Where generative tools genuinely save time, where they quietly cost it, and how to tell the two apart before committing budget.
Generative tools have genuinely changed parts of this job. They have also quietly added work in places that are harder to see on a timesheet.
Where the time is actually saved
First drafts of routine copy, variant generation for ad testing, summarising research calls, and writing the structured data and metadata nobody enjoys writing. These share a shape: high volume, low ambiguity, and a human reviews the output anyway.
Where it quietly costs
Anything requiring judgement about a specific client's position. A model will produce confident, fluent strategy that is indistinguishable from the strategy it would write for a competitor, because it has no access to what makes the business different. Editing that back into something true routinely takes longer than starting from a blank page.
The second cost is review load. Output volume rises faster than review capacity, and the failure mode is not obviously-wrong copy — it is plausible copy nobody checked.
How to tell the difference before committing budget
Ask whether a competent freelancer could do the task well knowing only what is in the brief. If yes, a model will probably help. If the task needs context that lives in your head, in the sales calls, or in last quarter's numbers, the tool will produce something that reads well and says nothing.
“The best marketing decision we made was agreeing what success meant before anyone spent anything.”
