Remember when everyone was telling you to start your prompts with “You are a world-class copywriter” or “Act like a math professor”? What about those dramatic threats like “This is very important to my career” or “Someone will die if you don’t give me a great answer”?

I hate to break it to you, but those techniques that worked so well with earlier AI models? They’re basically useless now for getting better accuracy.

Sander Schulhoff, the researcher behind The Prompt Report (a 76-page study analyzing over 1,500 papers on prompting techniques), put it bluntly in a recent conversation: “Role prompting does not work” for accuracy-based tasks. He even went viral on Twitter for saying exactly that.

But here’s the thing. While these old-school techniques have lost their magic, there are far more effective methods you can use instead. And the best part? They’re actually simpler to implement once you know what they are.

The role prompting reality check

Let’s start with role prompting. You know, when you tell ChatGPT or Claude something like “You are an expert marketing strategist” before asking your actual question.

The research is pretty clear on this. When scientists tested thousands of different roles across various tasks, they found that any performance improvements were statistically meaningless. We’re talking about accuracy differences of 0.01 percent. That’s not just small, it’s basically noise in the data.

Think about it this way: if telling an AI it’s a math professor actually made it better at math, wouldn’t that suggest something strange about how these models work? The truth is, it doesn’t activate some special “math region” of the AI’s brain like we might have hoped.

That said, roles aren’t completely worthless. They still work well for expressive tasks where style matters. If you want your AI to write like Tyler Cowen or sound more formal in an email, giving it a role can help. But for anything requiring accuracy, factual information, or problem-solving? Save yourself the tokens.

Why threats and bribes fell flat

The other technique that’s lost its effectiveness is the dramatic emotional manipulation. You’ve probably seen prompts like “I’ll tip you $20 if you do this well” or “My career depends on this answer.”

These used to work on earlier models, but modern AI systems are trained differently. They’re not sitting there thinking “Oh no, this human’s career is on the line, I better step up my game!” That’s just not how the training process works.

During development, these models aren’t told “Do a good job and you’ll get paid.” The reinforcement learning happens at a much more fundamental level. So your promises of tips or threats of consequences don’t trigger the response you’re hoping for.

What actually works instead

Instead of relying on these outdated tricks, focus on techniques that consistently improve performance. Here are the four most effective approaches you can start using today:

Give examples of what you want.

This is called few-shot prompting, and it’s the single most impactful technique you can learn. Instead of describing your writing style, paste in a few examples of your previous work. Instead of explaining what format you want, show the AI exactly what success looks like.

For instance, if you want help writing podcast episode titles, don’t just say “write catchy titles.” Give it 5-10 examples of titles that have performed well for you in the past, then ask it to create new ones in that style.

Break complex tasks into smaller steps.

This is decomposition, and it works exactly like it sounds. Before tackling a big problem, ask the AI: “What are some sub-problems that need to be solved first?” Then work through each piece systematically.

Imagine you’re asking for help with a customer return policy question. Instead of dumping all the details at once, you might first ask it to identify what information it needs (customer status, purchase date, product type, return timeframe), then gather each piece before making the final determination.

Ask for self-criticism and improvement.

After the AI gives you an initial answer, ask it to review and critique its own response. Then tell it to implement those improvements. This simple back-and-forth often catches errors and produces better results without you having to spot the problems yourself.

Provide relevant context upfront.

This might seem obvious, but most people don’t give AI enough background information. If you’re working on content for your company, include details about your industry, target audience, and company background. If you’re analyzing data, explain what the business does and why this analysis matters.

The more context you provide, the better the AI can tailor its response to your specific situation. Don’t worry about making your prompts too long, especially in conversational settings where you’re not paying per token.

Making it practical

Here’s what this looks like in practice. Instead of starting with “You are an expert email marketer,” try this approach:

First, give context: “I work at a B2B software company that helps marketing teams automate their workflows. Our typical customers are marketing directors at companies with 50-500 employees.”

Then provide examples: “Here are three emails that performed well for us in the past: [paste actual examples]”

Follow with your specific request: “Write a follow-up email for prospects who downloaded our automation guide but haven’t scheduled a demo yet.”

If needed, ask for refinement: “Can you review this email and suggest improvements for clarity and persuasiveness? Then rewrite it incorporating those changes.”

This approach gives the AI everything it needs to produce relevant, high-quality output without relying on outdated role-playing techniques.

A word of warning

One more thing to keep in mind. In conversational settings, you don’t need to apply all these techniques every time. Sander himself admits that for quick tasks, he often just types something like “writ emil” (yes, misspelled) and gets decent results.

The real value of these advanced techniques comes when you’re building products or systems that will process thousands of requests. In those cases, taking time to craft the perfect prompt pays dividends.

But for everyday conversations with AI? Start simple, then add techniques as needed.

The bottom line

Stop wasting time with outdated prompting techniques that no longer work. Role prompting and emotional manipulation might have been effective with earlier models, but today’s AI systems respond better to clear examples, structured thinking, and relevant context.

Focus on showing rather than telling, breaking down complex problems, and providing the information the AI needs to give you exactly what you’re looking for. Your results will improve dramatically, and you’ll save time by not relying on techniques that simply don’t work anymore.

Ready to level up your AI skills with techniques that actually work? I’d love to help you build a more effective AI strategy for your team. Connect with me on LinkedIn to explore how we can make AI work better for your business.