I Tested LLM Projects Feature Across ChatGPT, Claude, and Gemini — Here's What Actually Works
The biggest problem with AI assistants has always been context loss.
You start analyzing your marketing data in ChatGPT, switch to Claude for writing, jump to Gemini for research — and every time you switch contexts or start a new conversation, you're explaining everything from scratch.
Every. Single. Time.
Major LLM providers just solved this with Projects — persistent workspaces that maintain your files, instructions, and conversation history across multiple sessions.
I spent two weeks testing Projects across ChatGPT, Claude, and Gemini with real business workflows: uploading financial statements, configuring custom instructions, running complex queries, and switching between platforms to understand which implementation works best.
The feature fundamentally changes how you work with AI assistants, though each platform handles it differently.
Let me show you exactly how Projects work across major LLMs and which platform suits your specific workflow.
What Are LLM Projects?
Press enter or click to view image in full size
LLM Projects are dedicated workspaces that maintain persistent context across multiple conversations.
Think of a Project as a specialized environment for recurring work that requires consistent context, specific instructions, and access to reference materials.
The Universal Problem Projects Solve
Before Projects, every AI conversation started with a blank slate:
You'd upload the same files repeatedly Explain your preferences in every chat Lose valuable conversation history when starting fresh Waste time rebuilding context instead of doing actual work
Projects eliminate this friction by creating persistent workspaces where:
- Uploaded files remain accessible across all conversations
- Custom instructions apply automatically to every interaction
- Related chats stay organized under one umbrella
- Context carries forward without manual repetition
Platform Availability
As of January 2026, Projects are available across major LLM platforms:
ChatGPT: Available for Free, Plus, and Pro plans (rolling out gradually to free users)
Claude: Available for all paid plans (Pro, Team, Enterprise)
Gemini: Available in Gemini Advanced (part of Google One AI Premium)
The core concept remains consistent across platforms, though implementation details vary. Each provider adds platform-specific features while maintaining the fundamental value proposition: persistent context that eliminates repetitive setup work.
The Universal Pattern: How Projects Work
Press enter or click to view image in full size
Despite platform differences, Projects share common architectural patterns.
Three Core Components
Every Project implementation includes:
- File Storage System Upload documents, spreadsheets, images, and code that persist across conversations. The AI assistant can reference these files in any chat within the project.
- Custom Instructions Define behavioral rules, output preferences, and contextual guidelines that apply automatically to every conversation in the project.
- Organized Conversations Multiple chat threads exist within one project, each maintaining access to the same files and instructions while allowing you to explore different aspects of your work.
The Workflow Pattern
The universal workflow follows this structure:
Initial Setup:
- Create project with descriptive name
- Upload relevant files
- Configure custom instructions
- Test with verification queries
Daily Usage:
- Start new chat within project
- Ask questions or make requests
- AI responds with full context awareness
- Switch between chats while maintaining context
Maintenance:
- Add new files as needed
- Update instructions based on evolving requirements
- Remove outdated materials
- Organize conversations for easy reference
Why This Pattern Works
The Projects pattern succeeds because it mirrors how humans actually work:
You don't start every work session from zero. You have files, established processes, and accumulated knowledge. Projects give AI assistants the same persistence.
You organize work by domain. Marketing in one space, finance in another, development in a third. Projects enforce this natural separation.
You iterate on existing work. Previous conversations inform current tasks. Projects maintain this continuity without forcing you to manually provide context.
Platform-Specific Implementation: ChatGPT
Press enter or click to view image in full size
ChatGPT's Projects implementation emphasizes accessibility and simplicity.
Creating a ChatGPT Project
Navigate to the Projects tab in your ChatGPT sidebar. Click "New Project." Name your project with clear scope identification.
I created "Financial Advisor" to test the feature with profit and loss statements that I'd reference across multiple analytical conversations.
Key Features
Color Coding: Assign colors to projects for visual organization. I use light blue for financial work, green for marketing, orange for development. This visual hierarchy prevents confusion when managing multiple projects.
Model Selection: Choose which GPT model handles your queries — but with an important constraint. Once you upload files, you're locked to GPT-4. No access to GPT-4o or other models when working with uploaded documents.
File Management: Upload PDFs, CSVs, images, and code files. ChatGPT supports approximately 20-30 files per project, though OpenAI doesn't specify exact limits.
File Type Restrictions: Excel files (.xlsx) don't work — export to CSV first. Google Sheets require the same treatment. Word documents must convert to PDF.
Instructions Configuration
ChatGPT's instruction system evolved from the older "Custom Instructions" feature. The difference is scope — custom instructions applied globally, while project instructions only activate within that specific project.
Example instruction for my Financial Advisor project:
"Please point out any financial red flags or inefficiencies and suggest potential strategies for improvement."
This instruction now applies to every financial analysis within the project. ChatGPT doesn't just answer questions — it proactively identifies problems and recommends solutions.
Chat Organization
Every conversation within a project maintains full context with uploaded files and instructions. Click "New Chat" to start fresh while keeping access to all project resources.
You can also migrate existing chats into projects. Find the chat in your sidebar, click the three dots, select "Add to Project," and choose your target project. The entire conversation moves into the project workspace.
Testing and Verification
After uploading a CSV with financial data, I tested with a specific query: "Which year had the highest net sales?"
ChatGPT responded: "2029 had the highest net sales at $119,218.41."
I verified this against source data. Correct down to the cent. The system read the file accurately and extracted precise information.
ChatGPT Projects Strengths:
- Intuitive interface with minimal learning curve
- Strong file parsing accuracy
- Effective instruction following
- Seamless chat migration from existing conversations
ChatGPT Projects Limitations:
- File uploads lock you to GPT-4 (no GPT-4o access)
- Limited file count per project
- No Excel or Google Sheets support (CSV only)
- Gradual rollout to free users
Platform-Specific Implementation: Claude
Press enter or click to view image in full size
Claude's Projects implementation focuses on team collaboration and knowledge management.
Creating a Claude Project
Access Projects from the Claude sidebar. Click "Create Project." Define your project with a clear name and optional description that helps team members understand its purpose.
Claude Projects emphasize knowledge base construction — you're not just uploading files for reference, you're building a domain-specific knowledge repository.
Key Features
Knowledge Base Management: Claude treats uploaded content as a structured knowledge base rather than loose files. The system indexes content for semantic search, making retrieval more intelligent than simple keyword matching.
Team Collaboration: Claude Projects support team access with role-based permissions. Multiple team members can work within the same project, sharing context and instructions seamlessly.
Custom Instructions: Define project-specific behaviors similar to ChatGPT, but with more granular control over response patterns. You can specify tone, output format, citation requirements, and domain-specific conventions.
File Support: Claude accepts PDFs, text files, code, and images. Like ChatGPT, Claude has file count limits, though Anthropic doesn't publish exact numbers.
Conversation Threading: Each project contains multiple conversation threads that maintain access to the project's knowledge base and instructions.
Claude Projects Strengths:
- Superior semantic search and retrieval
- Team collaboration features
- More flexible instruction system
- Better handling of complex, multi-document queries
Claude Projects Limitations:
- Available only on paid plans
- Steeper learning curve for optimal configuration
- Less visual organization than ChatGPT
- File upload format restrictions
Platform-Specific Implementation: Gemini
Press enter or click to view image in full size
Gemini's approach to Projects integrates with Google's broader ecosystem.
Creating a Gemini Project
Access Projects through Gemini Advanced (requires Google One AI Premium subscription). The interface mirrors Google Workspace conventions for familiarity.
Key Features
Google Workspace Integration: Gemini Projects connect directly to Google Drive, Docs, Sheets, and Slides. Instead of uploading files manually, you grant access to existing Google Workspace content.
This integration is powerful for teams already using Google's ecosystem. Your project automatically syncs with Drive folders, maintaining real-time access to updated documents.
Collaborative Context: Multiple users can work within shared projects that reference the same Google Workspace files, maintaining consistency across team members.
Custom Instructions: Similar to other platforms, Gemini supports project-specific behavioral rules and output preferences.
Multimodal Capabilities: Gemini handles text, images, and video content within projects, leveraging Google's multimodal model strengths.
Gemini Projects Strengths:
- Seamless Google Workspace integration
- Real-time document syncing
- Strong multimodal support
- Natural fit for Google ecosystem users
Gemini Projects Limitations:
- Requires Google One AI Premium subscription
- Less flexible for non-Google Workspace users
- Smaller user base means fewer community resources
- Feature set still evolving compared to competitors
Setting Up Your First Project: Universal Guide
Press enter or click to view image in full size
The setup process follows similar patterns across platforms with platform-specific optimizations.
Step 1: Define Project Scope
Before creating a project, clearly identify:
Purpose: What recurring work will this project handle? Domain: Marketing, development, analysis, content creation? Users: Solo or team collaboration? Duration: Ongoing or time-limited?
Concrete example: "Q1 2026 Marketing Campaign — Social Media Strategy"
This scope definition prevents project sprawl where everything gets dumped into generic "Work" projects that become unusable.
Step 2: Choose Your Platform
Select based on your workflow:
Choose ChatGPT if:
- You work solo or small team
- Need simple, fast setup
- Work primarily with PDFs and CSVs
- Want strong general-purpose AI
Choose Claude if:
- Complex multi-document analysis
- Team collaboration is critical
- Need superior semantic search
- Work with technical documentation
Choose Gemini if:
- Already use Google Workspace
- Need real-time document syncing
- Require multimodal capabilities
- Want ecosystem integration
Step 3: Upload Core Files
Start with 3-5 essential documents, not everything you might need.
Good initial uploads:
- Brand guidelines
- Recent project documentation
- Reference materials you consult frequently
- Data files for analysis
Avoid uploading:
- Archive materials you rarely reference
- Duplicate content in different formats
- Extremely large files that consume storage limits
- Outdated versions of documents
Step 4: Configure Instructions
Write instructions that define:
Tone and Style: Professional, casual, technical, creative Output Format: Structured lists, narratives, code with comments Domain Context: Industry terminology, company conventions Behavioral Rules: "Always include data sources," "Flag assumptions clearly"
Example instruction for a marketing project:
"Maintain our brand voice: professional but approachable. Include one emoji in social media posts. Always cite data sources in analysis. Format recommendations as actionable bullet points with implementation difficulty ratings (Easy/Medium/Hard)."
Step 5: Test with Verification Queries
Don't assume your setup worked. Immediately test with specific questions that require reading uploaded files:
"What's the budget allocation for Q1 mentioned in the project brief?" "Find the contact information for our primary vendor in the contracts folder." "Summarize the key findings from last quarter's performance report."
If the AI can't answer accurately, your files weren't processed correctly. Check file formats, re-upload, or try different file types.
Step 6: Iterate and Refine
Use the project for actual work and note what's missing:
Files you keep needing but haven't uploaded Instructions you're repeating manually Output patterns you want consistent Context that's still getting lost
Update your project configuration based on real usage, not theoretical needs.
Real-World Use Cases Across Platforms
Press enter or click to view image in full size
Projects shine in scenarios requiring persistent context and repeated reference to specific information.
Marketing Campaign Management
Platform Choice: Claude or ChatGPT
Setup:
- Upload brand guidelines, past campaign data, competitor analysis
- Configure instructions for brand voice consistency
- Add image assets for reference
Usage Pattern:
- Generate social media content maintaining brand voice
- Analyze campaign performance against historical data
- Create ad copy variations following established patterns
Why Projects Help: Every piece of content maintains brand consistency without manually specifying voice, style, or guidelines in each request.
Software Development Workflows
Platform Choice: ChatGPT or Claude
Setup:
- Upload codebase documentation, API specifications, style guides
- Configure code formatting and commenting preferences
- Add architecture diagrams and decision records
Usage Pattern:
- Generate code following team conventions
- Debug issues with full codebase context
- Review pull requests against established patterns
Why Projects Help: Code suggestions match your team's standards and architectural decisions without repeated context setting.
Business Financial Analysis
Platform Choice: ChatGPT (tested extensively with P&L data)
Setup:
- Upload financial statements (CSV format)
- Configure analysis instructions (identify inefficiencies, suggest improvements)
- Add industry benchmark data for comparison
Usage Pattern:
- Query specific metrics across time periods
- Identify trends and anomalies
- Generate reports with consistent formatting
Why Projects Help: Every analysis maintains access to complete financial history and applies consistent analytical framework.
Content Creation and Editing
Platform Choice: Claude (superior for long-form content) or Gemini (if using Google Docs)
Setup:
- Upload style guides, past articles, editorial guidelines
- Configure tone, structure, and formatting preferences
- Add audience personas and content strategy
Usage Pattern:
- Draft articles following established style
- Edit existing content for consistency
- Generate variations maintaining voice
Why Projects Help: Content maintains consistent voice and style across multiple pieces without manual specification.
Research and Documentation
Platform Choice: Claude (best semantic search) or Gemini (Google Workspace integration)
Setup:
- Upload research papers, articles, reference materials
- Configure citation and sourcing requirements
- Add domain-specific terminology glossaries
Usage Pattern:
- Query across multiple sources
- Synthesize findings from disparate documents
- Generate summaries with proper citations
Why Projects Help: AI maintains access to complete research library and applies consistent citation standards.
Advanced Features and Best Practices
Press enter or click to view image in full size
Power users leverage Projects beyond basic file storage and instructions.
Cross-Platform Strategies
Don't limit yourself to one platform. Use each for its strengths:
ChatGPT: Quick analysis, general tasks, brainstorming Claude: Complex analysis, technical writing, multi-document synthesis Gemini: Google Workspace integration, collaborative editing
Create parallel projects across platforms for the same work domain. When one platform's limitations block you, switch to another while maintaining similar context through synchronized file uploads.
Instruction Optimization
Effective instructions are:
Specific: "Format code with inline comments explaining logic" beats "Write good code"
Contextual: Include domain terminology and company-specific conventions
Actionable: Define clear behaviors rather than vague preferences
Testable: Write instructions you can verify with specific queries
Bad instruction example: "Be helpful and professional"
Good instruction example: "Maintain professional tone. Start responses with brief summary, then detailed explanation. Include examples for technical concepts. Flag assumptions clearly. Cite sources for factual claims."
File Management Strategies
Organize by recency: Keep frequently referenced files; archive old materials
Version control: Include version numbers or dates in filenames
Format optimization: Convert files to platform-supported formats before upload
Size management: Split large documents into focused sections
Regular pruning: Monthly review to remove outdated content
Conversation Organization
Descriptive naming: "Budget Analysis Q1 2026" not "New Chat 1"
Thematic grouping: Related queries in same chat, new topics in new chats
Archive completed work: Move finished projects to separate workspace
Search optimization: Use consistent terminology for easy retrieval
Testing and Verification Workflows
After any configuration change:
- Test with specific query requiring new context
- Verify accuracy against source materials
- Check instruction application in responses
- Adjust and retest until behavior matches expectations
Never assume configuration worked without verification.
Limitations You Need to Understand
Press enter or click to view image in full size
Projects have boundaries that affect workflow decisions.
File Type Restrictions
Universal limitations:
- Excel files don't work (export to CSV)
- Google Sheets require CSV conversion
- Complex formatting often gets lost
- File size limits vary by platform
Platform-specific:
- ChatGPT: CSV, PDF, images, code
- Claude: Text-based formats, PDFs, code, images
- Gemini: Native Google formats preferred
Storage and Count Limits
No platform offers unlimited file storage within projects:
ChatGPT: ~20-30 files (unofficial limit) Claude: Undisclosed limits, but constraints exist Gemini: Tied to Google Drive storage quota
Hit these limits and you're forced to:
- Remove old files
- Create new projects
- Optimize file sizes
- Split work across multiple projects
Model Constraints
ChatGPT: File uploads lock you to GPT-4 (no GPT-4o access) Claude: Projects work with latest Claude model Gemini: Tied to Gemini Advanced capabilities
This means if you need specific model features (speed, capabilities, cost), Projects might force compromises.
Context Window Limitations
Even with Projects, AI models have finite context windows. Uploading 50 documents doesn't mean the AI can process all 50 simultaneously in one query.
The system prioritizes relevant content, but extremely complex queries spanning many documents may exceed processing capabilities.
No Cross-Project References
Projects are isolated. You can't reference files from one project while working in another.
Need data from multiple projects simultaneously? Your options:
- Consolidate into one project (faster limits)
- Work outside Projects entirely
- Manually copy content between projects
Collaboration Limitations
ChatGPT: No native team collaboration (Plus/Pro are individual accounts) Claude: Team features on Team/Enterprise plans only Gemini: Collaboration through Google Workspace sharing
Solo users have full Projects access. Teams need paid collaboration features.
Final Thoughts
Projects fundamentally change how you work with AI assistants by eliminating context loss and repetitive setup.
The feature works well across ChatGPT, Claude, and Gemini, though each platform optimizes for different use cases:
For solo users needing quick setup: ChatGPT offers the most accessible entry point with intuitive interface and rapid configuration.
For teams handling complex analysis: Claude provides superior semantic search, collaboration features, and sophisticated instruction systems.
For Google Workspace users: Gemini's native integration with Drive, Docs, and Sheets creates seamless workflows if you're already in that ecosystem.
Getting Started Recommendations
Beginners: Start with ChatGPT. The interface is most forgiving, and you can experiment freely without complex configuration.
Intermediate users: Try Claude Projects for work requiring deep document analysis or team collaboration.
Advanced users: Use all three platforms strategically, leveraging each for specific strengths while maintaining parallel project structures.
The investment in proper project setup pays off within days through time savings and improved output quality.
Whether you're analyzing business data, managing marketing campaigns, or developing software, Projects turn AI assistants from one-off tools into persistent collaborators that understand your work context.
Have you tried Projects features across different LLM platforms?
Which implementation works best for your workflow?