I use models every week. I will not let them pretend to be the strategist. That sentence is the whole article. The rest is how to live it without looking like a Luddite, and without handing your budget to a system that cannot feel a close rate.
The useful jobs are boring. Cluster comments. Draft twenty hooks. Summarise what changed in the account. Flag a spend spike at 11 p.m. Those jobs save hours. They do not choose the offer. They do not decide what “good” means. They do not raise budget because the model is confident.
Confidence is not a conversion
A model will write a plan in the tone of a senior hire. That is a style, not a result. If you cannot point to a qualified conversion the plan is supposed to create, you have a memo. Memos are cheap now. Judgment is not.
This is why I refuse the sentence “the model said we should spend more.” It does not belong in a business that pays salaries. Spend is a human decision because the downside is cash, not a worse paragraph.
Write prompts like standard work
If a prompt cannot be shown to a junior marketer, it is not ready. Version it like an ad. Put the kill rule in the prompt: what to never claim, which numbers are off-limits, which offers are live. OpenAI’s prompting guide is a clean starting point for the craft. It is not a strategy.
Google — Responsible AI practices
Where I actually use it
Audits: I paste qualitative feedback and ask for clusters, not conclusions. I name the leak. Creative: I ask for twenty variants of a claim we already believe. I kill eighteen. Reporting: I ask for a first pass of the Monday note against the three-number scorecard. I rewrite the decision line myself.
I do not ask a model which audience to buy. I do not ask it whether to scale. I do not let it talk to a client without a human reading the sentence that commits money. Google’s responsible-AI notes are worth a single read if only to remember that the vendor is not going to sit in your board meeting.
What SMEs should do instead of panicking
You do not need an “AI strategy.” You need one SOP: drafts in, human out, number on the page. If you cannot name the number, do not add a tool. Read the scorecard essay, then come back. Marketers who document their loop will outrun teams that collect subscriptions.
If you want the longer desk view — agents, creative volume, and why the commercial no stays human — it lives on the AI page of this site. Pair it with the leak essay so you are not automating a broken path.
A practical stack for an SME: one model for drafts, one sheet for the three numbers, one human who is allowed to say no. That is cheaper than five subscriptions and a “prompt library” nobody opens. Tools on this site are listed as tools, not as a shopping list. Pick one. Write the SOP. Then ignore the rest until the number moves.
When a vendor sells you an AI agent that “runs your ads,” ask who owns the kill rule. If the answer is the agent, you have hired a very confident intern with your credit card. Also read the scorecard essay so you know which number the intern is not allowed to touch.
Also read: Find the leak before you scale the spend
Also read: A founder scorecard with three numbers