Most bad AI answers are honest responses to vague questions. The model filled the gaps you left with the most average guess it could make.
This AI prompting guide for beginners shows how to close those gaps. You don't need jargon, templates from the internet or a paid course. You need the same instincts you already use when you hand work to a new employee: explain the situation, say what you want, describe what finished looks like and mention what to avoid.
Why Prompts Fail
When people say an assistant is "useless", their prompt usually looks like this: write a newsletter about our spring sale. The model has no idea who reads the newsletter, what is on sale, how long it should be, what tone the business uses or what it must not promise. So it writes something generic, cheerful and long. Then it invents a percentage discount, because sale newsletters usually have one.
Nothing went wrong with the tool. The prompt asked for an average newsletter and got one.
| Vague prompt | What the model has to guess | Better prompt |
|---|---|---|
| Write a newsletter about our spring sale | Audience, items, dates, tone, length, discount | Write a 150-word email to past customers of our garden centre announcing that seed potatoes and fruit trees are in stock from 1 March. Friendly, no exclamation marks, no discount mentioned. |
| Summarise this | Who it is for, what matters, how short | Summarise this supplier email in three bullet points for my business partner: what changed, what it costs us, what we need to decide. |
| Make a formula | Which columns, what logic, what software | Write a spreadsheet formula for column D that shows "Overdue" when the date in column B is before today and column C is blank. |
The Four Parts in Practice
Context: the part people skip
Context is everything the assistant can't see: your business, your customer, the history. One or two sentences is often enough. If you have a past example you liked, pasting it in counts as context and is usually the single biggest improvement you can make.
Task: one job per prompt
Use a clear verb: draft, rewrite, summarise, list, compare, explain. Asking for three different things in one prompt tends to get you three half-done things. Break big jobs into steps and run them one after another in the same conversation.
Format: describe the finished shape
Length, structure, reading level and where it will be used. "A table with three columns", "under 80 words for a text message", "bullet points I can paste into a slide". The format instruction prevents most of the editing you'd otherwise do by hand.
Limits: say what not to do
This is where you protect the business. Tell it not to invent prices, not to promise dates, not to mention competitors, to use British or American spelling, or to ask questions if information is missing. That last instruction alone cuts down invented details noticeably.
End important prompts with: "Before you start, ask me up to three questions if anything is unclear." You'll be surprised how often the questions reveal what you forgot to say.
Build a Prompt
Fill in the fields and watch the prompt assemble. Nothing you type leaves your browser; copy the result into whichever assistant you use. Leave out real customer names and numbers.
Prompt builder
Runs in your browser onlyStart with the task. The prompt assembles here as you type.
Iterate, Don't Restart
The first answer is a draft. The skill that separates comfortable users from frustrated ones is the follow-up. Instead of deleting the chat and trying again, tell the assistant exactly what to change:
- "Shorter. Cut the second paragraph."
- "Less formal. Write it the way I'd say it on the phone."
- "You invented a delivery date. Remove it and say we'll confirm by Friday."
- "Give me three different opening lines to choose from."
Each correction teaches the conversation what you want. After two or three rounds, copy the final prompt and the corrections into a note. That becomes a tested prompt you can reuse, which is the habit at stage two of the AI learning path.
Give It Examples
Showing beats telling. If you want product descriptions in your house style, paste two you wrote and say "match these". If you want a table, show the column headings. Models are very good at copying a pattern once they can see it.
Context: I own a small bike repair shop. Below are two service reminder texts I've sent before that customers liked. [PASTE EXAMPLE 1] [PASTE EXAMPLE 2] Task: Write a reminder for a customer whose bike has been ready for pickup for a week. Format: one text message under 300 characters, same tone as the examples. Limits: no storage fees mentioned, no emojis. Use [NAME] instead of a real name.
The messy middle: when examples backfire
Examples are copied closely, sometimes too closely. If both samples mention a sale, the new text may mention one too. Pick examples that share only the qualities you want repeated, and add a limit such as "do not reuse the specific offers in the examples".
The Bottom Line
Give context, one clear task, the format and the limits; then correct the draft instead of starting over.
- Paste a past example whenever you have one.
- Ask it to ask you questions before big jobs.
- Save prompts that worked, with a note on when to use them.
Once your prompts are reliable, the next skill is checking what comes back. Read how to check AI output for errors, and keep private details out using the AI data privacy guide.
Independent guide. Product names, features and data terms change often; check each vendor's current documentation before you rely on a feature. Nothing here is legal, security or professional advice.
