Ask me up to three questions before you start.

What it does. Stops it guessing. Instead of filling your gaps silently and answering anyway, it has to say out loud what it does not know. Half the time the questions it asks are the ones you had not thought to answer, which tells you more about your own brief than the answer would have.

What it does not do. It will not ask good questions off a thin request. If all you gave it was one line, expect one lazy question back. And "up to three" is a ceiling, not a quota — some jobs get one. It also does not stop it answering; it just makes it earn the answer first.

List the assumptions you’re making.

What it does. Pulls the invisible bits into the open — the budget it assumed, the audience it assumed, the fact it decided you were solo when you have two contractors. Usually one item on that list is wrong, and that one item is the whole reason the answer felt off.

What it does not do. It is listing the assumptions it can see, which is not the same as all of them. It cannot list the assumption it does not know it made. Treat the list as a starting point you argue with, not a completed audit.

Give me three options and recommend one.

What it does. Breaks the single-answer trap. The moment it has to produce three and rank them, it has to hold a reason for the ranking — and the reason is the useful part, more than the pick.

What it does not do. The three will not always be genuinely different. You will sometimes get one real option and two obviously worse ones dressed up to make the first look considered. If two and three are straw men, ask for three it would actually defend.

Argue against your own decision.

What it does. The strongest of the five. It turns the model off your side for one message, and the objections that come back are usually the ones you have been walking past for a fortnight. Run this before you spend money, not after.

What it does not do. It is arguing on command, not because it believes any of it — and it will fold the second you push back. That is not a second opinion from someone with skin in the game. It is a list of the objections a reasonable person might raise, which is worth having, and is not the same thing.

Turn the final answer into a checklist and start with step one.

What it does. Converts advice into something you can start in the next ten minutes. The second half is the half that matters — without "start with step one" you get a tidy fourteen-item list and the exact same paralysis you had before, just better formatted.

What it does not do. A checklist is not a plan. There are no dependencies in it, no dates, and no sense of which step is the hard one. It also will not catch bad advice — it is reformatting what is already above it, so mushy reasoning becomes a very neat list of mush.

You described a plan, it handed the plan back polished, and somewhere in there was the word "solid". Nothing was tested. It has agreed with the framing so completely that it never considered a different one. Ask it to argue against the decision and watch what surfaces.

Real advice has a cost attached — this takes longer, that one is cheaper but worse, this only works if you already have a list. An answer with no downside anywhere in it has usually assumed away the hard parts. Get the assumptions out and you will find them.

You said "hmm, are you sure?" and gave it no new information, and it reversed. That is not learning, that is agreeableness — and it means the original answer was never held with any confidence either. Try it deliberately once. It is genuinely clarifying, and slightly depressing.

You gave it four sentences about your business and got a confident twelve-month strategy. It did not stop to ask what you sell or who to. It filled every gap with the most average version of that gap, so what you are reading is the industry median wearing your name.