insider hot take

AI Doesn't Suck, Your Prompts Do (And Here's a Simple Fix)

AI was advertised as a game-changer for marketing, but anyone who has tried it knows the frustration: great models producing mediocre results. It's not the technology, it's us. The saying "garbage in, garbage out" still applies as much today as it did 10 years ago. The true strength lies not in using the latest LLM model, but in creating prompts that make AI think like a strategist rather than a parrot.

I have been using LLMs extensively since the early days, building AI workflows for various tasks, including GTM tracking, marketing ideation, in-depth analysis, and CRM automation. Basic prompts like "write a social post" lead to generic responses. But effective prompts can create gold! For example, try: "Act as a CRO (conversion rate optimization) expert with 10 years of experience in SaaS funnels. Analyze this landing page copy for friction points, then suggest three A/B tests. Include benchmarks and expected lift." This works because you define a role, set constraints, provide context, and specify the format - just like how top agencies brief their teams. Research shows that structured prompts can improve accuracy by 40-70% for copywriting, idea generation, and even data analysis.

The truth is, humans struggle with this. We tend to be vague, biased, rushed, and forget to add important details. That's why I use AI to help me craft my prompts; a method called meta-prompting, and you should too! Assign your favorite LLM a role, share your objectives and goals, and let it do the rest. For example: "You're a prompt engineer for marketers. Goal: create 5 LinkedIn hooks for a B2B webinar on AI personalization. Audience: CMOs overwhelmed with tools. Include emotional triggers, questions, and statistics. Aim for a 20%+ increase in engagement." Another way to take your prompts to the next level is to add "Ask me clarifying questions first." Adjust it, test it, and refine it, and then test out your new prompt. This method naturally formats your inputs in the best manner possible by encouraging the use of chain-of-thought reasoning (step-by-step breakdowns) and few-shot examples (your proven successes).

It may sound ridiculous to use AI to make prompts for AI, but it works! For my teams, this reduced ideation time by 80% and improved output quality overall. In three years, those who master prompts will be the ones leading their teams.

advertisement

advertisement

Don't believe it? Try it for yourself! Treat prompts like ad copy: A/B test your outputs and compare results with and without meta-prompting. 

Let AI rewrite your prompts for you, and watch as you become the most dangerous person in the room.

If you’re interested in submitting content for future editions, please reach out to our Managing Editor, Barbie Romero at Barbie@MediaPost.com.

Next story loading loading..