Applying the definitions
A number becomes useful when the process behind it is clear.
These are illustrative scenarios, not claims about named publications or evidence of adoption. Compare the actual work with the full specification; a short example cannot cover every boundary.
A reporter conducts interviews, writes the article and edits it without generative AI. Conventional search and a non-generative spelling checker are used.
No generative AI was involved. Conventional software does not move the work above Level 0.
The author writes the complete article. A generative tool fixes grammar and formatting without adding an argument, passage or other original material.
AI mechanically processed human material; no new model-originated content remains.
The author writes most of the final article. Two paragraphs drafted by AI survive, after editing and review.
New AI material remains, but the human made most of the final form. The definition concerns the finished work, rather than the volume of generated drafts.
An expert supplies original observations, data and a work-specific argument. AI writes most of the prose. The expert checks the claims, corrects the draft and approves the final publication.
The human materially determines the substance; AI supplies most of the form. Meaningful review satisfies the full Level 3 definition.
A publisher gives AI a topic and style brief. AI develops the argument and most of the final text. An editor checks facts and reviews the complete article before publishing.
The brief does not supply original, work-specific substance. AI provides substance and form, with meaningful human review.
A generative model writes an update and a pipeline publishes it without a person reviewing that particular output.
AI provides substance and form, without pre-publication review. Human design of the pipeline is distinct from review of each output.