AI is genuinely useful for the first pass
Writing an audit summary from scratch, drafting a report narrative from raw metrics, sketching a social post from a one-line brief. These are all blank-page problems. AI is genuinely good at them, and using it well means I ship better work faster than I would otherwise.
AI is also confidently wrong
The same tools will invent statistics that don’t exist, quote sources they can’t verify, and write in a flat, generic tone that reads like every other piece of AI content on the internet. They will confidently assert things that are subtly incorrect, get the specifics of a client’s business wrong, and produce content that ranks initially and then gets hit by helpful-content updates.
AI is genuinely useful for the first 80% — the blank-page problem. The second 80% is where a human earns the fee.
The line: first-drafted, human-reviewed, never AI-published
My rule is simple. AI gets used to draft. I get the final pass. No client ever receives anything that I haven’t read, refined where the AI was wrong, and signed off on. No social post ever goes live without a personal review of tone, accuracy, and brand fit. No report ever lands in a client’s inbox with a single line I didn’t write or rewrite.
That’s the line. It costs me time, which is why my price is what it is, but it’s the difference between an agency that uses AI to ship more work and an agency that uses AI to ship better work.
What this looks like in practice
Audit summary: first drafts from my structured notes; I rewrite the sections that read generic, add the local context the AI didn’t have, verify any numbers. Report narrative: first drafts from the raw metrics; I rewrite the "so what" that actually tells the client what to think. Social post: first drafts 10 from the calendar brief; I pick the 4 that work and rewrite them in the client’s voice.
In every case, what reaches the client is something I’ve personally read end-to-end. That’s the work.