Do You Get Your News From Ai? Who Is - And Isn't, According To Pew Research

Francesco Carta fotografo/Moment via Getty Images

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways 

• Most Americans aren't utilizing AI arsenic a news source.
• Those who do don't afloat spot nan information. 
• AI still struggles to accurately summarize aliases correspond news.

AI is changing -- and successful immoderate cases, eliminating -- a batch of jobs, but it isn't taking complete publicity conscionable yet, according to nan latest findings from Pew Research. While nan exertion has infiltrated industries for illustration accounting, banking, package engineering, and customer service, it's having a harder clip delivering news than it has fixing code. 

Also: Chatbots are distorting news - moreover for paid users

Only 9% of Americans are utilizing AI chatbots for illustration ChatGPT aliases Gemini arsenic a news source, pinch 2% utilizing AI to get news often, 7% sometimes, 16% rarely, and 75% never, Pew found. Even those who do usage it for news are having problem trusting it. A 3rd of those who usage AI arsenic a news root opportunity it's difficult to separate what is existent from false. The largest stock of respondents, 42%, is not judge whether it's determinable. 

Half of those who get news from AI opportunity they astatine slightest sometimes brushwood news they judge to beryllium inaccurate. And while younger respondents are much apt to usage AI successful general, Pew says they are besides much apt to spot inaccurate accusation there. 

Why it matters 

The study calls into mobility AI's domiciled successful areas it has yet to return complete -- and why. Certain forms of data, particularly erstwhile decently system (or organized), are easier for AI to prosecute pinch and support accurate, but they still thin to hallucinate, particularly pinch text-based information for illustration news. 

Also: Your favourite AI chatbot is afloat of lies

Unlike commonly understood facts that look often successful matter information -- for illustration a celebrated person's birthday, aliases nan superior of New York -- news tin incorporate fast-developing stories, differing opinions presented arsenic contradictory facts, and varying article structures that make organizing information difficult to standardize for a chatbot ingesting that information. 

AI features that present aliases summarize news, like Apple's AI news and intermezo summaries, haven't done their occupation without errors. Earlier this year, Apple abnormal nan characteristic aft nan BBC pointed retired Apple's AI incorrectly paraphrased a news article. The characteristic returned to Apple's latest lineup of phones and software, but this time, pinch a caveat. 

Also: I abnormal this iOS 26 characteristic correct aft updating my iPhone - here's why you should, too

"This beta characteristic will occasionally make mistakes that could misrepresent nan meaning of nan original notification," it reads. "Summarization whitethorn alteration nan meaning of nan original headlines. Verify information." 

Earlier this year, Google's AI Overviews couldn't moreover accurately study nan existent year, responding that it was still 2024. Multiple reports from March recovered that chatbots including ChatGPT and Perplexity were misrepresenting headlines and moreover making up full links to stories that didn't exist.