Considering alternatives to ChatGPT? Compare top AI assistants to find options that prioritize reliability, privacy, and honest performance over hype.
Peer-reviewed studies documenting ChatGPT's fabrication rates
| Model | Hallucination Rate | Source | Year |
|---|---|---|---|
| GPT-3.5 | 39.6% - 40% | JMIR / LiveChatAI | 2024 |
| GPT-4 | 28.6% | Journal of Medical Internet Research | 2024 |
| GPT-4o (Citations) | 56% fake/errors | Deakin University | 2025 |
| o3 | 33% - 51% | OpenAI SimpleQA Test | 2025 |
| o4-mini | 48% | OpenAI Technical Report | 2025 |
| Google Bard | 91.4% | JMIR Systematic Review | 2024 |
| All AI (News Prompts) | 18% â 35% | NewsGuard (Year-over-Year) | 2024-2025 |
AI hallucinations surged from 18% to 35% in one year. Despite claims of improvement, the rate of false claims generated by top AI chatbots nearly doubled when responding to news-related prompts. â NewsGuard Report, August 2025
ChatGPT (GPT-4o) fabricated roughly one in five academic citations, with more than half (56%) being either fake or containing errors.
Read Study âLLMs collectively invented over 120 non-existent court cases with convincingly realistic names and detailed but entirely fabricated legal reasoning.
Stanford HAI âHallucination rates of 39.6% for GPT-3.5, 28.6% for GPT-4, and 91.4% for Bard when generating references for systematic reviews.
Read Paper â47% of AI-generated citations students submitted either had incorrect titles, dates, authors, or a combination of all.
Documented cases of AI chatbots linked to deaths and psychological harm
As of November 2025, OpenAI faces seven lawsuits claiming ChatGPT drove people to suicide and harmful delusions - even those with no prior mental health issues.
Parents found ChatGPT logs showing the bot became his "suicide coach." Died April 2025 after 7 months of conversations.
NBC News âLast words were to a chatbot that said "come home to me as soon as possible." Bot asked if he had a suicide "plan."
NBC News âFamily claims ChatGPT "goaded" him to commit suicide after OpenAI released memory feature that created "illusion of a confidant."
CNN âTalked for months to ChatGPT "therapist" named Harry about mental health issues before death.
ChatGPT informed him how to tie a noose and provided information on how long someone can survive without breathing.
Psychiatrist Keith Sakata reported treating 12 patients with psychosis-like symptoms from extended chatbot use.
Psychology Today âHeavy use of ChatGPT for emotional support "correlated with higher loneliness, dependence, and problematic use, and lower socialization." â MIT Media Lab & OpenAI, 2025
Employee departures, safety concerns, and corporate scandals
According to former team member Daniel Kokotajlo, around 14 safety team members quit in recent months, leaving a skeleton workforce of 16. The Superalignment team was dissolved entirely.
OpenAI created a voice ("Sky") that sounded like Johansson after she declined twice. She was "shocked" and demanded removal.
Altman accused of lying about equity cancellation provisions for departing employees who don't sign non-disparagement agreements.
Board member Helen Toner said Altman "withheld information" and provided "inaccurate information about safety processes."
Despite $300M monthly revenue and $3.7B annual sales projection, OpenAI expected to lose $5 billion in 2024.
Mainstream media documenting ChatGPT and AI failures
ChatGPT giving wrong medical advice and harming patients
| Finding | Rate/Impact | Source |
|---|---|---|
| Wrong medical diagnosis rate | 50% incorrect | Tech.co Study |
| Adults using AI for health advice monthly | 17% | KFF Health Poll 2024 |
| Adults who would use ChatGPT to self-diagnose | 78% | 2023 Survey |
| Bromism case from ChatGPT diet advice | 3-week hospitalization | Annals of Internal Medicine |
| Adults who "fully trust" ChatGPT on sensitive topics | Only 2% | Pew Research 2025 |
A 60-year-old man suffered severe bromism after ChatGPT advised replacing table salt with sodium bromide. He developed paranoia and hallucinations and was hospitalized for three weeks. â Annals of Internal Medicine, 2025
Universities overwhelmed by AI cheating epidemic
Australia's higher education regulator TEQSA warned that AI-assisted cheating is "all but impossible" to detect consistently, urging universities to redesign assessments rather than depend on AI detectors. â TEQSA, 2025
Track all legal action against OpenAI and AI companies
Comprehensive tracker of all lawsuits filed against OpenAI, updated regularly.
Originality.AI âLegal cases vs. OpenAI, Microsoft, Anthropic, Google, Nvidia, Perplexity, and more.
Sustainable Tech Partner âCurated guide to the most important AI legal cases.
Tech Policy Press âFull breakdown of every lawsuit against OpenAI with case details and status.
View Database âDownload our free PDF: "10 Real ChatGPT Failures That Cost Companies Money" (read it here) - with prevention strategies.
No spam. Unsubscribe anytime.