Patient with AI prompts: how to get good output

"Hey ChatGPT, write a blog post about climate change"

That's how most people use AI. Short question, no context, and then they wait for a miracle. When the miracle doesn't show up, AI gets the blame. "It doesn't work for my work." "Too superficial." "Writes generic stuff."

That's not fair. Most mediocre AI output isn't down to bad AI, it's down to bad prompts. And that's good news, because prompting is something you can learn. Faster than you think.

This piece is not another "10 prompt tricks" listicle you've already seen in a carousel. It's about the mindset you bring to Claude, ChatGPT and Gemini, and why patience is what actually moves the needle.

A prompt is not a search query

Google is fast. Type a few words, get a list, click through, done. AI works fundamentally differently, and if you don't feel that difference, you'll keep getting output you can't use.

AI wants to know more about you. It asks questions back, or it should. Why are you looking for this? For whom? In what style? What does the reader already know? It's a dialogue where you and the model arrive at something together. The model learns from what you say and builds on it, inside the conversation.

With Google you give a question and get a list. With AI you give context and get a conversation partner. Different workflow, different mindset. Anyone who keeps prompting like they're Googling will keep getting Google-shaped output forever.

Treat AI like a new colleague

Imagine: new colleague, first day on the job. You ask for a report on last week's project. Would this be enough: "Just put together a report on that project"?

Of course not. You'd explain the goal, who's reading it, which details matter, in what form (slides, memo, email), with what deadline. With AI it's exactly the same. More so, in fact: AI knows nothing about your company, your clients, your cultural context, your brand voice. Nothing at all, except what you put in the prompt.

Instead of giving commands, take the time to spell out what you need and what matters most in the result. A good prompt for a blog post looks something like this:

"Write an 800-word blog post about [topic] for [audience]. The tone is informal but expert, comparable to the writing style of [example]. Avoid buzzwords like 'leverage' and 'empower'. The reader should know three concrete actions by the end. Start with a surprising statistic or example. End with a question for the reader."

That's not a more complex prompt. That's just the context you'd give a new colleague. Refusing to do this is asking, unconsciously, for generic output.

Clarify what you don't want

It helps enormously to be explicit about what you don't want. No jargon if it has to stay simple, no long sentences if you want an active style, no technical details if it has to be accessible. No Title Case, only sentence case. No em-dashes or hyphens used for emphasis. No echoing of the main theme in the closing paragraph.

By saying out loud what you don't want, you help AI find the right direction faster. Think of it as drawing a map and marking the places to avoid. With free ChatGPT this already works well. With Claude and the Pro versions it works even better, because they hold more instructions across a long conversation.

And yes, the fact that I'm writing this piece without em-dashes myself is the same request I make of every AI. It works for them too.

Reverse the roles

Usually you ask AI something. Try the opposite. Let AI question you.

Try: "Play my critical manager. What five things are missing from my proposal?" Or: "Pretend you're a journalist interviewing me about [topic]. Ask sharp questions that test my assumptions." Or: "Game rules: I have an idea, you play a potential client who isn't immediately convinced. Keep asking until you really understand what I'm offering."

What comes back is often more useful than what you'd have produced on your own. You see your plan through a different lens, gaps in your story become visible, and you get questions you'd never ask yourself.

For an SME owner in a small town this is an underrated technique. You don't always have a sparring partner on tap. AI can do that job, if you instruct it well. No meeting needed, no lunch, no calendar Tetris.

Perfection in one go? Forget it

Every AI "mistake" is a chance to refine. Say: "This part is good, but adjust that next bit." Or: "The tone isn't right yet. Make it more direct." Or: "The second paragraph feels too clichéd. Rewrite it with a concrete example."

That's how you and your AI tool get to know each other inside a single conversation. A long chat with good iterations almost always beats one perfectly engineered opening prompt. Invest in the conversation, not in the first line.

With Claude and ChatGPT-Plus you can also set system prompts or "custom instructions". You record your preferences once instead of repeating them in every chat. For recurring work, email replies, social posts, summaries, that's gold.

Advanced technique: chain-of-thought

For heavier tasks it almost always works to explicitly ask AI to think out loud first. "Before answering, write out your reasoning. Then give the final answer." That's called chain-of-thought prompting and it produces noticeably better results for analyses, decisions, and multi-step problems.

Example. Instead of "write a marketing plan for my bakery", try: "For my bakery in [location]: first write your analysis of the target audience, then your hypothesis about the main levers, then your proposed marketing mix with reasoning, then a priority schema. End with a 30-day action plan."

It costs more tokens and a bit more wait time. The output is almost always better reasoned, and you can argue with AI about one part without the rest collapsing.

The bottom line

Quality output from AI doesn't fall in your lap. It needs time to grow, step by step, just like with real people.

Every interaction teaches you more about what the model can and can't do. At the same time, the model, inside the conversation or via custom instructions, learns your preferences and style. Sometimes you have to take a step back to take two steps forward.

The owners and professionals who are actually getting somewhere with AI in 2025 aren't the ones with the most tools. They're the ones who learned to treat AI like a colleague: with context, with patience, and with iteration.

Want to know where AI delivers the most value inside your organisation? Take the free AI Scan and get a personal analysis in your inbox within one minute.