Why Your AI Prompts Get You Rubbish Results (And How to Fix It)
I saw something last week that perfectly sums up the biggest mistake I see people and businesses making with AI. And it came from a major player in the space: OpenAI—the people behind ChatGPT.
They released a “prompt pack” for enterprise teams, and it was filled with two-sentence prompts.
It treats some of the most powerful tech on the planet like a slightly better Google search. And honestly, too many of us are doing the same thing.
When you give AI a generic, low-context prompt, you get a generic, low-value answer. It’s a simple case of garbage in, garbage out.
This is fine when you need a quick throwaway answer, but when you need a quality response you need to provide a quality prompt.
Take this example from the marketing section:
“Research how top 5 competitors structure their blog content strategy. Include tone, topics, frequency, SEO focus, and CTAs. Provide URLs, takeaways, and a table summarising common and standout tactics.”
What’s missing? Everything. A prompt this vague forces the AI to spit out a summary, not an actionable result you can actually use. You’re treating AI like a search engine for facts, not a partner for getting work done.
So, what actually works?
Think of it like briefing a new intern. You wouldn’t just tell them, “look into our competitors.” You’d give them context: who we are, who our competitors are, what we’re looking for, and how you want the final report to look.
The same goes for AI. A great prompt isn’t short; it’s rich with detail because it’s built to do a real job. This idea of giving AI rich detail is sometimes called ‘prompt engineering’ or ‘context engineering,’ but you don’t need a fancy term. You just need a good framework.
Here’s a simple one I use—the COSTAR model—to craft prompts that deliver actual value.
- Context: Give the AI a role.
- Objective: State the specific goal.
- Steps: Outline the exact process.
- Task: Define the output format.
- Audience: Who is this for?
- Restrictions: What should it not do?
Let’s Rebuild That Marketing Prompt with COSTAR
Instead of that vague, two-sentence prompt, let’s delegate the task properly.
C – Context
You are an expert Content Marketing Strategist. You have been engaged by XYZ, an Australian B2B SaaS company that specialises in AI-powered workflow automation for small to medium-sized businesses. Our primary audience is non-technical business owners. You are analysing our top five competitors to inform our content strategy.
O – Objective
Your objective is to conduct a detailed competitive analysis of our competitors’ blog content to identify common patterns, standout tactics, and strategic gaps that XYZ can exploit.
S – Steps
- For each of the 5 competitor URLs provided, analyse their 10 most recent articles.
- For each competitor, extract data on: Tone of Voice, Core Topics, Posting Frequency, SEO Focus, and Primary CTAs.
- Synthesise your findings to identify common and standout tactics.
T – Task (Output Format)
Provide your analysis as a concise report in markdown, including an Executive Summary, Individual Competitor Analysis, a Comparative Analysis Table, and 3-5 Strategic Recommendations.
A – Audience
The final report is for the Head of Marketing at XYZ. The language should be professional, data-driven, and focused on strategic business outcomes.
R – Restrictions
- Do not analyse articles published more than 18 months ago.
- Ignore non-blog content like press releases.
- All written output must be in Australian English.
See the difference? You’re no longer asking a question; you’re delegating a task with clear instructions. The first prompt gets you a list of facts. The second gets you a strategic analysis you can actually act on.
The best part? You can use AI to help you craft these prompts. I’ve shared the COSTAR meta prompt template here on my site.
If you need help getting the most out of AI, reach out to me to discuss how I can help.