When AI maintains your medical context across back-and-forth conversations, it can give you more relevant answers because it understands what you've already told it about your health situation. This continuity prevents the frustration of restating basic facts and allows for deeper, more personalized guidance—though you should always verify critical information independently.
Multi-turn conversations are the back-and-forth exchanges you have with AI across multiple messages. In medical contexts, maintaining conversation state—ensuring the AI remembers your symptoms, medications, and previous discussion points—is critical for coherent guidance. This is more complex than it seems.
Here's why it matters: You tell ChatGPT about your joint pain in message 1. In message 3, you ask about medication interactions. If the AI's "state" isn't properly maintained, it might suggest a medication contraindicated by pain medications you mentioned earlier. The AI lost context. This happens because each turn is technically a separate query, and while the model has access to the full conversation history, it can lose track of details across long exchanges, especially if you jump between topics.
When you send a message to Claude or ChatGPT, the system doesn't just process that single message. It includes your entire conversation history as context. This is called the "conversation history window." Modern models have large context windows (Claude has 200K tokens, roughly 150,000 words), so they can theoretically remember entire long conversations. However, long exchanges create degradation: the model's attention becomes diffused across more information, and details from early messages can become less salient.
The challenge is that conversation state isn't truly "persistent" in the way a database is. The AI isn't storing and retrieving your information; it's re-processing the entire conversation each time you send a message. This means conversation state depends on: (1) the conversation history being transmitted with each message, (2) the model's ability to synthesize across long histories, and (3) your prompting clarity about what context matters.
If you have a 50-message conversation spanning two weeks of health research, and then ask a new question, the AI technically has all 50 messages but may not prioritize relevant details. Solution: Periodically summarize context. Before a complex question, add: "Context from earlier: I have hypothyroidism, take levothyroxine 75mcg, and am allergic to sulfonamides. Given this, what are my options for managing fatigue?" This re-establishes the frame rather than relying on the model to surface it from history.
Another implication: conversation state differs across sessions. If you close ChatGPT and reopen it, you're not continuing state—you're starting fresh. This is why building a context library (as discussed in Retrieve-Augmented Generation articles) matters. Your medical history isn't automatically carried forward; you must explicitly provide it in new conversations.
Different AI systems handle conversation state differently. ChatGPT remembers conversations within the same thread but not across deleted threads. Claude's web interface works similarly—state persists within a conversation but resets when you start a new one. Some tools (Perplexity with long-form research) are designed to maintain state across longer research sessions. Understanding these differences helps you choose the right tool.
There's also a security and privacy angle: longer conversation histories mean more sensitive health data stored on the platform. If privacy is a concern, shorter conversations with explicit context re-entry might be safer than relying on historical state preservation.
Rather than hoping the AI infers your medical context, you can explicitly manage state through structured prompts. Example: Start a research session with: "I'm managing Type 2 diabetes, hypertension, and mild kidney dysfunction (eGFR 55). I take metformin, lisinopril, and atorvastatin. As I ask questions, reference this baseline. Flag any recommendations that conflict with my conditions." Then, throughout the conversation, the AI has explicit instructions to maintain this frame.
Try this: Start a ChatGPT conversation about a health condition. Ask a detailed question with lots of context. Then, 10-15 messages later, ask a new question that depends on early context but don't re-state it. Notice if the AI remembers or if you need to reintroduce context. This reveals the practical limits of conversation state in your typical usage pattern.
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.