I’m on my way back from a shoot. Hands on the wheel, my head churning with fresh thoughts and impressions from the location, so I switch on a chatbot and simply talk to it. I tell it what I saw, try out the first sentences of the piece out loud, list what I still need to check. The bot writes it all down, organises the notes and comments on my ideas as we go. Sometimes a conversation like this produces a rough draft, sometimes just a list of observations that helps me with further work on the piece. Before, by the time I got to the newsroom, many fleeting but very good ideas had simply slipped away. When I had to come back to a subject a few days later, I had to rediscover a lot of it.
Because I have used artificial intelligence almost every day since ChatGPT came out, I find it hard to relate to the media debate about AI, that abstract discussion of ethics and hallucinations. Not because these aren’t problems – they are, and very serious ones – but because the debate too often lacks balance. Lack of understanding and ignorance mix with a sense of superiority over a machine that keeps getting things wrong, and with fear of a malicious algorithm that threatens us by spreading disinformation. Mocking, derisive narratives alternate with sounding the alarm, but none of these narratives brings us any closer to understanding what the problem really is.
A model doesn’t make things up because it wants to deceive us, but because it has to answer. What it answers comes from pure probability. If it’s raining, the pavements are… that’s right, 99% wet. There is also a tiny chance that the next word will be “slippery” or a few other words, but the task is simple. When we ask a model for Michael Jordan’s detailed statistics from the 1997 finals, or for something that is typically niche knowledge in medicine or law, the model will also try to answer in the way that seems most plausible to it. But if it doesn’t have that knowledge, it may answer wrongly. Instead of the exact depth of a lake or the height of a building, it will give an approximate value or the average for that class of objects. Without exact knowledge, a language model will plausibly guess statistics based on the player’s results it knows, or a plausible-sounding diagnosis. Once the artificial intelligence has picked the first token, the rest is downhill. The coherence of each next token with the previous one makes a smooth narrative, and that is what it is rewarded for. This coherence is dangerous, because it looks like the truth, but it isn’t.
From a journalist’s point of view, that’s a nightmare. If you don’t understand how this technology works, what it can and cannot do, using AI in journalistic work is a straight road to trouble and to losing your credibility. At best, a waste of time. You want to speed up research, but in the end you spend twice as long on it, because you have to verify every sentence, every fact the assistant found. It doesn’t have to be that way, though, if you use AI wisely – to look for inspiration, to map sources or to pull structured information out of the many-page documents on your desk. AI models really can make journalistic work easier and simpler, if we know when they can be trusted and then take the time to verify ourselves.
In research, AI is bringing about a real revolution. Open-source intelligence (OSINT) tools, until recently fiendishly expensive and reserved for the biggest intelligence agencies or market giants, have been radically democratised. Anyone can build an agent for their own needs that will extract and structure the key data from many-page documents in seconds. Final fact-checking always stays with a human, but simply locating in an instant where a specific piece of information sits in terabytes of digital data cuts an analyst’s working time many times over, and now it is also within reach of every journalist. We must remember, though, that even top-shelf tools used by governments and armies can be as brilliant as they are deadly dangerous when human scepticism is missing. A few weeks ago the world was a step away from a war between great powers when a military analyst trusted a civilian chatbot that combined classified data with OSINT. The model hallucinated a report claiming that a Chinese merchant ship was carrying nuclear weapons for Iran; fighter jets and boarding teams were already in the air, and the strike was called off at the last moment only because a human spotted the error in the machine’s reasoning in time. On the other hand, we hear that it was advanced AI models, integrating intelligence reports, drone data and satellite images in real time, that enabled the Americans to carry out the daring operation to capture Nicolás Maduro. Without this technology, such a complex operation would probably have been impossible to carry out without disproportionately high costs.
Naturally, in journalism too, the temptation quickly arises to go a step further and hand more and more work to artificial intelligence. If the machine can read, let it write the text itself. My attitude to this is mixed, because for now artificial intelligence simply cannot do for me what matters most in making a piece. No model can take a handful of raw interviews and arrange them into a sensible television “ladder” – a precise structure in which the voices of the people in the story weave naturally with the author’s narration and respond intelligently to the footage that has been shot. Trying to give an agent enough context to generate a script close to what I expect would currently take work out of all proportion to the final result. Perhaps it is a matter of time and of further iterations of the models, but so far I haven’t found a tool that could capture this level of complexity. AI copes quite well with short news items and press releases – provided someone checks them afterwards – but in longer forms I almost always sense a lack of individual style and flow.
On the other hand, I am not a dogmatic opponent of using artificial intelligence in creative work. Writing, let’s be honest, is partly a purely mechanical skill. What really counts is the idea, the original view and what the author has to tell us. A good, unconventional idea and original thinking will sooner or later break through the layer of form. I know all too many journalists who have had nothing new to tell us for years, and whose texts look deceptively like ready-made pieces generated by an algorithm – with no living idea in them at all. They lack that particular spark, that unconventional intelligence and human flash that the mathematics of language models still cannot capture.
When I read an article, I usually know within the first few sentences whether its content resonates with me. To be honest, whether it was written by a human or a machine matters less to me than one question: does this text bring new value into my life? Can it change the way I see a problem? Besides, contrary to popular belief, it is very hard today to tell human text from machine text categorically at the level of syntax alone. Testing AI-text detectors, I quickly found that perfectly structured, logical and rhythmic columns by well-known commentators, written as far back as 2013, are sometimes flagged by these programs as one hundred per cent machine-made. It works both ways, too. The ZeroGPT detector judged a passage of the US Declaration of Independence to be almost 98% the work of a machine. So I consider condemning texts solely on the basis of how they sound to be largely pointless.
As a reader, I judge the value of a text, not its author, but public trust rests on who answers for the text. That’s why I understand the opposite, purist approach: the view that, for ethical reasons, journalism should remain a machine-free zone, as an area of special public trust. The only thing is that such a solution would require a firm social contract and mechanisms that would unquestionably guarantee a certificate of human authorship. Such mechanisms still do not exist: the existing certificates of human authorship are voluntary and have not become a standard in the media.
In the end, even though journalism is listed today near the top of the professions most threatened by AI, I believe that in the age of artificial intelligence real journalists will do quite well. First, because sooner or later a moment of systemic reckoning will come – the value of human control over information will be taken seriously enough to get the right legal regulation. The first rules already exist, though they go in a different direction: from 2 August 2026, the EU AI Act requires AI-generated texts on matters of public interest to be labelled, unless they have been through editorial review and someone takes responsibility for them. So the law asks who is responsible, but gives the reader no way of checking who actually wrote it.
Second, our skills scale perfectly in the world of technology. Working on my own app and on training a model made me realise one key thing: creating, both in technology and in the media, is in my view 90% a process that we journalists mastered long ago. It is a repeated cycle of two things: research and iteration. You gather information. You verify. You synthesise. You improve. And you start again. For people who want to learn and to understand the reality around them, artificial intelligence opens up great possibilities. One capable journalist supported by agentic models can do today what yesterday took a whole editorial team. That is a real chance to increase public scrutiny of the processes going on in the world.
Besides, journalism and artificial intelligence are linked by something that lies very deep at the foundations of how they work: attention. In AI it is precisely the attention mechanism – described in the groundbreaking scientific essay “Attention Is All You Need”, with which the current revolution began – that has made modern models so effective. Machines learn how to weigh and connect what matters most in a thicket of data. Reporters do exactly the same. Journalism is largely the ability to pick the key information out of a wide context in an instant and put it in the lead, often in extremely difficult, chaotic conditions at a press conference or on a shoot. Models try to imitate this process and are getting better and better at it, but in the end they hit a wall: no access to physical, tangible reality. Algorithms can only analyse what has already been written or digitised. We, thanks to trained attention, can interpret the multidimensional world around us as it happens and decide in a split second what really matters to our audience. That is why it will be very hard for machines to overtake us in this field. It also means that journalists can find themselves perfectly at home in a new role – as experts who understand AI, catch its shortcomings and correct its hallucinations. Evaluating and analysing content in RAG systems, and in the future designing, checking and testing whole agentic systems or the models themselves, may turn out to be a very natural development path for many of us, and a fascinating professional pivot.
So who will lose their jobs? Above all, those for whom journalism long ago stopped being a mission and became a matter of repeating templates. People who write almost the same text on the same subject every year. I don’t see a single reason why any newsroom should employ them when an algorithm will do the same faster and for free. I have no illusions that the market always weeds out the derivative. Sometimes it weeds out the cheapest, or those who happened to be standing nearest the door. In the long run, though, competition from machines will sweep from the board, above all, those who reject development and refuse to learn. Of course, it is hard to predict the direction of such a fast-developing technology. But I believe that if artificial intelligence is used in journalism ethically – with the natural and overriding right of people to reliable information in mind – it will, paradoxically, strengthen our profession in the end. It will become a tool that does not replace people but multiplies their agency. In a world where producing information is getting cheaper and cheaper, a journalist’s value is shifting from producing text itself towards recognising reality, establishing what is true and taking responsibility for what gets published.
When I come back from a shoot, I talk to the bot and dictate the first sentences to it. It will write them down, fix the commas, perhaps suggest a better word order. But I was the one on the shoot, and it is still me who decides what goes into the piece and what the viewer sees on screen. The time saved on boring, repetitive sentences, which AI helps me automate, I can give to the people in my reportages.