AI-generated content

The label covers anything a generative model produced from a user's instructions: articles and social posts, product photos, voice recordings, video clips. Content is often mixed, with a person drafting and a model editing, or a model drafting and a person revising, and there is no agreed threshold at which something stops being "human" content. Disclosure rules differ by platform and jurisdiction, so the term describes how something was made and says nothing on its own about whether it is accurate or acceptable.

Synthetic media is the subset that depicts or imitates real things: a cloned voice, a realistic image of an event, a video of a person. When the aim is to make an audience believe a person said or did something they did not, it is a deepfake. Much synthetic media is harmless or disclosed, such as illustrations, dubbing and stock imagery.

A documented case: in May 2023 an AI-generated image appearing to show smoke rising near the Pentagon circulated widely on social media, including through some paid-verification accounts. Local officials said no incident had occurred and the image was debunked within minutes to hours, but it was shared before then. A hypothetical example: a manufacturer finds a realistic but fabricated photo of a damaged product being shared by customers, with no original source and no photographer.

Fabricated images, audio and quotes can reach an organization's audience before the organization has seen them. Teams need a verification routine: trace the earliest appearance, check for provenance data (see content provenance), look for an original source, and decide whether to respond publicly or wait. Detection tools exist but none is fully reliable, so a verdict rests on tracing and reporting as much as on software.

AI-generated content also feeds back into what AI systems say. Low-quality volume is covered under AI slop, and models can repeat errors in it as fact; see AI hallucination. To see how AI assistants currently describe a brand, the free AI Perceptions check samples answers from a subset of providers.

Whether something is AI-generated says nothing about whether it is accurate, and human authorship guarantees nothing either. Detectors that claim to separate the two err in both directions, including flagging real writing as machine-made, so a detector score is not evidence on its own.

Search engines do not treat AI authorship as a violation by itself. Google has said its concern is content produced at scale mainly to manipulate rankings, whoever or whatever wrote it.