AI-generated content is no longer a novelty in news media — it is infrastructure. Dozens of major publishers now use large language models to produce first drafts of earnings reports, sports recaps, and weather updates. Some do so transparently; many do not. At the same time, synthetic voiceovers are replacing human narrators in video news packages, and AI image generation is being used to illustrate stories where no photograph exists. The volume of AI-assisted content has grown faster than the disclosure norms designed to label it.

Recognising AI involvement in text often comes down to a set of structural signals. AI-generated prose tends toward a particular kind of fluency: highly consistent paragraph length, a preference for hedged or summary-style claims, an absence of the syntactic irregularities that characterise human writing under deadline pressure. It rarely makes the kind of idiosyncratic word choices a journalist makes from lived experience. Transitions tend to be explicit ("Furthermore," "In conclusion,") rather than implied. If a piece reads as though it could be describing any similar event — competent but oddly interchangeable — that is a meaningful signal.

Synthetic voiceovers in video present a different set of cues. AI voice models still struggle with prosodic naturalness: the pauses, emphasis shifts, and micro-corrections that human speakers produce organically. Breath patterns are often absent or artificially inserted at regular intervals. Emotion is frequently applied uniformly across a segment rather than varying with content. When a voiceover sounds authoritative but oddly affectless — technically correct but emotionally flat — it warrants a closer look at the production credits.

Lenstrum's AI Involvement tab analyses any URL for AI-generation probability. Try it free →