How to spot AI-generated content
No method is 100% reliable, and automated detectors produce documented false positives. Useful clues: factual inconsistencies, visually aberrant details, lack of primary sources, recent publication accounts. The AI Act will require labeling of synthetic content, with general application in August 2026.
No method is one hundred percent reliable, and automatic detectors produce documented false positives and false negatives. Detection therefore relies on a body of clues, not a single verdict. Useful signals include factual inconsistencies, aberrant visual details, lack of primary sources, and the publication context. European regulation will also require the labelling of artificial content, with general application expected by August 2, 2026, which will make the task easier without making it unnecessary.
Clues in text
Smooth but empty generalities, no precise and verifiable examples, very regular structure, uniformly neutral tone, claims without sources. A generated text tends to sound good without committing to anything verifiable. Also be wary of precise but unverifiable references, a possible sign of fabrication.
Clues in images
Aberrant details in hands, fingers, teeth, ears. Illegible or distorted text within the image. Inconsistent reflections, shadows, and perspectives. Backgrounds that warp. These clues are fading as technology improves, but remain useful today.
Clues in video and voice
Imperfect lip synchronization, strange blinking, unnatural transitions. Voices with overly regular intonation or missing breaths. Here too, quality is increasing rapidly, calling for caution.
The most reliable reflex
Do not just try to detect the machine: verify the information. A piece of content, whether generated or not, stands by its source and its truthfulness. Trace back to the origin, cross-check, date it. Fake content is often betrayed less by its form than by the absence of a solid source behind it.
What the law changes
European regulation requires that artificial content be labelled, notably images, sounds, and videos that are generated or manipulated. This labelling obligation will help the public, but does not eliminate the need for vigilance around non-compliant content.
Frequently Asked Questions
Are automatic detectors reliable?
Not absolutely. They make mistakes both ways. Do not base an accusation on their result alone.
Does a watermark guarantee the origin?
It helps, but may be absent on non-compliant content or removed. It is not absolute proof.
How to protect my loved ones from fakes?
By sharing these reflexes and reminding them to verify the source before believing or sharing.
ICIA Resource
ICIA offers MentivisOS ICIA, a free lifelong platform to learn AI with a personalized learning path.
Discover MentivisOS ICIA