and your preferences - without the onboarding time.

Why This Matters

Without projects, every conversation starts from zero. You constantly repeat:

• "I work at a B2B software company..."
• "Our target audience is..."
• "I prefer concise responses..."

With projects, you set this context once. Every conversation in that project already knows.

---

Setting Up a Project

Step 1: Create the Project

1. Go to claude.ai
2. Click Projects in the left sidebar
3. Click New Project
4. Give it a descriptive name (e.g., "Marketing Content", "Client Proposals", "Research Assistant")

Step 2: Add Custom Instructions

Custom instructions tell Claude how to behave, what to assume, and how to format responses.

Click Edit Project Instructions and add your rules (see templates below).

Step 3: Upload Relevant Documents

Add documents Claude should reference:

• Click Add Content or drag files into the project
• Supported: PDF, DOC, TXT, CSV, and more
• Size limits apply (check current limits at claude.ai)

Step 4: Start a Conversation

Click New Chat within the project. Claude now has access to your instructions and documents.

---

ChatGPT Memory: A Different Kind of Persistence

ChatGPT has a feature called Memory that works differently from Projects. Understanding the difference helps you use both effectively.

Projects vs Memory

Feature	AI Projects	ChatGPT Memory
Scope	Specific to one project	Applies to ALL conversations
What's stored	Documents, instructions	Facts about you
How it's set	You configure it explicitly	ChatGPT remembers from conversations
Use case	Specialized work contexts	Personal preferences

How Memory Works

ChatGPT Memory stores facts you tell it (or that it learns from conversations):

• "I'm a product manager at a tech startup"
• "I prefer bullet points over paragraphs"
• "My team uses Python, not JavaScript"
• "I'm allergic to peanuts" (if you're asking for recipes)

These facts then influence ALL future conversations - not just within a project.

Managing Memory

To view/edit memories:

1. Go to ChatGPT Settings
2. Click Personalization → Memory
3. Review what ChatGPT has stored
4. Delete anything incorrect or outdated

To add memories manually:

• Tell ChatGPT directly: "Remember that I prefer concise responses"
• ChatGPT will confirm it saved the information

To prevent memory:

• Say: "Don't remember this" before sharing something
• Or disable Memory entirely in settings

Using Projects + Memory Together

The most effective setup uses both:

• Memory for your general preferences and background

  • Your role and company
  • Communication style preferences
  • Technical stack you use

• Projects for specific work contexts

  • Marketing project with brand guidelines
  • Client project with their requirements
  • Research project with relevant papers

This way, ChatGPT always knows your general preferences (Memory), and has specialized context when you're in a specific project.

Claude Note

As of now, Claude does not have an equivalent "Memory" feature - it only has Projects. Claude's persistent context is entirely project-based.

---

Custom Instructions That Work

Copy and customize these templates for your projects.

For a Business Analyst

You are helping a business analyst at [COMPANY TYPE].

About my role:
- I analyze [TYPE OF DATA/DECISIONS]
- I present findings to [AUDIENCE - executives, stakeholders, etc.]
- My deliverables include [reports, dashboards, recommendations, etc.]

When I ask for analysis:
- Start with the key insight, then supporting details
- Quantify when possible (percentages, comparisons, trends)
- Flag assumptions you're making
- Suggest what additional data would strengthen the analysis

Formatting preferences:
- Use tables for comparisons
- Use bullet points for lists of findings
- Keep summaries to 3-5 key points unless I ask for more detail

Industry context:
[Add relevant industry information, terminology, competitive landscape]

---

For a Writing Assistant

You are my writing assistant for [TYPE OF CONTENT].

My writing style:
- Tone: [Professional but approachable / Casual and conversational / Formal / etc.]
- Length preference: [Concise - get to the point / Detailed - thorough explanations]
- Voice: [First person / Third person / We (company)]

My audience:
- [Who they are]
- [What they care about]
- [Their expertise level]

Brand guidelines:
- We say: [preferred terms, phrases]
- We don't say: [avoided terms, competitor names, etc.]
- Our values: [key messages to reinforce]

When writing:
- Match the style of documents I've uploaded
- Suggest alternatives when something feels off-brand
- Ask clarifying questions before writing long pieces

---

For a Research Assistant

You are my research assistant for [DOMAIN/TOPIC].

My background:
- My expertise level: [Beginner / Intermediate / Expert]
- I'm researching for: [personal learning / work project / decision-making]

How I want information presented:
- Start with a summary, then details if I ask
- Cite sources when making factual claims
- Distinguish between established facts and emerging research
- Flag when information might be outdated

My research focus:
- Primary topics: [List main areas of interest]
- I'm particularly interested in: [Specific aspects]
- I'm less interested in: [What to de-emphasize]

When I ask questions:
- If you don't know something, say so rather than guessing
- Suggest related questions I might want to explore
- Point me to authoritative sources for deeper reading

---

What to Upload

The right documents make your project significantly more useful.

High-Value Uploads

Document Type	Why It Helps
Brand guidelines	AI matches your voice and style
Strategy docs	AI understands your goals and priorities
Past work examples	AI can match your existing quality and format
Product/service info	AI can answer questions accurately
Competitor analysis	AI can position against alternatives
FAQ / Common questions	AI can reference standard answers

How to Get the Most Value

• Be selective - Upload 5-10 highly relevant documents, not 50 tangentially related ones
• Use good documents - If your brand guidelines are outdated, AI will reference outdated info
• Update periodically - Remove stale documents, add new relevant ones
• Name files clearly - "Q3-2024-Strategy.pdf" is better than "Document1.pdf"

---

Privacy Note

Before uploading anything to AI Projects, consider:

What Happens to Your Data

• Documents go to the provider's servers (Anthropic for Claude, OpenAI for ChatGPT)
• Both ChatGPT and Claude train on your data by default (all consumer tiers)
• You can opt out in settings (see Module 7 for details)
• Team/Enterprise plans don't use your data for training
• Even with training disabled, data still passes through their servers

What NOT to Upload

Never upload these

• Passwords, API keys, or credentials
• Customer personal data (names, emails, addresses)
• Financial account numbers
• Health or medical records
• Anything covered by NDA that prohibits sharing
• Proprietary source code (if your company policy prohibits it)

What to Be Careful With

• Internal strategy documents (could leak competitive info)
• Employee information
• Unpublished financial data
• Client-specific details (anonymize first)

When in Doubt

• Check your company's AI usage policy
• Ask your manager or legal team
• Anonymize sensitive details before uploading
• Use Team/Enterprise plans for better privacy controls

---

Key Takeaways

• Projects save repetition - Set up context once, use it in every conversation
• Custom instructions shape behavior - Tell AI how to respond, what to assume, what to avoid
• ChatGPT Memory is different - It applies to ALL conversations, not just one project
• Upload strategically - High-quality, relevant documents make AI more useful
• Privacy matters - Think before uploading anything sensitive or confidential
• Create multiple projects - Different projects for different types of work (marketing, research, client work)