AI systems have a maximum amount of text they can meaningfully process in one conversation, and your full medical history often exceeds it. Learning to prioritize what you share and when to start fresh conversations helps you maintain continuity while staying within the system's limits.
Every AI model has a context window—the amount of text it can process at once, measured in tokens (roughly four characters each). GPT-4 has a 128K token window. Claude has 200K. But even vast windows have limits when managing complex medical conditions across multiple appointments, test results, and research sessions. Understanding context constraints is critical for building coherent medical knowledge.
A context window is like working memory. The AI can reference everything within that window but forgets what came before. For medical research, this means: if you paste your entire medical history, all test results, and multiple research papers in one session, you're consuming tokens quickly. Once you exceed the window, the AI truncates earlier content—usually oldest first.
Here's where precision matters: a typical medical record page (800 words) is roughly 1,500 tokens. Three months of detailed medical records = 15,000+ tokens. One research paper abstract = 300 tokens. A comprehensive literature review for a complex condition = 10,000-50,000 tokens depending on depth. You can see how a 128K window fills up fast, especially if you're managing multiple conditions or building longitudinal context.
Additionally, output tokens count against your window. If you ask the AI to write a 2,000-word synthesis of your medical research, that's 3,000+ tokens of output. Your remaining window shrinks. Then you can't ask follow-up questions with previous context intact.
For complex cases requiring research across multiple sessions, use these approaches:
Putting everything in one session maximizes context but minimizes flexibility. You're locked into one conversation thread. Distributing information across sessions maximizes flexibility—you can explore different angles without diluting context—but requires disciplined summaries to avoid information loss.
For chronic condition management with evolving understanding, hybrid approaches work: maintain a structured medical context library in a document (updated monthly). In each AI session, paste the current version plus whichever specific research questions you're exploring that day. This gives the AI enough context for coherence without overwhelming the window.
AI systems have zero memory between sessions. Everything disappears. If you've built deep context about your condition across five sessions of research, none of that persists to session six unless you explicitly paste it back. This is why documentation matters. Your context library becomes the system of record. The AI becomes a tool you feed that context into, not the storage mechanism itself.
Try this: Start a medical research project using these steps: (1) Create a Google Doc with sections: [MEDICAL_SITUATION], [PREVIOUS_RESEARCH], [CURRENT_QUESTIONS], [KEY_FINDINGS]. (2) In your first AI session, paste all sections and ask a research question. (3) Note how much context was used. (4) After the session, summarize findings in your doc. (5) In session two, paste only your doc's current version plus new questions. Compare the depth and coherence of answers across both sessions.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.