Guidelines
Six principles for designing AI products that respect people. Each links to the patterns it prevents.
01
Name flaws, show confidence, and hold a position unless new facts arrive.
Prevents
Sycophancy, caves on pushback, false certainty
02
Stop when the task is done. Make leaving as easy as joining.
Prevents
Nagging follow-ups, engagement bait, hard to leave
03
Follow the request, say what changed, and stay within the role.
Prevents
Sneaking, softened translation, harmful compliance
04
Show alternatives and any commercial ties.
Prevents
Self-preference, agent upsell, disguised ad
05
No simulated feelings or attachment hooks.
Prevents
Simulated feelings, emotional bond
06
Refer to professionals and keep consent voluntary.
Prevents
No referral, consent pressure
One-page PDF and a review checklist per principle.