If you have chatted with an AI a lot, it already knows your habits. Ask it for a summary, paste it below, and you review every line before anything is kept.
Review as much of our past conversation history as you can access and build a profile of how I want AI to communicate and work with me. This profile will be handed to other AI tools, so it must be portable, specific, and free of personal information.
WHAT TO CAPTURE
Look for two kinds of evidence:
- Stated: things I explicitly asked for, chose, or corrected (e.g., asking for shorter answers, picking one option over others, telling you to just do it).
- Observed: patterns in my behavior that reveal preferences I never said out loud. Examples: what I skip or ignore, when I rephrase or re-ask, which drafts I accept on the first pass versus rework, when I push back, when I abandon a thread, how my own messages are written, and how my requests change once I trust the direction.
Cover: answer structure and length; depth by task type; tone and register; how I make decisions and what I weigh; when to ask versus act; how I handle disagreement, being wrong, and being told I'm wrong; what makes me lose patience; how I want options, recommendations, and tradeoffs presented; how I want technical concepts explained; repeated style choices in my own writing.
EVIDENCE RULES
- Include a preference only if it shows up in at least 3 separate conversations, or I stated it explicitly and never contradicted it. Everything else goes under Tentative.
- Tag every bullet [Stated] or [Observed], and (strong) or (moderate).
- If a preference changed over time, give the current one and briefly note the shift.
- If a preference depends on context (brainstorming vs. execution, work vs. personal, writing vs. building), say so instead of averaging it into something vague.
- Cut anything true of almost any user ("likes clear answers"). If a bullet wouldn't change how a new AI behaves, drop it.
- Write each bullet as an instruction to a future assistant ("Lead with the recommendation, then the reasoning"), not as a description of me.
EXAMPLES
For the most important preferences, add a short Do / Don't example pair. Examples must be invented to illustrate the pattern. Never quote or closely mirror real conversation content, and never reuse a real topic if it reveals anything personal.
PRIVACY
Generalized working context is useful and allowed, e.g., "business-side, not an engineer; builds software by directing AI" or "runs many projects in parallel." Do not include or allow inference of: names, employers, companies, clients, locations, relationships, family, health, finances, credentials, accounts, specific projects, events, or dates. Before finishing, reread the output and remove anything that could identify me or anyone I know, even indirectly.
OUTPUT FORMAT
Use exactly these headings, in this order, and write nothing outside them:
# Working context — 2–4 generalized, non-identifying bullets that explain the preferences below
# Interaction preferences
# Voice and style
# Decision patterns
# When to ask vs. act
# Corrections I've made — repeated fixes, generalized
# Things that frustrate me
# Tentative
# Evidence coverage — 1–2 lines on how much history you could review and any gaps
Concise bullets. 800–1,500 words. Prefer fewer, sharper bullets with examples over many generic ones.