Master AI for Your Business & Career
🔹 AI Practical Book – AI for the Rest of Us Real-world strategies for professionals (Available on Notion Press)
🔹 AI Consultation Services – Tailored guidance for businesses navigating AI transformation
🔹 Corporate AI Training – Hands-on workshops for teams ready to leverage AI tools effectively
📧 Contact us: | LinkedIn
AI Reality Series Part 8: Context Windows & Projects – The Memory Problem No One Talks About
Series Navigation: Catch Up on AI Realities
This is Part 8 of our ongoing AI Reality Series. If you're new here, we've been cutting through the hype to give you honest, practical insights into how AI actually works:
Part 1: AI Myths vs Reality – Separating fact from fiction
Part 2: Prompt Engineering Fundamentals – How to actually communicate with AI
Part 3: Real-World Limitations – Where AI breaks down
Part 4: The Hallucination Problem – Why AI confidently lies
Part 5: Bias in AI Systems – The hidden prejudices in algorithms
Part 7: Why Different Tools Give Different Answers – Model architecture matters
Coming Next: Part 9 will tackle Data Privacy, Conversation Ownership, and How to Choose the Right AI Tool for Sensitive Work
✨ Download this article as a PDF — ideal for offline reading or for sharing within your knowledge circles where thoughtful discussion and deeper reflection are valued.
The Invisible Wall Every AI User Hits
The 2 AM Frustration
It's 2 AM. You've been working with ChatGPT for the past hour, building a detailed project plan. The AI has been brilliant – understanding your requirements, refining your timeline, suggesting improvements. You're in the zone. Flow state achieved.
Then you ask one more question, and suddenly... it's like talking to someone with amnesia.
"I don't have information about the project you mentioned."
Wait, what? We just spent an hour on this! You scroll up – everything is there. The entire conversation. But somehow, the AI has forgotten it all.
Welcome to the invisible wall every AI user eventually hits: the context window limit.
What Just Happened? Understanding AI's "Working Memory"
Think of an AI's context window like a whiteboard in a meeting room. At the start of your conversation, the board is empty. As you talk, everything gets written down – your questions, the AI's responses, uploaded files, images.
The problem? The whiteboard has a fixed size.
Once it's full, something has to give. The AI starts "erasing" older parts of the conversation to make room for new content. It's not being forgetful or glitchy – it's literally running out of memory space.
This isn't a bug. It's architecture.
And here's what almost nobody tells you: different AI tools have wildly different whiteboard sizes – and the marketing claims don't always match reality.
The Token Truth: Why AI Doesn't Count Words
Before we compare tools, you need to understand one critical concept: AI doesn't think in words. It thinks in tokens.
What Is a Token?
A token is a chunk of text that AI processes as a single unit. Sometimes it's a whole word. Sometimes it's part of a word. Sometimes it's just punctuation.
Examples:
"Hello" = 1 token
"Artificial" = 2 tokens (Art + ificial)
"Understanding" = 3 tokens (Under + stand + ing)
"!" = 1 token
Why tokens instead of words?
Tokens give AI flexibility. When it encounters an unusual word like "counterintuitive," it doesn't need to have seen that exact word before. It breaks it into familiar chunks: "counter" + "intu" + "itive" – pieces it recognizes.
This is brilliant engineering. But it creates an invisible inequality.
🚨 The Language Penalty: Why Indian Users Hit Limits Faster
Here's something most AI companies won't emphasize: English is the most token-efficient language.
Because AI models were primarily trained on English text, English words compress better. Other languages – especially those with different scripts – require more tokens for the same meaning.
Real Example:
English: "Good morning, how are you?" = ~6 tokens
Hindi: "सुप्रभात, आप कैसे हैं?" = ~12-15 tokens
Tamil: "காலை வணக்கம், எப்படி இருக்கிறீர்கள்?" = ~15-18 tokens
Same meaning. Double or triple the token cost.
This means:
✗ Non-English conversations hit the context limit faster
✗ Free tier limits expire quicker
✗ In paid plans, you're charged more tokens for the same conversation
For Indian professionals switching between English, Hindi, and Tamil in conversations: Your effective context window is smaller than advertised. This isn't discrimination – it's a side effect of how tokenization works. But it's real, and it matters.
The 2026 Context Window Reality Check
Let's cut through the marketing and look at what you actually get:
Key Reality Check: The numbers provided here represent the most usable data collected from the web through AI research and user reports. While these limits reflect typical free-tier experiences, users accessing the models via API may be able to utilize significantly higher context windows than those listed above.
What this means for you: If you're on free tiers, expect to hit memory limits within 10-20 exchanges, depending on message length. Claude Free gives you the most breathing room. ChatGPT Free is the most restrictive.
💰 TOKEN ECONOMICS: The Hidden Cost
Quick Callout Box:
Most AI tools charge separately for input tokens (what you send) and output tokens (what AI generates).
Why this matters:
Output tokens cost 2-3x more than input tokens
Free tiers have strict usage limits beyond just context window
The advertised context window might exist technically, but usage caps kick in first
Reality: Free tiers are designed to give you a taste, not to support regular, sustained use. For professional workflows, paid plans become necessary quickly.
Sam Altman's Two Visions: Compute vs Memory
OpenAI's CEO Sam Altman has made two significant statements about AI bottlenecks, often confused as one:
Statement 1: The Compute Bottleneck (September 2025)
"AI's future isn't bottlenecked by talent or ideas – it's bottlenecked by compute. Over the next 1-2 years, scarcity could force brutal trade-offs: Do we allocate GPUs to cure cancer? Or to power free global education?"
Context: This was about hardware – the physical chips and energy needed to run AI models.
Statement 2: The Memory Breakthrough (December 2025)
"The real AI breakthrough will not be better reasoning, but total memory. An AI that can remember every conversation, email, and document across a person's lifetime."
Context: This is about persistent memory – AI that remembers you across sessions, devices, and months or years.
The difference matters: Context windows (what we're discussing today) are temporary working memory within a single conversation. Altman's vision of "total memory" is about permanent memory that follows you everywhere.
We're not there yet. But the race is on.
Projects in AI Tools: Like Having a Shared Notebook (But Not Quite a Magic One)
Imagine you're working on a big project—like planning a family trip or writing a book. You want all your notes, documents, and ideas in one spot so the AI can remember them no matter which chat you're in. That's what Projects (or Spaces/Collections) promise: a central place to organize chats, share files, and keep important details handy across conversations.
It sounds like having one big, smart notebook where everything connects automatically.
The simple truth: It helps a lot with organization and files, but it doesn't make the AI remember every past chat perfectly across different threads. Each chat is like a separate page in the notebook—some pages share the same reference section (files), but they don't automatically copy notes from other pages.
Here's an easy-to-read breakdown (February 2026 reality for regular users):
Quick Comparison Tables
Table 1: Basic Features
Table 2: Quick Reality Check – Projects in AI Tools
The File Upload Trap: A Common Headache (Pain Point Box)
Many people hit this wall and get frustrated—here's why in plain words:
You create a Project → Upload File A and File B at the start → Everything's great in your first few chats.
Then in Chat #3 you add File C (either attach it mid-chat or upload it to the project later).
What you hope: File C magically appears in all future chats.
What often happens (especially in ChatGPT):
The old chats (like Chat #1 and #2) usually don't see File C right away.
Why? When a chat starts, it "takes a snapshot" of the project's files at that moment. Later additions don't update old chats automatically.
To use the new file? You often have to start a brand-new chat—but then you lose the full history from the old one.
It's like adding a new chapter to your notebook after some pages are already written—the old pages don't flip to include it unless you rewrite or start fresh.
Claude handles this better because its Knowledge Base is designed for always-on access, and it uses smart search (RAG) to pull what it needs without reloading everything.
Workarounds That Actually Help (Simple Tips)
Upload everything upfront — Add all files you think you'll need to the project before creating chats. (This avoids 90% of the pain.)
Use summaries as bridges — At the end of a good chat, ask the AI: "Summarize our key decisions and facts so far." Copy that summary → Paste it into new chats or upload as a file to the project.
One long chat vs. many short ones
One long chat → Keeps perfect memory but eventually gets slow or hits limits.
Many short chats → Fresh starts, but you repeat yourself unless files/summaries help.
Pick the right tool for your job
Need rock-solid document access across many sessions? → Claude Projects.
Just want folders + some style rules? → ChatGPT Projects.
Quick searches grouped by topic? → Perplexity Spaces.
Bottom Line
Projects are super useful for keeping things tidy and sharing files/instructions—but they're not a perfect shared brain that remembers every detail from every chat forever.
They're more like a shared filing cabinet: Great for references (files, rules), okay for light connections (some memory in ChatGPT/Claude), but you still manage continuity yourself with summaries or careful planning.
No tool fully solves "I wish the AI just remembered everything without me repeating"—yet. But Claude comes closest for document-heavy work, ChatGPT is solid for everyday organization, and Perplexity fits best for fast fact-finding.
Try a small test project in each one with your real work. You'll quickly see which feels like the helpful notebook you need!
🛠️ Practical Strategies: Making Context Windows Work for You
1. Custom Instructions: Your Secret Weapon
Instead of relying on Projects, use Custom Instructions (available in ChatGPT Plus, Claude Pro):
Set your role, preferred output style, and key context once
It applies to every conversation automatically
Doesn't consume your context window
More reliable than Projects in ChatGPT
When to use Projects vs Custom Instructions:
Custom Instructions: Consistent behavior across all chats
Projects: Specific knowledge base (documents, code, research) you need to reference repeatedly
2. The Context Reset Decision Framework
Signs you should start a fresh conversation:
AI starts giving generic responses
It contradicts information from earlier in the chat
You've uploaded multiple large documents
The conversation has veered into multiple unrelated topics
Signs you should continue:
Building on a specific analysis
Iterating on a document or code
The entire conversation history is relevant context
Pro tip: Before starting fresh, ask the AI to summarize key decisions or information. Copy that summary into your new chat.
3. Images & Multi-Modal Context: The Hidden Token Eater
When you upload an image to an AI chat, it doesn't just "see" the image. It converts the image into tokens.
Rough estimates:
A simple screenshot: 500-800 tokens
A detailed image: 1,000-2,000 tokens
A high-resolution photo: 2,000-4,000 tokens
What this means: If you're on ChatGPT Free (8K token limit) and upload two detailed images (4K tokens), you've used half your context window before typing a word.
Strategy: Use images strategically. If you're having a long conversation, upload images only when necessary. Describe simple concepts with text instead.
📱 Mobile vs Desktop: The Context Trap
Important reality check: Mobile apps for AI chatbots often have reduced functionality compared to browser versions.
Common mobile limitations:
Shorter effective context windows
Fewer model options
Limited file upload capabilities
Simplified interfaces that hide advanced features
For Indian users (often mobile-first): If you're working on complex tasks, use the browser version whenever possible. The mobile app is great for quick queries, but serious work needs the full desktop experience.
Exception: Claude's mobile app maintains feature parity better than ChatGPT's.
The Conversation Graveyard Problem
Here's a pain point almost every AI user faces: hundreds of orphaned conversations with no way to find what you need.
You remember having a great conversation about market analysis last month. But which one? You've got 200+ chats. No search. No organization. Just endless scrolling.
Projects were supposed to solve this. But as we've seen:
ChatGPT Projects don't actually share context
You still need to manually organize
Search within Projects is limited
Current best practice:
Name your conversations descriptively (most platforms let you rename chats)
Use Projects/Collections as folders, not as context-sharing magic
Export important outputs to external documents
Accept that AI conversations are ephemeral – treat them like phone calls, not email archives
The future: We need better conversation management tools. This is where AI platforms should innovate next.
Key Takeaways: Context Windows in Practice
✅ Five Honest Bullets for AI Users
Free tiers hit context limits in 10-20 exchanges. Plan accordingly. Don't expect marathon conversations without upgrading.
Non-English languages consume 2-3x more tokens. If you're working in Hindi, Tamil, or other Indian languages, your effective context window is significantly smaller than advertised.
ChatGPT Projects don't auto-share context across chats. Use them as folders with custom instructions, not as magic memory solutions.
Images eat massive chunks of context. A couple of screenshots can consume half your available window on free tiers.
Marketing claims don't equal usable reality. Gemini's "1 million tokens" means little if the web interface caps you at 32K.
The Memory Problem, Solved (With Realistic Expectations)
Context windows aren't going away – they're fundamental to how AI works today. But now you know:
Why AI suddenly "forgets" mid-conversation
Which platforms give you the most working memory
How to work around the limitations
Why your language choice affects your experience
The invisible wall is still there. But at least now you can see it coming.
As AI evolves toward Sam Altman's vision of "total memory," these limitations will fade. Until then, treat AI conversations like whiteboard sessions: brilliant in the moment, but save the important stuff before you erase the board.
Looking Ahead: Part 9 Preview
Context windows determine how much AI can remember within a conversation. But what about between conversations? And across platforms?
In Part 9, we'll tackle the questions that matter for serious AI users:
Who actually owns your conversation data?
What happens to your chats when you delete them?
How do you choose the right AI tool for sensitive business information?
What are the real privacy trade-offs?
Practical governance for AI tools – because understanding how your data is used is just as important as understanding how AI thinks
📚 Read More from the AI Realities Series
Part 1 - 2026: The Year We Stop Asking If AI Works, and Start Asking If We're Using It Right
Part 3 - AI, Charts, and the Meaning Gap
Part 5 - Precision Prompts: How to Set Clear Guardrails for Professional AI Workflows
Part 6: Why AI Thinks Differently - The Shift from Rules to Probability
Part 8: Context Window Reality(You are here)
Part 9: The Privacy Problem – Choosing AI Tools for Sensitive Work (Coming soon)
📥 Download & Share
[Download PDF Version] - Save this guide for offline reference
Share this article: Help fellow AI users understand why their chatbots "forget" mid-conversation!
🔗 Twitter | LinkedIn | WhatsApp
🤝 Connect with Kannan M
Radha Consultancy | Chennai, India
AI Trainer | Management Consultant | Author
🌐 Blog: radhaconsultancy.blogspot.com
💼 LinkedIn
🐦 Twitter
📘 Facebook
🎥 YouTube: Radha Consultancy Channel Subscribe to our channel
📧 Contact us
📞 send request via contact us blog form
#AIContextWindows #ChatGPTMemory #AIReality #ContextWindowLimits #AIForgets #TokenLimits #ChatGPTvsClaudevGemini #AIProjects #AIMemoryProblem #UnderstandingAI
