Here's the thing nobody tells you about AI prompts: a single line of instructions almost always comes back generic. You ask ChatGPT or Claude for help, it gives you something usable-ish, and you spend ten minutes fixing the tone, adding context it didn't have, and cutting the parts that don't sound like you. Tomorrow, you start from zero again.

The fix isn't a better one-liner. It's giving the AI a real role, your actual context, and the exact steps to follow, before you ever hit enter. Below are five prompts we actually use, each shown two ways: the version most people type, and the version that gets it right the first time. Copy, paste, and swap in your own details.

1. Following up with a lead who's gone quiet

What most people typeWrite a follow-up email to a customer who hasn't responded.

You'll get something polite, generic, and forgettable, the kind of email that reads like it was sent to a thousand people, because it basically was.

What actually worksYou are a follow-up specialist for a small [your industry] business. A prospect named [name] inquired about [service/product] on [date] and has not responded since. Write a short, warm follow-up text and a follow-up email. Reference their original request specifically. Do not apologize for following up. End with one easy next step, not a hard sell. Keep the email under 100 words and the text under 30.

Swap the bracketed parts for your business and the specific lead. The more specific the inputs, the less generic the output.

2. Turning a long document into something you can actually use

What most people typeSummarize this contract.

You'll get a summary, technically. But it usually misses the one clause that actually mattered, because the AI has no idea what you're worried about.

What actually worksYou are reviewing this document on behalf of a small business owner who is not a lawyer. Read the attached [contract/quote/proposal] and produce: 1) a 3-sentence plain-English summary, 2) a bulleted list of every dollar figure, deadline, and obligation mentioned, 3) anything unusual compared to a standard agreement of this type, flagged clearly. Do not use legal jargon. If something is ambiguous, say so instead of guessing.

Paste the document text (or attach the file if your AI tool supports it) after this prompt.

3. Writing a job posting that attracts people who'll actually stay

What most people typeWrite a job posting for a customer service rep.

You'll get a generic posting that sounds like every other generic posting, which is exactly why it attracts generic applicants.

What actually worksYou are a hiring manager at a [size] [industry] business in [location]. Write a job posting for a [role] that filters for the right person rather than maximizing applications. Include: what a normal Tuesday actually looks like in this role, the one skill that matters most and why, a realistic pay range framed honestly, and one sentence about what this role is not (to discourage the wrong applicants from wasting everyone's time). Keep the tone direct, not corporate.

The "what this role is not" line does more filtering work than the entire rest of the posting.

4. Turning scattered reviews and complaints into an action list

What most people typeWhat do my reviews say?

Without the reviews pasted in, or with just a vague summary requested, you get vague generalities back. Garbage in, garbage out.

What actually worksBelow are [number] recent customer reviews and complaints for my business. Group them into recurring themes, ranked by how often each theme appears. For each theme, give one specific example quote and one concrete, low-cost fix I could make this month. Ignore one-off complaints that don't repeat. End with the single theme I should fix first and why.

Paste your actual reviews below the prompt: Google reviews, support tickets, even a rough list from memory works.

5. Writing a service page that search engines and humans both like

What most people typeWrite SEO content for my HVAC company website.

You'll get keyword-stuffed filler that reads like it was written for a search engine, because it was, and it'll rank about as well as it deserves to.

What actually worksWrite a service page for a [industry] business in [city, state] offering [specific service]. Write for a homeowner who has a problem right now and is comparing 3 local options. Open with the problem, not the company. Include the specific service area, one clear differentiator, and one direct call to action. Naturally include the phrases "[service] near me" and "[service] in [city]" without it reading like keyword stuffing. Keep it under 350 words.

Read it out loud after. If it sounds like an ad instead of a person, tell the AI to rewrite it in plain language and try again.

The pattern behind all five

Every strong prompt above does the same three things: it gives the AI a role, it gives it your specific context instead of asking it to guess, and it tells it exactly what shape the answer should take. That's the whole difference between a rough draft and something you'd actually send.

It's also, not coincidentally, the same thing we do before building any automated system: understand the actual role a task plays in your business, the real context around it, and the exact steps that need to happen, before a single line of anything gets built. A good prompt is a tiny, one-time version of that. A real automated system does it permanently, without you having to open ChatGPT at all.