Imagine waking up one morning to a message on your phone claiming that a radiation leak has occurred at a nuclear facility in Ghana.
There is a photograph. Perhaps there is even a voice recording that sounds like a senior official warning people to stay indoors. Within minutes, the message is being forwarded from one WhatsApp group to another.
Is it real? Is it old? Who can confirm it?
For Ghana, as the country advances its nuclear power programme, such a scenario is not simply a question for scientists and emergency authorities. It is also a test of journalism—and of the public trust that will be needed long before any nuclear emergency occurs.
The age of artificial intelligence has made that challenge more complicated.
At a recent virtual training on responsible AI in nuclear journalism hosted by Nuclear Power Ghana, participants were reminded that nuclear reporting carries consequences beyond getting a story wrong. Coverage can influence how people respond to warnings—whether they take appropriate action, panic, ignore a genuine warning or act on false information.
For journalists, this means that speed cannot come at the expense of verification.
AI can make the newsroom faster. It can help a reporter understand unfamiliar nuclear terminology, work through lengthy documents, prepare interview questions, transcribe interviews, compare records and monitor emerging claims.
But the same technology can also manufacture convincing falsehoods.
Generative AI can produce fabricated facts, invented quotations, false citations and synthetic photographs, audio and documents that may look and sound authentic. A fluent answer from an AI system is therefore not evidence.
The training’s central message is simple: AI output is a starting point, not a source. Any claim generated or surfaced by AI must be traced to a document, dataset or identifiable human source and independently verified before it becomes part of a published story.
That discipline becomes particularly important when information about radiation begins circulating online.
A screenshot showing an unusual radiation reading, for example, may look alarming to someone unfamiliar with radiation measurements. AI can help a journalist understand what the numbers and units mean and prepare questions. It cannot establish from a screenshot that a leak has occurred.
The responsible approach is to check the original monitoring record and seek confirmation from the regulator, the operator and an independent scientist. If the facts cannot be established before deadline, the journalist should report only what is confirmed and explain what remains unknown.
But verification is only half of the job.
How the verified information is communicated can determine how the public receives it.
A session on nuclear risk communication highlighted an important distinction: a hazard is something with the potential to cause harm, while risk concerns the likelihood that harm will occur as a result of exposure to that hazard.
For the ordinary reader, however, nuclear risk is rarely just a calculation.
Radiation cannot be seen, felt or detected by human senses. The possible consequences can be serious, while memories of nuclear accidents and weapons continue to influence public perceptions. Trust in institutions, familiarity, fear, uncertainty, media attention and concerns about children and future generations can all shape how people interpret nuclear information.
That is why simply telling people that “there is no risk” may not be enough.
It may sound reassuring, but it does little to answer the questions people actually have: What happened? What does the reading mean? Who checked it? What is the possible impact? What precautions are being taken? What is still uncertain?
Good nuclear journalism should not dismiss those fears. Neither should it turn fear into evidence.
Instead, it should translate complicated science into language people can understand while preserving the facts, the limits of the evidence and the uncertainty surrounding an event.
The same principle applies to nuclear institutions.
Sweeping assurances such as declaring a technology “100 per cent fail-safe” may sound comforting, but they can undermine credibility. Clear explanations of safety systems and how they are independently assessed can provide a more credible basis for public confidence.
For journalists, maintaining that confidence does not mean becoming cheerleaders for Ghana’s nuclear programme.
Nuclear power can be reported as a national development issue while still being subjected to rigorous scrutiny. Journalists must ask difficult questions about safety, regulation, financing, siting, environmental impacts and accountability, while giving institutions and experts a fair opportunity to respond.
That requires more than waiting for a crisis before establishing relationships with sources.
Newsrooms need to know who to call when a rumour breaks. Regulators, nuclear institutions, scientists and other credible experts need channels through which accurate information can reach the public quickly.
AI can assist with monitoring the information environment, including tracking emerging narratives and synthetic media. But journalists still have to establish what is true. The recommended verification chain is clear: AI output, a named source, primary evidence, independent checking and human editorial judgment.
Institutions, too, have a responsibility to communicate before they are under pressure.
Risk communication is more than issuing statements. It involves listening to public concerns and engaging people in dialogue. The nuclear risk communication training captured the principle succinctly: trust is built before you need it.
That may be Ghana’s most important lesson as its nuclear ambitions move forward.
When the inevitable rumour, disputed claim or frightening image eventually appears on someone’s phone, public reaction will depend partly on what people have learned to expect from the institutions and journalists they rely on.
Technology may power a nuclear plant, but when uncertainty strikes, trust will determine whom people believe.
By: Christian Kpesese
