Picture a joiner’s Monday morning. Before 9am he has used AI three times: to draft a reply to a customer chasing a quote, to check whether a supplier’s invoice matches what he ordered, and to ask his web person for a small fix to his site’s contact form. Three different jobs, same tool. Prompt all three the same way and at least two come back wrong or useless. The trick isn’t learning three different tools, it’s noticing that the same handful of prompting habits show up differently depending on what you’re trying to get done.
We cover those habits properly in How to write a prompt that actually works. If you haven’t read it yet, start there. Here we take three situations you’ll actually hit, writing something, checking something, and fixing a piece of tech, and show what good prompting looks like in each.
Writing: emails, job ads, social posts
Writing tasks are usually the first thing people try with AI, and the gap between vague and good is easiest to see here. Ask for something vague and you get cheerful, empty adjectives back. Ask properly and you get something you could nearly send.
Say a cafe owner needs a job ad for a part-time barista. The vague version: “Write a job advert for a barista.” What comes back reads like every job ad ever written, “passionate individual”, “fast-paced environment”, no pay, no hours, nothing that sounds like an actual cafe in an actual town.
The better version gives it the details, an example to match, and a limit: “You’re a small independent cafe owner in Stockport hiring a part-time barista, three days a week, £11.50 an hour, weekend availability essential. Write a job ad in the same tone as this old one we posted [paste old ad]. Keep it under 150 words, friendly rather than corporate, and mention that Gaggia machine experience is a bonus, not a requirement.” That pasted example does more work than any instruction about “tone”, because it’s now copying something real instead of guessing at a vibe. The same approach works for an overdue-invoice chaser or a social post: give it the facts, a real example to match, and a length.
Analysis: reviewing a quote, summarising a contract, checking numbers
Analysis prompts care less about tone and more about precision, and this is the category where you never take the first answer on trust. AI can sound completely certain while getting a number wrong.
A bathroom fitter gets a supplier quote for materials on a refit. The vague version: “Check this quote.” That gets a vague summary back, with no real way to know if it caught anything.
The better version gives it a role, something to compare against, and a specific task: “You’re a quantity surveyor helping a small bathroom fitting business check supplier quotes for errors. Here’s this month’s quote [paste it]. Compare the line items against the materials list I actually ordered [paste list], flag anything missing or priced differently to last time [paste old invoice], and add up the total yourself so I can check it against theirs.” Naming the role sharpens the kind of checking it does, and the old invoice means it’s comparing real numbers, not guessing what’s normal. “Add it up yourself” is there on purpose: you’ll check that total by hand anyway, but a second calculation often catches the transposed digit before you do.
Commissioning code and technical work
For a non-technical owner, this is usually the scariest category, a broken form, a slow page, an idea for a simple script to save an hour of admin. The fear is sounding daft, or getting a “fix” that breaks something else. The habits don’t change here, you just describe the outcome, not how to build it.
A plumber’s contact form isn’t showing a confirmation message. Vague version: “My contact form doesn’t work properly, can you fix it.” That could mean almost anything, and whoever picks it up, a developer or an AI coding tool, is left guessing.
Better: “On my website’s contact page, when someone submits the form, I want them to see a message saying ‘Thanks, we’ll call you within 24 hours’ instead of the blank white screen it shows now. Nothing else on that page needs to change. Before you touch anything, tell me what’s currently happening when the form is submitted, and check with me before making any changes near the booking or payment pages, they need to stay exactly as they are.” That’s context, a small specific outcome, and a boundary on what must not be touched. Inviting it to check first matters most here: a change made on a hunch can break something that was working fine, somewhere you never mentioned.
Where to take this next
All three examples lean on the same eight habits laid out in How to write a prompt that actually works. To build the skill properly rather than picking it up example by example, our course Writing prompts that work walks through all eight in more depth, with exercises drawn from real business tasks. Want the structure done for you instead? Our Prompt Builder asks the right questions and puts a properly formed prompt together, whichever of these three jobs you’re on.
Prefer to work through this live with people who do it for a living? Our AI Automation Masterclass in Manchester covers exactly this. Tickets are normally £20. This one’s free, a limited-time offer to launch the series.