I've spent the last few years interacting with (LLMs), with , and more than occasionally going down the rabbit hole when I encounter new AI technology.
Based on all that experience, I've learned that the quality of the AI output depends way more on than on the tool itself. AI tools are all using state-of-the-art models, and while some are better than others for certain tasks, the real difference comes down to how you interact with them.
But you don't have to reinvent the wheel every time you interact with an . There are prompt patterns you can follow that'll help you get what you want faster.
Table of contents:
Why do prompt patterns work?
LLMs rely on patterns to provide answers to queries: fundamentally, they work by predicting the next most likely word in a sequence. (You can learn more about that in Zapier's article about .)
That means that by understanding the underlying patterns in the training data and how to use them in your prompts, you can get higher quality responses.
8 AI prompt templates to get you better outputs
Let's dive into some of the easiest prompt patterns you can use to get your AI underling to give you what you need. You can use them with any AI chatbot or with , which can start with your prompts and then take action on your behalf across your entire tech stack.
1. Cognitive verifier pattern
With the cognitive verifier pattern, you tell the AI to break your question down into smaller parts and get feedback before proceeding.
Example:
Why it works:
The cognitive verifier pattern forces the AI to break down complex questions into smaller, more manageable components. Instead of rushing to answer, it gathers additional context that leads to more informed responses. It's similar to the "reflection" prompt pattern, where you query the LLM to "Take a moment to think about your answer before responding."
It's almost like how you'd talk to a kid—or maybe a high schooler who's taking an important test. It forces the AI into a more deliberate and structured reasoning process that leverages self-evaluation and feedback loops.
When to use this prompt pattern:
This prompt pattern helps you answer complicated questions more accurately. Use it when dealing with complex scenarios (such as legal contracts), getting personalized recommendations, and for technical problem-solving.
2. Template with placeholders
Instead of giving the AI an example, giving it a template with placeholders can help prevent hallucinations and force it to come up with unique content instead of mimicking yours.
Example:
Some examples of what a template with placeholders may look like include:
Email: "Dear [NAME], your [PROJECT] is [STATUS] as of [DATE]..."
Report: "Executive Summary: [SUMMARY] Key Findings: [FINDINGS]..."
Per of (then) Verblio, at , placeholders are a great way to proactively prevent hallucinations. She gave the example of wanting to use AI to create unique location pages for a local SEO client.
Here's what the prompt for something like that might look like:
In Megan's experiment, without the placeholders, the AI hallucinated incorrect details about the specific location (e.g., "Find us at Route 59, next to the McDonald's"). Using the placeholders also allows humans to more easily go through and fact-check.
Why it works:
Using a template with placeholders forces AI to follow a specific format to create consistency across outputs. For example, if you're using and want the final output to follow a specific format for each workflow, this is the prompt pattern you want to incorporate in the instructions. It's a great way to marry the best of AI with the best of human capabilities.
Use cases:
The template with placeholders prompt pattern can be helpful in business communications, email templates, daily reports, and .
3. Flipped interaction pattern
In the flipped interaction pattern, you make the AI ask you questions instead of the other way around. Whether this makes you feel more in control or more likely to experience an AI uprising is up to you—but it works.
Example:
Why it works:
With the flipped interaction pattern, the AI takes the lead in gathering information through structured questioning. It helps you discover blind spots you might not have considered.
Use cases:
The flipped interaction pattern is great for requirements gathering, decision-making, and troubleshooting.
4. Persona pattern with specific expertise
The persona pattern is a pretty common technique. But instead of just telling the AI that it should act as an expert, try giving it even more specific expertise.
Example:
Why it works:
The persona pattern narrows the scope of how an LLM responds. It helps the LLM focus on relevant training data instead of the broader data that powers its operations.
Use cases:
The persona pattern with specific expertise is helpful for professional analysis, industry-specific advice, and technical consulting.
5. Expert panel (multi-persona)
The expert panel pattern takes the persona pattern one step further by asking a panel of experts instead of just one expert.
(Alternatively, you can have your LLM identify the most relevant types of experts for this problem.)
Example:
Why it works:
The expert panel prompt pattern compels AI to perform a thorough analysis that considers the problem from multiple viewpoints. It helps remove bias (to a degree) and often reveals solutions you wouldn't have considered.
Use cases:
The multi-persona prompt pattern is helpful for complex decision-making, strategic planning, and technical troubleshooting. I personally default to this pattern when vibe coding, whether planning the technical stack or requirements, or when I hit a difficult blocker while building.
6. Step-by-step framework (recipe)
The step-by-step framework prompt pattern is also known as the recipe prompt pattern among prompt engineers.
Example:
Why it works:
This framework works because it requires the AI to give you every step: requirements, preparation, instructions, and beyond—just like following a recipe.
Use cases:
The recipe prompt pattern is helpful for developing skills, planning a project, or performing a difficult task. You can also use a modified version to plan an itinerary. I've used it to help with basic trip planning, to surface information about how to get around, where to eat, which attractions to visit (and when), apps to install for tickets, and price ranges to consider.
7. Fact-check list
I will always encourage a human fact-check, but you can have AI do the first go-through to find obvious errors or misleading statements. (I've found that I can put the same text through an AI fact-check multiple times and get different results each time.)
Example:
Why it works:
Using a fact-check list prompt pattern is a good way to help prevent hallucinations when working with AI. If you're really concerned about hallucinations, though, you'll want to use technology, which involves using specific knowledge sources.
Use cases:
The fact-check list prompt pattern can be useful in journalism, research, and business analysis—but again, always follow up with a human fact-check, especially if you're working on sensitive content.
8. Interview/test preparation
The interview and test preparation prompt pattern is a great way to practice for just about anything, using the AI as an expert conversation partner.
Example:
Why it works:
The interview/test preparation prompt pattern uses your knowledge sources as a primary basis. But because AI is able to use its training data to extrapolate, it can come up with questions and angles you may not have considered. It's especially handy for interview prep if you use voice mode in your AI chatbot.
Use cases:
The interview/test preparation prompt pattern helps you prepare for unexpected questions—even if it's not for an interview or test. Since it's personalized based on your requirements, it's more relevant than any random test you can find online.
Use prompt patterns for better outputs
With an understanding of these patterns and how to use them, you can break surface-level AI patterns and get the detailed answers you're seeking. And if that isn't enough, here's a bonus prompt pattern you can use to help in most situations:
Act as a prompt engineering expert. I want to ask you about [topic], but I'm not sure how to phrase it for the best results. Help me rephrase this question by asking clarifying questions to include all the context you'd need: [your original question].
Happy prompting!
Related reading:
