The six stages
Most content programmes move through the same sequence regardless of industry: research, brief, draft, edit, publish, report. Generative tools are sold as improving all six at once. In practice the effect varies enormously, and averaging it into “30% faster” hides where the time actually goes.
Where it genuinely helps
Research synthesis
Condensing twenty sources into a structured summary is the strongest case. The work is mechanical, the output is checkable against the sources, and a wrong summary is obvious to anyone who read the material.
First-pass structure
Turning a brief into a heading outline removes the blank page without committing anyone to a sentence. Writers rearrange it immediately, which is the point — arguing with a bad outline is faster than producing a good one from nothing.
Where it is a wash
Drafting body copy
Generated prose arrives fast and then costs the time back in editing. Whether you come out ahead depends almost entirely on how specific the brief was. A vague brief produces text that is fluent, on-topic and worthless, and rewriting that is slower than writing.
Publishing and formatting
Real but small gains — alt text suggestions, meta descriptions, heading cleanup. Worth having, not worth restructuring a team around.
Where it makes things worse
Editing for judgement
An editor decides what to cut, what to argue harder, and what claim needs a source. Those are opinions about risk and audience. Automated editing optimises the sentence and leaves the argument untouched, which produces smooth text that says less than the draft did.
Reporting
Generated commentary on performance data is confident and frequently wrong about causation. It reads plausibly enough that people stop interrogating it, which is worse than no commentary at all.
A rule that holds up
These tools help most where the output is verifiable and hurt most where it is a judgement call. A summary can be checked against its sources in a minute. A claim about why traffic dropped cannot be checked at all without doing the analysis yourself — at which point the generated version saved you nothing and cost you scepticism.
Apply that test before adding a tool to a stage. If nobody on the team can tell within a minute whether the output is right, automating that stage moves risk rather than removing work.
General information only, not professional advice. Placeholder article written for this build.