How To Write The Best AI Prompts

AI tools like ChatGPT, Claude and Gemini have become part of the everyday for many businesses. They’re being used to draft emails, summarise meetings, write first-draft copy and pull together quick research. But there’s a noticeable difference between the people getting genuinely useful results and the people getting bland, generic output. Most of the time, the difference isn’t the tool. It’s the prompt.

Here’s how to write prompts that actually work…

Give it context before you give it the task

The biggest mistake we see is people typing a request with no background. “Write me a LinkedIn post about our new product” will give you something passable, but it won’t sound like you, and it won’t speak to your audience.

A better prompt tells the AI who you are, who you’re writing for, and what you’re trying to achieve. Something like, “I run a B2B accountancy practice in the Midlands. I’m writing a LinkedIn post for small business owners about a new payroll service we’ve launched. The tone should be helpful and reassuring, not salesy.” That gives the AI something to work with, and the output improves dramatically.

Be specific about what you want back

AI tools are good at following instructions, but they’re not very good at guessing. If you don’t tell it the format, length or style, you’ll get whatever the model thinks is most likely to be helpful, and that’s usually a generic middle ground.

Tell it the length (300 words, three paragraphs, five bullet points). Tell it the format (a list, an email, a table comparing two options). Tell it the tone (formal, conversational, technical). The more specific you are, the closer the first draft will be to what you actually need.

Show it an example

If you have a piece of writing that captures the style you want, give it to the AI as a reference. This is one of the most underused techniques, and it transforms results.

Paste in a previous blog post, a sample email, or even a competitor’s piece of content, and ask the AI to write something new in that same style. AI models are very good at pattern matching, and a single example often gets you a better result than several paragraphs of style instructions.

Break big jobs into smaller steps

If you ask an AI to “write a full marketing strategy for our business”, you’ll get something that looks impressive on the surface and falls apart on closer reading. The task is too big, and the AI has had to guess too much.

Instead, break the work into stages. Ask it to research your audience first. Then ask it to suggest three angles for a campaign. Then ask it to draft messaging for the strongest angle. Each step builds on the last, and you stay in control of the direction. This also makes it much easier to spot where the AI has gone off track.

Treat the first answer as a starting point

The best prompters don’t expect a perfect answer first time. They expect a useful first draft, and then they refine it.

When you get the output, tell the AI what’s working and what isn’t. “This is too formal, can you make it warmer” or “this missed the point about pricing, can you rework the second paragraph” will get you a better version quickly. It’s a conversation, not a one-shot request.

The bit prompts can’t fix

Better prompts will get you better drafts, but they won’t replace expertise. AI can produce a confident-sounding paragraph on a topic it doesn’t really understand, and it can miss the nuance that someone with real experience would catch. The output still needs an expert pair of eyes to check accuracy, tone, brand fit and whether it actually says something useful.

This is how we use AI at Zinc. Our team uses these tools every day to speed up the foundation work, and our specialists then bring the judgement and craft that AI can’t. It’s the combination that produces the best results, not either on their own.