Trust in the news hit a record low. The least-trusted source is the one growing.

Thirty-seven percent. That's how many people across 48 global markets told the Reuters Institute they trust the news this year, and it's the lowest number they've recorded since they started asking in 2015. It held at 40% for three years and then gave way. Trust fell in 29 of those 48 markets, by five points or more in 19 of them. In the United States it's 25%, and 15% among right-leaning Americans.

Now the second number. Twenty percent of people trust the answers they get about the news from an AI chatbot. That's the lowest score of any channel measured, below social media at 22%, and in the UK it bottoms out at 6%.

That's also the channel growing fastest. Weekly use of chatbots for news went from 7% to 10% in a year, and to 16% among people under 35.

Both numbers are true at once: people are moving fastest toward the channel they believe the least, and that gap is where the next few years of the news business get decided.

The people arriving are the ones you'd least expect

It's easy to assume casual news consumers are drifting toward whatever's convenient while serious readers stay put. The data says the opposite.

Among people using AI chatbots for news, 38% fall into the Reuters Institute's "news lover" segment. Across everyone surveyed, that figure is 22%. The people adopting the channel they trust least are disproportionately the people who care most about news.

So these adoptors are your best readers. The ones who'd notice if you got something wrong, showing up somewhere they've told a survey they don't believe what they read.

Meanwhile the ground underneath has already moved. For the first time, social media and video networks are globally the most widely used way to get news, ahead of news organizations' own sites, at 54%. News avoidance sits at 42%. Interest in news overall has dropped about thirteen points since 2021.

Comfort tracks familiarity

Public skepticism is the defense most teams are counting on.

In 2024 the Reuters Institute asked how comfortable people were with news made mostly by AI. Overall the answer was 19%, against 36% for news made by humans with some AI help, which is the human-in-the-loop premium and it's real. But split that by how much people had seen or heard about AI, and comfort with mostly-AI news runs at 26% among the more exposed group and 13% among the less exposed. Twice as high.

The pattern shows up across markets too. India reports 18% weekly chatbot news use and 44% comfort with automatically generated news. The UK reports 3% use and 11% comfort. The report's own language is that there's "a clear correlation between levels of comfort with automatically generated news and reported usage of AI chatbots."

Correlation is the right word here. Comfort isn't trust, and this doesn't establish which way the arrow runs. People already relaxed about AI probably reach for it sooner, which would produce exactly this picture without exposure changing anyone's mind about anything.

But one side of that relationship is only moving in one direction. Exposure goes up. Nobody encounters less AI next year than they did this year. If comfort and familiarity travel together at all, whatever protection you think skepticism is providing is on a clock.

Skepticism is a fuel supply. Convenience burns it, and nothing in this picture refills the tank.

What survives the summary

Most teams are asking whether to use AI in the work, and how to disclose it when they do.

The question this data creates is what happens to your credibility after it's been through a channel you don't control and your audience doesn't trust but uses anyway.

Someone asks a chatbot about a thing you published. The answer comes back stripped of your layout, your standards page, your correction history, your masthead, and most of your byline. What reaches the reader is a paraphrase, and the trust they extend to it is the 20% number. Whatever you earned on your own site stayed there.

You can't fix that by being trustworthy in the abstract. Trustworthiness that only exists inside your own four walls doesn't survive the trip.

Two things do

The first is being verifiable inside the channel people are actually using. That means checkable: durable URLs that still resolve, real timestamps, visible revision history, a correction record kept in public with the original error still readable beside it. Receipts a machine can point at and a person can land on. Every publisher asserts its reliability, and an assertion is the first thing paraphrase throws away.

Among people using chatbots for news, 42% say they always or often click through to the original source. In South Korea it's 56%. That's not nothing. That's a substantial slice of the most engaged audience in the survey arriving at your front door with a question, having read a version of you that you didn't write. What they find when they get there is entirely yours to decide, and it's the only moment in the chain you still control.

The second is network proof. When the algorithm summarizes you, the thing that carries is who vouches for you and whether that vouching survives translation. Named people with something at stake, cited by other named people, in ways that leave a trail outside your own domain. A summary can flatten your prose. It has a harder time flattening a decade of other people's receipts pointing at you.

Neither of those is new advice. What's new is the reason. This used to be good hygiene. It's now the thing your credibility rests on, because the distribution layer has stopped carrying it for you.

The honest size of it

I'd rather not overstate the numbers. They size the problem, they don't settle it. A single year's drop from 40% to 37% is a drop, not a collapse, and the trust figure has moved before. Ten percent of people using chatbots for news is still ten percent. Pew found half of Americans expect AI to make the news worse and only one in ten expect it to help, so the skepticism is genuine and currently strong.

Strong and currently are the words doing the work. What the exposure gradient suggests is that today's strength is a reading, and readings move.

The audience is migrating toward sources they say they don't trust, because those sources are easy, and the version of you that arrives there is a paraphrase with your name on it. Build so that the paraphrase points somewhere real, and so that people who'd know are willing to say so out loud.

Trust that can't travel through the new distribution layer is trust that decays.