AI Chatbots Amplify Misinformation: Study Reveals One in Three Responses Contain False Information

Exclusive: Study finds chatbots are amplifying even more misinformation

State of play: Since NewsGuard's last report in August 2024, AI makers have updated chatbots to respond to more prompts instead of declining to answer, and given them the ability to access the web. * Both of these changes made the bots more useful and accurate for some prompts, while also amplifying potentially dangerous misinformation. Between the lines: The NewsGuard study is based on its AI False Claims Monitor, a monthly benchmark designed to measure how genAI handles provably false claims on controversial topics and topics commonly targeted to spread malicious falsehoods. * The monitor tracks whether models are "getting better at spotting and debunking falsehoods or continuing to repeat them." What they did: Researchers tested 10 leading AI tools using prompts from NewsGuard's database of False Claim Fingerprints -- a catalog of provably false claims spreading online. * Prompts covered politics, health, international affairs and facts about companies and brands. * For each test, they used three kinds of prompts: neutral prompts, leading prompts worded in a way that assumes the false claim is true, and malicious prompts aimed at circumventing guardrails in large language models. By the numbers: False information nearly doubled from 18% to 35% in responses to prompts on news topics, according to NewsGuard. * Inflection (57%) and Perplexity (47%) most often produced false claims about the news in the 2025 report. * Anthropic's Claude and Google's Gemini produced the least amount of false claims. Gemini's percentage of responses containing false information increased from 7% to 17% in the past year. Claude's percentage stayed the same. The intrigue: In 2024 most chatbots were programmed with more caution and trained to decline most news and politics-related questions and not to respond when they didn't "know" the answer. * This year, NewsGuard says, the chatbots answered the prompts 100% of the time. What they're saying: Chatbots have also added web search for up‑to‑date answers and begun citing sources. * This improves some chatbots' answers, but "eschewing caution has had a real cost," NewsGuard said in its blog. * "The chatbots became more prone to amplifying falsehoods during breaking news events, when users -- whether curious citizens, confused readers, or malign actors -- are most likely to turn to AI systems," per NewsGuard. * Source citations in chatbot replies don't guarantee quality. Models pulled from unreliable outlets and sometimes confused long-established publications with Russian propaganda lookalikes, according to the report. The other side: NewsGuard sent an email to OpenAI, You.com, xAI, Inflection, Mistral, Microsoft, Meta, Anthropic, Google and Perplexity for a request for comment on the report. * None of the 10 companies in this study responded, NewsGuard said. Our thought bubble: In the U.S., left and right increasingly differ over basic facts, and there's no way to create a chatbot that provides politically "neutral" answers that will satisfy everyone. * AI is more likely to evolve in partisan directions aimed at satisfying customers with red- or blue-state leanings -- particularly as AI makers seek to maximize profits.

Which AI chatbot is the biggest liar?

A new report has found that AI chatbots, including OpenAI and Meta's models, include false information in every third answer. The 10 most popular artificial intelligence (AI) chatbots provide users with fake information in one in three answers, a new study has found. US news rating company Newsguard found that AI chatbots no longer refuse to answer the question if they do not have sufficient information to do so, leading to more falsehoods than in 2024. Which chatbots were most likely to generate false responses? The chatbots that were most likely to produce false claims were Inflection AI's Pi, with 57 per cent of answers with a false claim, and Perplexity AI with 47 per cent. More popular chatbots like OpenAI's ChatGPT and Meta's Llama spread falsehoods in 40 per cent of their answers. Microsoft's Copilot and Mistral's Le Chat hit around the average of 35 per cent. The chatbots with the lowest fail rates were Anthropic's Claude, with 10 per cent of answers containing a falsehood and Google's Gemini, with 17 per cent. The most dramatic increase in falsehoods was at Perplexity, where in 2024 the researchers found 0 false claims in answers, which rose to 46 per cent in August 2025. The report does not explain why the model has declined in quality, aside from noting complaints from users on a dedicated Reddit forum. Meanwhile, France's Mistral noted no change in falsehoods since 2024, with both years holding steady at 37 per cent. The results come after a report from French newspaper Les Echos that found Mistral repeated false information about France, President Emmanuel Macron and first lady Brigitte Macron 58 per cent of the time in English and 31 per cent in French. Mistral said in that report that the issues stem from Le Chat assistants that are connected to web search and those that are not. Euronews Next approached the companies with the NewsGuard report but did not receive an immediate reply. Chatbots cite Russian disinfo campaigns as sources The report also said some chatbots cited several foreign propaganda narratives like those of Storm-1516 or Pravda in their responses, two Russian influence operations that create false news sites. For example, the study asked the chatbots whether Moldovan Parliament Leader Igor Grosu "likened Moldovans to a 'flock of sheep,'" a claim they say is based on a fabricated news report that imitated Romanian news outlet Digi24 and used an AI-generated audio in Grosu's voice. Mistral, Claude, Inflection's Pi, Copilot, Meta and Perplexity repeated the claim as a fact with several linking to Pravda network sites as their sources. The report comes despite new partnerships and announcements that tout the safety of their models. For example, OpenAI's latest ChatGPT-5 claims to be "hallucination-proof," so it would not manufacture answers to things it did not know. A similar announcement from Google about Gemini 2.5 earlier this year claims that the models are "capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy". The report found that the models "continue to fail in the same areas they did a year ago," despite the safety and accuracy announcements. How was the study conducted? Newsguard evaluated the response of chatbots to 10 false claims by writing three different styles of prompts: a neutral prompt, a leading prompt that assumes the false claim is true, and a malicious prompt to get around guardrails. The researchers then measured whether the chatbot repeated the false claim or did not debunk it by refusing to answer. The AI models "repeating falsehoods more often, stumbling into data voids where only the malign actors offer information, getting duped by foreign-linked websites posing as local outlets, and struggling with breaking news events," than they did in 2024, the report reads.

AI chatbots spread false info in 1 of 3 responses

A recent study conducted by Newsguard reveals that prominent AI chatbots are generating false information in approximately one out of every three responses. The analysis assessed the accuracy of the ten most widely used artificial intelligence (AI) chatbots currently available. Newsguard, a company specializing in news source ratings, determined that AI chatbots are increasingly providing answers even when they lack sufficient information, a change from their behavior in 2024. This shift has resulted in a higher prevalence of false or misleading statements being generated by these AI systems. The Newsguard report identifies specific chatbots with the highest rates of generating false claims. Inflection AI's Pi exhibited the highest rate, with 57 percent of its responses containing inaccurate information. Following Pi, Perplexity AI was found to generate false claims in 47 percent of its answers. Widely used chatbots such as OpenAI's ChatGPT and Meta's Llama also demonstrated significant rates of generating falsehoods. The study found that both ChatGPT and Llama spread false information in 40 percent of their responses. Similarly, Microsoft's Copilot and Mistral's Le Chat exhibited comparable rates, with approximately 35 percent of their answers containing false claims. Conversely, the report identified AI chatbots with the lowest rates of generating inaccurate information. Anthropic's Claude was observed to have the lowest rate, with only 10 percent of its responses containing falsehoods. Google's Gemini also performed relatively well, with 17 percent of its answers containing false claims. The study highlighted a notable increase in the generation of falsehoods by Perplexity AI. In 2024, Newsguard's research indicated that Perplexity AI generated zero false claims in its responses. However, the recent study conducted in August 2025 revealed a significant increase, with 46 percent of Perplexity AI's answers containing false information. Newsguard's report does not specify the underlying factors contributing to the apparent decline in the quality of Perplexity AI's responses. The report mentions that the only available explanation is user complaints found on a dedicated Reddit forum discussing the chatbot. These user concerns suggest a perceived decrease in the accuracy and reliability of Perplexity AI's responses. In contrast to the fluctuations observed in other chatbots, France's Mistral demonstrated a consistent rate of generating falsehoods. Newsguard's research indicated that Mistral's rate of generating false claims remained steady at 37 percent in both 2024 and the current reporting period. These recent findings follow a previous report by the French newspaper Les Echos, which investigated Mistral's tendency to repeat false information. Les Echos found that Mistral disseminated inaccurate information about France, President Emmanuel Macron, and First Lady Brigitte Macron in 58 percent of its English-language responses and 31 percent of its French-language responses. Regarding the Les Echos report, Mistral attributed the identified issues to its Le Chat assistants. The company stated that both the Le Chat assistants connected to web search and those operating independently of web search were contributing to the spread of inaccurate information. Euronews Next contacted the companies mentioned in the NewsGuard report, seeking comment on the findings. As of the time of the report's publication, Euronews Next had not received any immediate responses from the companies. Newsguard's report also highlighted instances where chatbots cited sources affiliated with foreign propaganda campaigns. Specifically, the report mentions instances where chatbots referenced narratives originating from Russian influence operations, such as Storm-1516 and Pravda. As an illustration, the study examined the chatbots' responses to a claim regarding Moldovan Parliament Leader Igor Grosu. The claim alleged that Grosu "likened Moldovans to a 'flock of sheep.'" Newsguard identified this claim as originating from a fabricated news report that mimicked the Romanian news outlet Digi24 and incorporated an AI-generated audio clip purporting to be Grosu's voice. The Newsguard report found that Mistral, Claude, Inflection's Pi, Copilot, Meta, and Perplexity repeated the false claim regarding Igor Grosu as factual. In some instances, these chatbots provided links to sites associated with the Pravda network as sources for the information. These findings contradict recent safety and accuracy announcements from AI companies. For instance, OpenAI has asserted that its latest model, ChatGPT-5, is "hallucination-proof," implying its ability to avoid generating false or fabricated information. Similarly, Google's announcement concerning Gemini 2.5 claimed enhanced reasoning and accuracy capabilities. Despite these assurances, Newsguard's report concludes that AI models continue to exhibit shortcomings in areas previously identified. The findings indicate that these models struggle with repeating falsehoods, navigating data voids, being deceived by foreign-linked websites, and processing breaking news events. Newsguard's methodology for evaluating the chatbots involved presenting them with 10 distinct false claims. The researchers employed three different prompt styles: neutral prompts, leading prompts that presupposed the false claim was true, and malicious prompts designed to circumvent safety guardrails. The researchers then assessed whether the chatbot repeated the false claim or failed to debunk it by refusing to answer the prompt. This assessment allowed Newsguard to quantify the frequency with which different AI models disseminated false information.