Baidu Ernie Bot Caught Surfacing Unverified Negative Brand Info

Small Brand Test Exposes Big AI Accuracy Gap

Baidu's Ernie Bot is under fire after a small business owner discovered it was surfacing unverified, defamatory content about their brand — while competitors like OpenAI's GPT, Google Gemini, and ByteDance's Doubao returned accurate, brand-aligned information. The incident highlights a growing concern: how AI models handle reputation data for small businesses with limited online footprints.

The founder of HUPPO, a niche fragrance brand based in China, decided to test how various AI chatbots perceived their brand. The results were stark — and sparked a heated discussion on V2EX, one of China's largest developer forums.

GPT and Gemini Got It Right — Baidu Did Not

When queried about HUPPO, most major AI models returned responses that closely matched the brand's actual identity and values. Ernie Bot, however, told a very different story.

Here is how each model performed:

• OpenAI GPT — Returned responses consistent with the brand's positioning and philosophy
• Google Gemini — Delivered accurate brand information aligned with the founder's stated mission
• ByteDance Doubao — Provided a faithful summary drawn from verified online sources
• Baidu Ernie Bot — Prioritized unverified negative posts from complaint forums over legitimate brand information

The founder reported that Baidu had also removed their brand's encyclopedia entry (Baidu Baike, China's equivalent of Wikipedia), further compounding the problem by eliminating the most authoritative source of brand information from its own ecosystem.

The Origin: A $0.70 Dispute Turned AI Nightmare

The negative content Ernie Bot surfaced traced back to a single customer dispute over roughly $0.70 in shipping fees. A buyer purchased a fragrance sample set for about $1.50 (10.9 yuan), with $0.70 (5 yuan) allocated to shipping. After requesting a return, the platform auto-refunded only the product cost.

The buyer then escalated the situation dramatically — filing complaints on Xiaohongshu (China's Instagram equivalent) and Black Cat Complaints (a consumer platform), posting what the founder describes as fabricated defamatory claims. The buyer ultimately received a platform-mediated payout and kept the samples at zero cost.

Despite providing evidence, the brand owner spent months fighting to get the false posts removed. This type of 'one-post defamation' is notoriously difficult for small businesses to combat.

Why This Matters Beyond China

This case raises urgent questions for any business operating in an AI-driven information landscape. Large language models are increasingly becoming the first point of contact for brand discovery, and their training data choices have real commercial consequences.

Key takeaways for Western businesses and AI developers:

• Training data quality matters — Models that ingest unmoderated complaint forums without verification risk amplifying defamation
• Small businesses are disproportionately vulnerable — A single bad-faith post can dominate a thin data profile
• Platform ecosystem lock-in creates compounding risks — Baidu removing the brand's wiki entry while its AI surfaces complaints creates a one-sided narrative
• AI-powered brand perception is the new SEO battleground — Companies may soon need 'LLM optimization' strategies alongside traditional search optimization

A Warning Sign for the AI Search Era

As AI chatbots replace traditional search for millions of users, the stakes for data accuracy are rising fast. Google and OpenAI have invested heavily in source attribution and grounding techniques to reduce hallucinations and misinformation. Baidu's Ernie Bot, despite being one of China's most prominent LLMs, appears to lag in filtering unverified consumer complaint content.

For small brands worldwide, the lesson is clear: your AI reputation is now as important as your Google ranking — and potentially harder to control.