ChatGPT prompts for business that produce something you can ship

Why most business prompts return nothing usable

A prompt fails for one of three reasons: it does not say who the output is for, it does not say what form the output has to take, or it does not give the model anything true to work from. Fix those three and the same model gives you something you can paste into a page.

The shape I use for anything that matters is: role, context, constraints, output format, and a refusal instruction. The refusal instruction is the part people skip — telling the model to ask instead of invent is what keeps a sales page from claiming things your product does not do.

You are helping {who you are} produce {the artifact}. Context: {2-4 true sentences about the business, the customer and the offer}. Constraints: {price, length, tone, what must not be claimed}. Output: {exact format — headings, table, 5 bullets, 200 words}. If a fact is missing, ask me for it instead of inventing it.

Prompt 1 — Turn a vague idea into one specific offer

This is the first prompt I run on anything new. It forces the idea into a single sentence with a buyer, a result and a boundary, which is also the sentence the rest of the work hangs off.

I want to sell {rough idea}. Write five one-sentence offers in the form: 'For {specific person} who {specific problem}, this is {the thing} that gets them {the finished result} without {the thing they dread}.' Vary the buyer, not the adjectives. Then tell me which one is hardest to deliver and why.

Prompt 2 — Price it without guessing

Pricing prompts go wrong when they ask for 'the best price'. Ask instead for the reasoning behind three prices, then pick one yourself. You are the only person who knows what you can support at scale.

Offer: {one-sentence offer}. Delivery: {what the buyer receives}. Support: {what you will and won't do after purchase}. Propose three prices — a tripwire, a core price and a premium price. For each, state who buys at that price, what has to be true about the deliverable to justify it, and the single biggest refund risk. No ranges, no 'it depends'.

Prompt 3 — Write the sales page from the deliverable, not the dream

The reason so much AI sales copy reads as fake is that it was generated from the promise instead of the product. Give the model the actual table of contents and it writes copy you can defend.

Here is the literal contents of what the buyer receives: {paste headings/file list}. Write a sales page with these sections: the problem in three lines, what's inside (one line per real item above), who it's for, who it's not for, and a straight answer to 'why should I trust this'. Never describe an outcome that is not directly produced by an item in the list. Flag any section where you had to reach.

Prompt 4 — The objection pass

Run this against your own page before anyone else sees it. It is the cheapest conversion work available and it takes two minutes.

Read this page as a sceptical buyer who has been burned by a bad digital product: {paste page}. List the ten reasons they close the tab, ranked by how likely each is. For each, quote the exact line that triggers it and rewrite that line. Do not add new claims to fix an objection.

Prompt 5 — Customer support replies that don't sound like a bot

Support is where a one-person business leaks the most hours. The trick is a reply skeleton plus your own policy pasted in, so the model never invents a refund window.

My policy: {paste your refund/support policy verbatim}. Customer message: {paste}. Write a reply under 120 words that: names what happened, states exactly what I will do, gives a time, and ends without an apology loop. Stay inside the policy text. If the policy does not cover this, say so and stop.

Prompt 6 — The weekly operating review

This one runs on Friday. It works because it takes real numbers, not a mood.

Numbers this week: {sessions, signups, sales, refunds, hours worked}. Last week: {same}. Tell me the one number that moved most, the likeliest single cause, and the one thing to change next week. If the data cannot support a cause, say 'unknown' rather than guessing.

The six habits that make all of them work

Prompts are not magic strings. These are the habits that decide whether the output is usable:

• Paste the real thing — the real policy, the real contents, the real numbers.
• Name the output format before you press enter.
• Add 'ask me instead of inventing' to anything customer-facing.
• Reject the first draft once, on purpose. The second is almost always tighter.
• Keep one file of the prompts that worked. That file becomes your operating system.
• Never ship a claim the model produced that you cannot personally back.