What is the story about?
A new study from US-based digital marketing agency Legal Guardian Digital has attempted to answer exactly that by ranking some of the world’s most widely used AI assistants on reliability. And in a result that may surprise many users, the winner was not OpenAI’s ChatGPT.
Instead, Perplexity AI claimed the top spot, outperforming larger and more mainstream rivals including ChatGPT, Google Gemini and Anthropic Claude.
The report assessed AI chatbots using a mix of practical metrics, including hallucination rates, uptime reliability, response consistency and customer satisfaction. In short, researchers wanted to find out which chatbot was least likely to confidently invent facts while also remaining stable enough for regular everyday use.
The results suggest that smaller AI platforms may quietly be overtaking their bigger rivals in areas that matter most to users.
According to the report, Perplexity recorded the lowest hallucination rate among major AI chatbots at 13 per cent, well below the industry average of 22 per cent. The platform also maintained a 100 per cent uptime score during the testing period, meaning users reportedly experienced no service outages.
That combination helped Perplexity secure the highest overall reliability score of 85 out of 100.
Coming in second was xAI’s Grok, backed by Elon Musk. Grok reportedly achieved a hallucination rate of 15 per cent while also maintaining perfect uptime reliability.
Chinese AI assistant DeepSeek placed third with a 14 per cent hallucination rate and 99.52 per cent uptime. The report also highlighted that DeepSeek remains free to use despite outperforming several paid competitors.
Meanwhile, Claude ranked seventh in the reliability study and reportedly suffered more outages than competitors, while Google Gemini landed in eighth place. Meta AI finished ninth.
Despite remaining the world’s most recognisable AI chatbot, ChatGPT ranked sixth overall in the study’s reliability index. Researchers claimed the chatbot generated inaccurate answers in roughly 30 per cent of cases, nearly double the error rate recorded for DeepSeek.
Even so, ChatGPT continued to perform strongly in customer satisfaction, receiving a user rating of 4.7 out of 5.
The findings arrive during a period of growing scrutiny around AI trustworthiness and user fatigue. According to figures from market intelligence firm Sensor Tower, US uninstalls of ChatGPT’s mobile app surged by 295 per cent on February 28 compared with the previous day. That represented a dramatic jump from the chatbot’s usual day-to-day uninstall rate of around 9 per cent over the previous month.
Austin Hunt, chief executive of Legal Guardian Digital, said many users still assume ChatGPT is automatically the best AI assistant because it entered the market early and became the most famous. But according to the study, popularity alone may no longer be enough to guarantee trust.
Instead, Perplexity AI claimed the top spot, outperforming larger and more mainstream rivals including ChatGPT, Google Gemini and Anthropic Claude.
The report assessed AI chatbots using a mix of practical metrics, including hallucination rates, uptime reliability, response consistency and customer satisfaction. In short, researchers wanted to find out which chatbot was least likely to confidently invent facts while also remaining stable enough for regular everyday use.
The results suggest that smaller AI platforms may quietly be overtaking their bigger rivals in areas that matter most to users.
Perplexity leads on accuracy and uptime
According to the report, Perplexity recorded the lowest hallucination rate among major AI chatbots at 13 per cent, well below the industry average of 22 per cent. The platform also maintained a 100 per cent uptime score during the testing period, meaning users reportedly experienced no service outages.
That combination helped Perplexity secure the highest overall reliability score of 85 out of 100.
Coming in second was xAI’s Grok, backed by Elon Musk. Grok reportedly achieved a hallucination rate of 15 per cent while also maintaining perfect uptime reliability.
Chinese AI assistant DeepSeek placed third with a 14 per cent hallucination rate and 99.52 per cent uptime. The report also highlighted that DeepSeek remains free to use despite outperforming several paid competitors.
Meanwhile, Claude ranked seventh in the reliability study and reportedly suffered more outages than competitors, while Google Gemini landed in eighth place. Meta AI finished ninth.
ChatGPT still dominates popularity, but concerns remain
Despite remaining the world’s most recognisable AI chatbot, ChatGPT ranked sixth overall in the study’s reliability index. Researchers claimed the chatbot generated inaccurate answers in roughly 30 per cent of cases, nearly double the error rate recorded for DeepSeek.
Even so, ChatGPT continued to perform strongly in customer satisfaction, receiving a user rating of 4.7 out of 5.
The findings arrive during a period of growing scrutiny around AI trustworthiness and user fatigue. According to figures from market intelligence firm Sensor Tower, US uninstalls of ChatGPT’s mobile app surged by 295 per cent on February 28 compared with the previous day. That represented a dramatic jump from the chatbot’s usual day-to-day uninstall rate of around 9 per cent over the previous month.
Austin Hunt, chief executive of Legal Guardian Digital, said many users still assume ChatGPT is automatically the best AI assistant because it entered the market early and became the most famous. But according to the study, popularity alone may no longer be enough to guarantee trust.
