In this latest post in the FAQ series, I am sharing some responses to a radio interview about AI’s impact on journalism.

Q: Is the continuous growth of AI-generated content online a danger for journalism?

It is certainly a problem yes, in three ways: it makes reporting harder, it makes it harder to support journalism financially, and it makes it harder for audiences to trust your reporting.

It makes reporting harder because a journalist has to deal not only with a lot more misinformation but also a lot more information full stop.

It makes the business model more difficult because there’s more competition from AI-generated slop.

And that’s compounded by the fact that audiences are less likely to trust your reporting — either because they might dismiss it as AI generated, or because they might have already begun to believe AI misinformation that your reporting contradicts.

Q: Is the industry equipped to handle it?

I don’t think any industry is equipped to handle how AI is changing the information environment, and I think it will take some time for society to adjust to a world of unlimited cheap and unreliable information.

But journalism is probably in a better position than most industries to handle that change because it has rules and workflows that a lot of other industries do not.

For example, journalists are trained to research both sides of the story and seek a right of reply, and to find more than one source for key facts. A high value is put on speaking to real people and capturing raw footage of events. Reporters are encouraged to be sceptical and apply verification and factchecking skills, and their stories go through editors who are supposed to look at the story dispassionately. These are not qualities that all industries have.

Also, the industry has been exploring applications of artificial intelligence for over a decade, so there’s technical knowledge there too which many organisations will lack.

The biggest challenge is the collapse of the financial basis for journalism and I think there will have to come a point where more public or charitable funding is used to support it.

Q: When AI can fabricate convincing news content at scale, who is ultimately responsible for the integrity of what we read and watch online?

Primarily it’s whoever created that content. When the printing press came along someone might have asked who is responsible for these books that are being published? And of course we know the answer now is the authors and the publishers.

But technology companies are also responsible, because AI is not a neutral tool like a printing press.

ChatGPT is a collection of recipes for predicting text, and so OpenAI, which owns ChatGPT, has responsibility for the way it has written those recipes because it can change them. It has changed them, and does change them, regularly.

If you try to create racist or other offensive content in ChatGPT or Gemini or Claude, for example, it will generally refuse to do so. If you create images in Gemini a watermark is added, so it can be identified as AI generated. If you ask Claude for medical advice it will recommend consulting a health professional. These are editorial choices by the technology company.

So while individual crimes involving fabrication might be tackled by identifying and charging the creator, the ultimate responsibility for any systemic social harm will lie with technology companies who haven’t considered key risks and taken steps to prevent that. And there will need to be regulation and enforcement to ensure that happens.

Q: What does this mean for the public’s trust in journalism?

We have always had to earn that trust, and that has become more and more important as competition for the public’s attention has increased.

Part of that trust is about building understanding: not only explaining how we actually do journalism, and why we do it that way, but also listening to our audiences to understand what they need and why.

In order for the public to trust journalism they have to see how it is distinct from other sources of power, and holds that power to account. They have to see how it is on their side, rather than looking down on them.

Q: When fabricated content goes viral, what happens to people’s ability to make informed decisions?

I think we already had a problem with people being able to make informed decisions, as it became easier to find information that confirmed a person’s existing beliefs rather than information that challenged those.

Ultimately the more misinformation there is, the more likely it is that people are either making decisions based on flawed information, or unable to make decisions because they can’t be 100% sure a certain piece of information is true.

I think we are going to learn some quite painful lessons in the next few years because most people think they won’t fall for this stuff, and yet most people do. I am an intelligent person who trains people in verification and factchecking, and I have retweeted information that is not true.

There are two key things that we need to remember: firstly, that you will fall for misinformation at some point, and probably already have, that is certain.

Secondly, slow down and look for counter-evidence.

Q: For an ordinary person watching the news or scrolling their feed, what practical steps can they take right now to check the veracity of what they are seeing?

There’s a great piece of research in psychology that finds when we look at an image of a crowd of faces, we pay attention to the angry faces more: our eyes spend more time on those.

Our brains are designed to pay more attention to threats, and social media algorithms have learned this, so they prioritise material that triggers anger and fear because we spend more time looking at it, and are more likely to react to it.

So the first practical step people can take is to slow down. When we come across new information two parts of our brain processes it: the first part is fast and instinctive. It makes a decision whether to pay attention and what to do with the information. That’s the part that shares or likes or comments on a chat message or video.

Don’t share. Don’t like. Don’t react.

The key thing is to let the information move through to a second phase, the part of the brain that actually processes that information rationally.

Now, ask yourself if that update is triggering some core emotion. Is it confirming something you believed? Is it making you angry or afraid? Why? The short answer is: because it keeps you on that site or app for longer. But is it true?

The main way to check veracity isn’t to look for clues that something is fake. An image or video might be real but from a different time or place. Something might be factually true but misleading because it’s missing context. Trusted friends and authority figures will share this stuff, so don’t use that as a primary signal.

The main way to check veracity is to look for evidence that challenges it.

For example you can use a website called TinEye to see where an image has appeared before (you can also freeze frame videos and take screenshots). You can put the information into Google with the word ‘hoax’ or ‘factcheck’ to see if it’s already been debunked.

You can try to identify the original source, rather than whoever passed it on to you. That might give you clues about their independence, expertise or agenda.

Q: And people can use AI tools to verify as well?

Using AI to verify material is very dangerous. We’ve seen a big increase in the last few years of people using AI tools to check if an image or a video is real, and very often they get it wrong.

Part of the reason is that people ask the wrong question. If you ask AI “is this fake?” then you have to remember tools like ChatGPT and Grok are sycophantic, they are designed to please, so they will say “yes” even if they’re not certain. They are also overconfident.

The key thing with AI in general is to always ask a neutral question. Ask it what evidence there might be to help you identify whether something is true or false. Ask it what methods you can use.

AI is best used to challenge you and open up other ideas and techniques. Don’t use it to confirm things because it will just tell you what you want to hear.

Never rely on AI’s answer as the end of a process – it can only ever point you to next steps. Remember that it is not a factual tool, it is a language prediction tool. So it can predict the language of a piece of advice around verifying something, but it cannot know if something is real or fake, because that’s not what AI is designed to do.