As AI tools embed deeper into daily workflows, the gap between good and bad outputs often comes down to one skill: prompting. We asked three industry voices what separates sharp prompts from vague ones, and why treating AI like a capable colleague, not a search engine, is the key to getting usable results.
Nick Myers, Chief Strategy Officer, OLIVER
“The problem with prompting is that people treat it like search, when it’s actually briefing. If you can’t brief well, you won’t prompt well, and vice versa. Good prompting takes clarity of thought and the instincts of a good teacher. I find the Feynman technique helps; explain the task as if to a smart novice, and where you stumble is where the brief is broken. But prompting is also unlike briefing, in a way that matters. A shoddy brief to a person gets pushback. A shoddy brief to a machine gets sycophancy. It tells you the work is wonderful and quietly fills the gaps itself. Agentic AI raises the stakes. An agent doesn’t stop to query a thin brief.
“It runs long sequences across other agents, apps and websites, burning tokens as it goes, and completes the work at speed in the wrong direction. Bad briefs always wasted budget. Bad prompts waste it faster. So, getting better, and sharing what works, is imperative. The barrier is incentive. People fear that AI, or more precisely someone who can use AI, will displace them, and so the best prompters keep their methods to themselves. Organisations must build a culture of safety in which the good ones teach how they think through a brief, and in which there is no stigma in saying “I used AI to do this.”
Chloe Parker, Partner & AI Champion, Clarity
“Good prompting isn’t about finding a magic phrase, it’s about treating the AI like a very capable but very literal new team member. The more context, constraints and examples you give it, the better the output. Most people prompt like they’re searching Google, when they should be prompting like they’re briefing a colleague.
“For example: ‘write me a LinkedIn post about what a good AI prompt looks like’ gets you AI slop. ‘Draft a 150-word LinkedIn post exploring what a good AI prompt looks like for comms leaders, sceptical tone, one stat, no jargon, end on a question, and use these two previous posts as an example of style and tone,’ gets you something usable.
“The best prompts read like a proper brief: they set the objective, the audience, the tone, and what ‘good’ looks like. Strip any of those out and you’re leaving the model to guess, and guesswork is where mediocre outputs come from.”
Darek Baczyk, SEO Technical Lead, Wildcat Digital
“Good prompting looks a lot like good delegation. If you handed a task to a new starter with one vague line and no context, you’ll get a rushed, generic result from AI too. People on my team who get consistently good outputs give the model what they’d give a junior colleague: the actual goal, any constraints, one example of what “good” looks like, and who the output is actually for. Skip that, and you’re asking the tool to guess, and it will, confidently, every time.
“That approach shifts with the tool and the task. For anything needing real reasoning, untangling a messy problem or weighing up options, I want to give more context and treat it like a conversation, going back and forth until it’s right. For quick, repeatable jobs, a tighter, templated prompt gets there faster. And once you move from a single back-and-forth chat to an agent working through several steps on its own, the brief has to be tighter still, because there’s no moment mid-task to catch a wrong assumption. You’re front-loading your judgement instead of applying it as you go.
“The skills gap inside most organisations isn’t really about AI. It’s that not everyone is a natural brief-writer, and most workplaces never taught anyone to be one. Ask a colleague something vague and they’ll usually ask a follow-up question before they start. AI won’t. It just answers what you asked, however unclear, which makes the gap between strong and weak briefers far more visible than it used to be.
“There’s no way to guarantee the best output every time, but you can shorten the odds. Treat the first response as a draft, not the answer. Show the model one concrete example rather than describing “good” in the abstract. Build a shared bank of prompts for your team’s recurring tasks, so nobody’s rewriting the same brief from scratch every week. On my own team, we took that further and automated a big chunk of client reporting entirely, using exactly this kind of structured, repeatable brief behind the scenes. The tool didn’t get smarter. We just stopped re-explaining the same task differently every time, and that alone is now saving dozens of hours a month.”
