Technology · 22 March 2026

Three Prompts Slash Claude's AI Hallucinations Dramatically

People using Claude AI can make its answers more accurate. A Redditor discovered three simple system prompts, listed in official Anthropic documents, that reduce the model's tendency to create false information, or "hallucinations." These instructions involve allowing the AI to say "I don't know" and requiring it to verify facts using direct quotes and citations, though this may reduce creative output.

By BriefTea Editorial · 22 March 2026 at 5:29

Reported by XDA · How we write briefs

Get the app

More briefs

WorldTurkey, Saudi Arabia and Pakistan to discuss new mutual protection deal on Monday WorldIran's unseen supreme leader urges Gulf countries to fight a shared enemy BusinessIceland supermarket boss tells the Chancellor to cut business taxes for employers GeneralNigel Farage's £5m donor has major stakes in arms industry PoliticsTreasury officials advise John Healey to tax banks and oil companies more HealthCorrecting hearing loss can cut dementia risk by almost half Browse all stories →