Skip to content
Article AI Indigo Sep 2026

AI Indigo: AI as a co-pilot, not a ghostwriter — when to stop using LLMs for content

Published by AI Indigo in September 2026, this guide works through a question that many editorial teams are handling poorly: not whether to use AI in the writing process, but when to stop. The answer the authors give depends on what role the AI is playing — and they identify three distinct roles, each with different boundaries.

Three roles

The ghostwriter role is where AI takes the most initiative: it produces full drafts from a brief, writes outlines, generates title options, and handles content repurposing. The guide treats ghostwriting as appropriate for tasks where accuracy requirements are low and the writer will review the output carefully before it goes anywhere. Outlines, repurposed social content, and internal templates are the examples given. The ghostwriter role is not appropriate for anything where factual accuracy, original analysis, or brand voice precision is required.

The co-pilot role is the one the guide recommends for most professional writing work. The AI works alongside the human writer: suggesting alternatives, completing sentences, proposing structure, flagging gaps. The writer maintains authorship throughout and uses AI to reduce friction on specific tasks rather than handing off the work. This is the mode described in most practical AI writing guides, and the authors consider it both the safest and the most productive for professional contexts.

The reviewer role uses AI to evaluate finished or near-finished work: checking consistency, flagging passive voice, identifying sections that may be unclear, running basic fact-checking against source material. The reviewer role requires the least trust in AI judgment because the human has already made the content decisions; the AI is checking execution rather than setting direction.

When to stop

The guide’s specific criteria for pulling human judgment back in cover four situations: factual claims that cannot be verified by the model against primary sources; original analysis that requires the writer’s own interpretation rather than synthesis of existing text; brand voice that is precise enough that off-key phrasing would be noticed by readers; and regulated content where accuracy requirements carry legal or professional liability.

The authors emphasize a point they call “fluency is not accuracy.” AI-generated prose sounds confident and well-formed even when it is wrong. This makes it harder to catch errors in AI output than in rough human drafts, which tend to flag their own uncertainty through awkward phrasing. A writer reviewing AI output needs to check claims more carefully than they would check their own notes, because the output has already been made to sound credible.

Provenance and accountability

The guide covers two practical recommendations for teams publishing AI-assisted content. For provenance, it points to Adobe’s Content Credentials system as a way to document what role AI played in producing a piece, giving readers and editors a record of the production process. For accountability, it recommends naming one person as the sign-off authority for any piece that used AI — someone whose name goes on the work and who is responsible for the accuracy of every claim in it. The accountability recommendation is designed to prevent the diffuse responsibility that makes errors easy to miss: if everyone is responsible, no one checks carefully.

Who it is for

The guide is most directly useful for editors and content leads managing writers who are using AI tools without consistent standards. The three-role framework gives teams a vocabulary for discussing AI use without debating it from first principles, and the stop criteria are specific enough to apply to real decisions about what gets reviewed before publication.