Is it AI? How things are made today, and how to tell

Two habits have grown up around AI. One is to look at anything polished, a poster, a song, an essay, and say "that is the work of AI". The other is to paste a piece of writing into an AI detector, a tool that estimates whether text or an image was made by AI, and treat its score as a verdict. Both habits are understandable. Both mislead.

The first gets how things are made wrong. Most work today is made by people who decide what to make, use AI as one tool among several, and choose what to keep. Our guide to whether something is the work of AI shows how, and tells the story of South Africa's draft national AI policy, withdrawn in April 2026 because its references had been invented by AI and published without anyone opening them. The tool drafted. People skipped the check.

The second habit fails for a different reason. Most text detectors guess from how predictable the words are, and plain, careful writing looks predictable. That is a problem for anyone who writes in English as a second language, which describes millions of students and workers in our region. In one Stanford study, seven detectors labelled essays by non-native English writers as AI-written 61.3% of the time on average. Our guide to whether AI detectors work sets out the evidence, and what a teacher, an employer or an accused student can do instead.

For pictures and video there are newer signals. A watermark such as Google's SynthID is hidden in images some AI tools make, and Content Credentials, under the C2PA standard, attach a signed record of how a file was made. Both are useful when present, and neither proves anything when absent, because screenshots and forwarded copies usually lose them.

So the questions that work on everything are about people, not tools. Who decided what this says? Who checked it? Who answers for it if it is wrong? A message with no name, no source and no one to answer for it deserves doubt whether a person, a machine or both wrote it. A piece of work with clear answers deserves trust whatever tools helped make it.

Most of the time, whether AI helped make a poster or a post matters little. Three situations are different. In an election, a convincing fake picture or voice clip shared on the eve of a vote can do damage before anyone checks it. In health, a confident message about a cure or a vaccine, forwarded from a group, can change what people do. And in schools and universities, a wrong accusation of using AI can follow a student for years.

In each of these, slow down before you share or decide. Look for the earliest source, ask the three questions, and give the person accused the chance to show their work. Our guides set out how, step by step.

Do AI detectors work? How to tell if AI made it
Why an AI detector score is a guess, who it wrongs, and the checks that work better.

Is it the work of AI? How things are really made today
Why "AI made it" is rarely true, and the three questions to ask instead.