AI can write News in seconds, but can it tell the truth?

The world is currently undergoing a major disruption driven by the rapid development of Artificial Intelligence, commonly known as AI. Companies that sell AI tools have become some of the most valuable corporations of modern times, with valuations running into trillions of dollars. Their worth now exceeds the GDPs of many countries.

These companies are increasingly influencing social, commercial and political life, while also transforming entire industries.

One of the sectors most affected by this shift is the media industry. Journalism, which plays a crucial role in maintaining healthy and functioning democracies, is changing in ways that are not always visible to the public.

To fully grasp how AI is shaping our information environment and influencing politics, it is necessary to understand what generative AI, or GenAI, actually is and how it functions. As experts suggest, we need to “lift the bonnet” on the systems that will increasingly power the information we see and consume.

What powers generative AI?

The foundation of generative AI is data. The development of these systems begins with collecting enormous volumes of information, including text, images, videos and audio. This data is gathered by crawling and scraping the internet, drawing from journalism, academic research, public websites and text conversations.

These datasets are often expanded using collections of literature obtained through commercial licensing agreements with media repositories, sometimes without clear legal permission.

The legality of these data collection practices remains uncertain and has sparked high-profile copyright and privacy cases across the world. It has also triggered regulatory and policy debates, alongside criticism from creative professionals whose work has become the basis for the massive revenues earned by global AI technology firms.

How is data turned into AI training material?

Access to data alone is not enough to train generative AI systems. The raw material must be converted into structured training datasets through complex computational processes combined with human labour.

To make data useful, workers are required to label, clean, tag, annotate and process text and images. This creates meaningful connections that allow GenAI models to generate responses to user prompts.

Much of this work is outsourced to lower-cost countries such as Kenya, India and China. In these regions, workers are often paid low wages and operate under poor labour conditions. Once prepared, the datasets are used to train AI models through machine learning.

How do generative AI systems actually learn?

Machines do not learn in the same way humans do. What is known as machine learning is essentially a process of statistical pattern recognition.

Although training methods vary, most involve repeated adjustments to huge numbers of internal values within a model. This process is iterative, meaning it repeats until the system’s predictions closely match expected outcomes.

After training, models such as those behind ChatGPT can respond to prompts like “write a short news story on inflation figures” by producing a sequence of tokens, or word fragments, that statistically resemble similar content encountered during training.

Crucially, these systems do not understand the world they describe. They lack semantic knowledge and cannot grasp facts or concepts such as what inflation means or what a street protest looks like.

Instead, they function as pattern-modelling tools that predict what content is most likely to fit a given prompt. In simple terms, AI outputs are the result of scale and training data rather than genuine understanding.

What does generative AI mean for journalism?

The same predictive ability that makes generative AI impressive also makes it unreliable. Prediction is not the same as verification.

These systems can quickly write fluent text, summarise lengthy documents and rephrase complex material. They can also generate images that appear highly realistic. However, these outputs are created through prediction rather than fact-checking.

When trained on biased or incomplete data, generative AI is known to produce what is often referred to as “hallucinated” content. This material may look and sound convincing but can be inaccurate or misleading.

This distinction is critical for journalism, which relies on truth, verification and accountability rather than plausibility.

Why is AI a risk to trust in information?

A major concern for journalists and audiences alike is the difficulty of verifying AI-generated content. As more such material enters the information ecosystem without clear labelling or context, it becomes harder to distinguish between genuine reporting and simulated content.

This blurring of boundaries risks creating an environment where the difference between fact and fabrication becomes increasingly unclear.

The future of journalism will depend on whether institutions can adapt to and properly govern the use of AI. This involves developing new editorial standards, improving verification practices and making the data, labour and energy behind AI systems more transparent and accountable.

The issue is no longer whether AI will reshape journalism. That transformation is already underway. The real question is whether democratic societies can prevent AI from undermining trust in public institutions.

For those concerned about where information comes from and how it is produced, human ability to check and verify content cannot match the speed at which chatbots can generate flawed text, data and images.

Unless effective systems are created to ensure oversight and checks before sharing machine-generated content, society risks further erosion of a fundamental pillar of democracy: shared facts that enable rational thinking and informed decision-making.