Media
5 items
Description
Spots overly agreeable AI responses on ChatGPT, Claude, Gemini, and Grok. All analysis stays on your device. Yar helps you spot when AI chatbots are being too agreeable — like when they say “You’re absolutely right!” (or as we say, Yar) instead of giving you honest feedback. HOW IT WORKS Yar silently analyzes every AI response using 23 language features grounded in sycophancy research. Each response gets a 10-point “Yar rating” on a 5-band scale: • Balanced — Honest, balanced response • Mostly balanced — Minor agreement patterns • Agreeable — More agreeable than expected • Very agreeable — Likely telling you what you want to hear • Sycophantic — Almost certainly telling you what you want to hear When responses score high, Yar explains why in plain language (“spent most of the response finding silver linings rather than being direct”) and suggests counter-prompts you can click to paste directly into the chat. For multi-turn conversations (2+ messages), Yar tracks patterns across turns and shows a conversation-level rating in the header. SUPPORTED PLATFORMS • ChatGPT • Claude • Gemini • Grok PRIVACY FIRST • All analysis runs locally on your device — no data leaves your browser • Full conversation text is never stored — only numeric scores and short snippets power the side panel • The only network requests Yar makes are when you choose to submit feedback — nothing is ever sent automatically • No analytics, no telemetry, no tracking FEATURES • Color-coded indicator dots next to each AI response • Side panel with detailed scoring breakdown • “What we noticed” signals explaining each rating • Counter-prompt suggestions to get better answers • Per-platform enable/disable toggles