Signal Over Noise cuts through AI hype with weekly reality checks on what actually works. Written by a digital strategy consultant who tests every tool before recommending it, each Friday edition delivers honest reviews, practical frameworks, and real-world insights for professionals who need AI to work reliably.
Share
Why Your AI Prompts Aren’t Working (And How to Fix Them)
Published about 2 months ago • 3 min read
Signal Over Noise #10
July 11th, 2025
Dear Reader,
Earlier this week a friend sent me a message that got me thinking.
He’d been testing out a custom GPT I built months ago called Prompto - a structured prompt builder that forces you to break down what you’re actually trying to accomplish before throwing it at an AI. He was working on a CMS selection project for a client, comparing three to four different options and weighing up the pros and cons.
Prompt is a free tool designed to take simple prompts and make them "AI-ready".
He had put his initial question through ChatGPT without context or much detail and also tested it through Prompto, which is also using ChatGPT but with a custom set of instructions on how to format prompts.
“The outputs are quite similar,” he told me, “but the Prompto version is far more refined.”
Same task. Same AI model. Completely different results. The difference wasn’t the technology, it was the thinking that went into the prompt.
The “Just Figure It Out” Problem
Most people treat AI like a search engine with a personality. They type in whatever comes to mind, hope for something useful.
But here’s what I’ve learned from building automations and testing various AI tools over the past year: vague inputs create noisy outputs. When we ask AI to “just figure it out,” we’re not just delegating the task, we’re delegating our responsibility to think clearly about what we need. And this is a huge red flag on how not to use AI.
I see this constantly in my consulting work. Clients will show me AI outputs that are technically correct but completely miss the mark. Almost always, the problem isn’t the AI - it’s that nobody took time to clarify what they actually wanted.
Slowing Down to Speed Up
Using a structured approach like Prompto forces you to pause and consider:
What are you actually trying to accomplish?
What does good output look like for this specific task?
What context does the AI need upfront?
What doesn’t matter and can be left out?
This isn’t about making the AI work harder. It’s about making you think more clearly.
In my friend’s case, this meant getting proper comparisons between the CMS systems that aligned with his actual project requirements: security considerations, upgrade paths, editor usability. Instead of generic feature lists, he got analysis that directly informed his decision.
The Real Benefit
Here’s what I discovered when I started using structured prompting: it doesn’t just improve AI outputs. It improves your own thinking.
Structured prompting helps to improve your own thinking.
When you’re forced to articulate exactly what you need, you often realise you weren’t entirely sure yourself. The prompt-building process becomes a form of problem clarification.
I experienced this whilst building my social media automation earlier this year. Each time I refined a prompt for Claude or Perplexity, I understood my own content goals more clearly. The AI got better results because I finally knew what I was asking for.
Don’t Automate Unclear Thinking
The more capable these AI models become, the easier it is to get something that looks reasonable without putting in proper thought. But “looks reasonable” isn’t the goal - getting the right answer is.
When you take time to build a clear, well-structured prompt, you’re not just writing better instructions for an AI. You’re clarifying what actually matters to you.
This becomes especially important as AI integrates deeper into our workflows. We can’t afford to delegate our thinking entirely to machines, no matter how sophisticated they become.
Because ultimately, that’s what effective AI use comes down to: asking better questions so you can stop spinning your wheels on mediocre answers.
If you’re struggling with getting consistent results from AI tools in your work, I’d love to hear about it. Always up for a conversation about practical prompt engineering.
* I’m currently working on a version with even more guided structure for the public to use.
Stop switching between disconnected AI tools. In a 90-minute AI Action Plan Session, I'll show you how to set up the kind of orchestrated workflows I use daily - where your AI can read your files, update your systems, and execute complex tasks across all your tools. Let's design your unified AI workflow (or build a weird little thing!)
Signal Over Noise cuts through AI hype with weekly reality checks on what actually works. Written by a digital strategy consultant who tests every tool before recommending it, each Friday edition delivers honest reviews, practical frameworks, and real-world insights for professionals who need AI to work reliably.
Signal Over Noise #18 September 3rd, 2025 Dear Reader, Air Canada recently learned an expensive lesson about AI implementation. Their customer service chatbot provided incorrect information about bereavement fares, and when challenged, the company argued they weren't liable for their AI's mistakes. "My AI told you that." isn't much of a legal defense. The court disagreed. The ruling was clear: organisations remain accountable for AI-generated decisions. This case represents a broader pattern...
Signal Over Noise #17 August 27th, 2025 Dear Reader, Two weeks ago, I built a complete business intelligence tool in one afternoon. Not a prototype, but a fully functional system that scrapes social platforms for monetizable problems, analyzes them with AI, and outputs detailed business opportunities. It’s also something that I had no planned intention of building that day. The trigger? A fellow developer’s impressive SaaS demo that I was curious enough to use, but at a subscription price I...
Signal Over Noise #16 August 20th, 2025 TLDR: Watch the video above for a brief on this week's issue. Dear Reader, I’ve been publishing this newsletter every Friday for the last 10 months, but growth has completely stagnated. I could keep doing what I’m doing and hope for different results, or I could treat this like any other workflow problem and systematically fix it. So - it’s not an accident that you’re now receiving this on a Wednesday. Note: DALL-E and SORA have big problems displaying...