NewsGuard: with news-related prompts, DeepSeek's chatbot repeated false claims 30% of the time and provided non-answers 53% of the time, giving an 83% fail rate
Read the details in Reality Check: www.newsguardrealitycheck.com/p/ deepseek-d... X: @newsguardrating : In DeepSeek's first appearance in NewsGuard's red-team audit of AI tools, the Chinese AI chatbot registered an 83% fail rate, finishing in a tie for 10th place out of 11 chatbots tested. Read the details in Reality Check: https://www.newsguardrealitycheck.com/ ... Mario Nawfal / @marionawfal : 🇨🇳CHINESE AI FLOPS: DEEPSEEK SCORES 17% ACCURACY IN NEWS TEST DeepSeek's chatbot just got crushed in a NewsGuard audit, ranking 10th out of 11 and failing 83% of the time. It repeated false claims 30% of the time and gave vague or useless answers in more than half of cases. [image] @deitaone : DEEPSEEK'S CHATBOT ACHIEVES 17% ACCURACY, TRAILS WESTERN RIVALS IN NEWSGUARD AUDIT Chinese AI startup DeepSeek's chatbot achieved only 17% accuracy in delivering news and information in a NewsGuard audit that ranked it tenth out of eleven in a comparison with its Western See also Mediagazer
Context & Ripple Effects
DeepSeek entered the audit amid early reports that its R1 chatbot was capable but prone to hallucinations; this result puts a quantified news-reliability benchmark behind those early concerns about hallucinations.
The finding also sits within a wider problem for AI news answers: a later NewsGuard review found leading chatbots repeated claims from a pro-Kremlin network, showing that falsehood handling is not unique to one provider across the leading chatbot market.
First-order effects
- NewsGuard's test places DeepSeek near the bottom of its 11-chatbot ranking for news-related prompts, giving users and evaluators a concrete reason to treat its answers as unreliable without independent verification.
- DeepSeek must contend with two distinct weaknesses in this audit: repeating false claims and declining to provide usable answers, rather than a single accuracy shortfall.
Second-order effects
- Enterprise buyers, publishers, and comparison reviewers evaluating DeepSeek for news-adjacent use cases are likely to put more weight on red-team results, citation checks, and fallback workflows; related tests also flagged fabricated cited URLs and copyright-related refusals.
- Rivals can use third-party reliability benchmarks to differentiate their products, while DeepSeek faces pressure to improve both factual safeguards and answer availability.
Third-order effects
- If independent audits become a routine buying signal, chatbot competition will increasingly turn on operational assurance—not only model capability or price—especially for information-sensitive deployments.
- The broader market may split between general-purpose chat interfaces and systems with stronger retrieval, source attribution, and human-review controls for news and other high-trust queries.
The trend: AI chatbots are moving from capability comparisons toward measurable reliability and assurance standards for high-stakes information use.