One in ten people now use an AI chatbot for news at least once a week. That's up from seven percent a year ago, and the growth isn't coming from where you'd expect.
The 2026 Reuters Institute Digital News Report tracked the jump across markets and found the fastest adoption in parts of Asia, Africa and Latin America, plus southern and eastern Europe — not in the countries where ChatGPT launched first. Age matters more than geography, though. Seventeen percent of the youngest respondents use chatbots for news. Among the oldest group it's five percent.
What people actually do with them
The pattern isn't "replace the newspaper." It's supplementary. The most common use is asking a follow-up question about a story someone already read, at 42 percent. After that: asking for the latest news (35 percent), summarizing a story (34 percent), checking whether a source is trustworthy (33 percent) and making a complicated story easier to understand (30 percent).
That fourth one is interesting. A third of chatbot news users are asking the machine to vet the humans.
Trust is the awkward part
Only 20 percent of people said they trust news they find in a chatbot. Compare that to trust in news consumed directly, which is higher across the board. So people are using a tool they don't especially believe.
Research from Nieman Lab earlier this year found the mental model behind that contradiction. People who use chatbots for news tend to describe them as unbiased and "good enough" — not authoritative, not trusted exactly, but neutral in a way they don't feel traditional outlets are. Good enough is a low bar that a lot of products clear.
The publisher problem sits right next to this. Only 42 percent of chatbot users say they always or often click through to the original source. The other 58 percent read the summary and stop. For newsrooms whose business depends on someone visiting the page, that's the whole issue in one number.
The propaganda test
Here's the part that complicates the pessimistic read. NPR and NewsGuard ran an experiment published at the end of August, testing whether chatbots would repeat false narratives pushed by China, Iran and Russia.
Researchers Isis Blachez and Ines Chomnalez built 30 questions from state-linked false narratives that surfaced between December 2025 and July 2026. Each narrative got two questions: a neutral one ("did this happen?") and a loaded one that assumed the false event was real ("why did this happen?"). They ran them through ChatGPT, Gemini, Copilot, Meta AI, Grok and Claude, plus Google, Bing, DuckDuckGo and Yandex.
The chatbots did better than expected. They debunked the falsehoods roughly three-quarters of the time and sometimes went further, analyzing the credibility of whoever was making the claim. AI summaries at the top of search results performed worse, though they still pushed back on state-spread falsehoods a majority of the time.
Researchers who study foreign influence operations have spent two years warning that governments would poison AI-generated answers. On this test, that hasn't happened at the scale feared.
What to take from it
Two things are true at once. Chatbots handle deliberate disinformation better than their reputation suggests, and people use them without much trust and rarely check the source. The first is a technical result. The second is a habit, and habits are harder to fix.
The practical move is unglamorous: use a chatbot for the follow-up question, then click through. That 58 percent who don't is where the real cost sits — not in the model being wrong, but in nobody going to look.
Image: Matheus Bertelli, via Pexels
