System prompts set standing instructions that persist across all conversations, while session prompts apply only to the current exchange; for neurodivergent workflows, system prompts prevent you from re-explaining your processing style every single time, creating consistency that actually supports sustained effort. It's the difference between repeatedly onboarding a new assistant versus having one who knows how you work.
A system prompt is foundational instruction that applies across all conversations and models—your 'constitution' for how an AI should behave with you. A session prompt is context-specific instruction for a single conversation about a single goal. Think of system prompts as rules of engagement; session prompts are task briefs.
For neurodivergent users, this distinction matters enormously because consistency is a form of cognitive scaffolding. If the AI behaves unpredictably—sometimes verbose, sometimes terse; sometimes flag-happy, sometimes silent—you burn executive function just adapting to the variation. System prompts create guardrails that make the AI's behavior predictable across contexts.
A system prompt defines core parameters: 'When I ask questions, ask me clarifying questions before answering. Flag when I'm contradicting myself. Keep responses concise unless I explicitly ask for detail. If I'm overcomplicating something, tell me directly.' These rules apply everywhere you deploy the system—every conversation, every project, every AI tool (if you're using multiple tools and want consistency).
System prompts are especially valuable for people with ADHD because they prevent the 'wall-of-text' problem. If you're sensitive to information overload, you establish once that you want concise responses, and you don't have to keep negotiating that in every conversation. Similarly, if you tend to hyperfocus on implementation details before clarifying requirements, a good system prompt prevents the AI from enabling that pattern. It becomes a structural safeguard rather than something you have to self-regulate.
The strongest system prompts are specific to your neurotype. For example, an ADHD-specific system prompt might include: 'I often task-switch. Help me recognize when I'm drifting from my stated goal. Use specific language ('You said you were focusing on X; we're now discussing Y') rather than implicit redirection.' A dyslexic learner's system prompt might include: 'Provide visual structures for information (lists, tables, hierarchies). Spell out acronyms on first reference. Use sans-serif fonts when possible.' An autistic learner's system prompt might emphasize: 'Be explicit rather than implied. Don't use idioms. State social expectations clearly.'
A session prompt is the context you provide for a specific conversation: your immediate goal, current constraints, the documents or information relevant to this session, and how you want output formatted. Session prompts are where you layer in real-time information. They're more flexible and change with each conversation.
The critical insight: session prompts should build on system prompts, not contradict them. If your system prompt says 'keep responses concise,' your session prompt shouldn't then ask for 'exhaustive explanations.' This consistency prevents cognitive whiplash and makes the AI's behavior predictable.
This is where it gets technical: not all AI tools support explicit system prompts in the user interface. ChatGPT doesn't expose system prompts to free users, though paying users can create custom GPTs with persistent system instructions. Claude allows you to set system prompts directly (in the API or via character/persona setup). Gemini doesn't have explicit system prompt functionality, so you have to encode your system-level instructions in the opening message of every conversation, which is inefficient.
A workaround: create a 'master prompt' document in Notion or Google Docs that contains both your system rules and session-setup templates. At the start of each conversation, copy the relevant section into your first message. It's not as elegant as persistent system prompts, but it creates consistency across tools.
For power users, if you're using Claude via API or ChatGPT via API, you can integrate system prompts programmatically. But for most neurodivergent users managing ADHD, the Notion-template approach is simpler and more maintainable.
Your system prompt isn't fixed forever. Every month or two, review whether it's actually achieving what you intended. Are you still getting wall-of-text responses despite the 'be concise' instruction? Refine it. Clarify the rule: 'Responses under 150 words unless you explicitly ask for more detail.' Did the AI stop flagging task-switch drift? Maybe the rule wasn't specific enough. Revise: 'If my response indicates I'm now discussing something different from my opening goal, explicitly state: You originally wanted to [original goal]. We're now discussing [current topic]. Is this intentional?'
The best system prompts are iterative. After 10-15 conversations, you'll have enough data to refine the rules based on what actually works for your brain.
Try this: Write down 5 consistent frustrations you have with AI responses (e.g., 'too verbose,' 'doesn't ask clarifying questions,' 'sometimes forgets my preferences mid-conversation'). For each frustration, write one specific rule that would prevent it. Now compile those rules into a short 'system constitution' (150-200 words max). Paste this at the start of your next 3 conversations with ChatGPT or Claude. After those 3 conversations, review whether the rules actually worked. Refine the wording where they didn't. This refined version becomes your persistent system prompt.
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.