What Are AI Prompts? (& How to Make Them Better)

Note: This blog post was created around this interview with Anna Bernstein, prompt engineer at Copy.ai.

What are AI prompts?

An AI prompt is a question, command, or statement that you input into an AI model to initiate a response or action, harnessing the power of natural language processing.

The prompt serves as the starting point for the model's generation process, directing it towards generating content that is relevant to what you're asking or stating. With the rise of AI tools, the quality of the prompt can greatly affect the quality and relevance of the output generated by the AI.

In other words, a prompt is your way of instructing the AI on what you want it to do.

This could range from asking it to generate text, solve a problem, answer a question, and so on. Advanced AI models also cater to tasks such as image prompting or text to image models.

For instance, if you are using a text-based conversational AI like Copy.ai’s Chat, an example of an ai tool, you might use prompts like:

• "Tell me a joke."
• "What is the capital of France?"
• "Write a short story about a talking cat."

The AI interprets your prompt and attempts to generate a suitable and relevant output based on its training data and algorithms.

Two qualities of a strong AI prompt

When diving into the world of generative AI and ai tools, one of the key factors for success is crafting the perfect prompt. With millions of neurons buzzing in these advanced algorithms, it's crucial to know how to harness that power effectively.

This is where the expertise of prompt engineers and AI prompt engineering comes into play.

And that boils down to two critical qualities your prompts must have: they need to be both "rich" and "specific."

In today's post, we'll dive deep into what these terms mean, why they're crucial, and how they can set you up for a rewarding AI experience using various tools, including the generative ai tool.

Rich prompts

What do we mean by a 'rich' prompt?

In the simplest terms, a rich prompt is like painting a vivid picture with your words. It's about giving the AI model a full contextual palette to work with.

For those in the field of AI prompt engineering, mastering this concept is crucial.

For example, instead of asking the AI to "Write an email introduction," elevate your game by requesting, "Write an enthusiastic yet professional email introduction aimed at a prospect in the tech industry using your expertise in natural language processing."

See the difference? With the latter, you're equipping the AI with the crucial input data that enable it to craft a message that's on point.

Specific prompts

Now, on to our second factor in quality prompts: specificity.

While richness provides context, specificity zooms in on the task. It's all about delivering a clear-cut instruction that leaves no room for ambiguity.

If your prompt is vague, you're essentially asking the AI to fill in the blanks, and let's be honest, nobody wants a machine making assumptions about their work. This is where prompt engineers really shine, ensuring that AI tools have specific instructions to follow.

So, instead of vaguely asking for a "blog post about artificial intelligence," specify, "Write a 500-word blog post highlighting three recent advancements in artificial intelligence for a general audience using text to image model techniques."

Richness and specificity aren't standalone concepts, either; they work best in tandem. Think of them as two sides of the same coin. It's important for those who create prompts to understand this.

Richness sets the stage, providing the AI with a comprehensive backdrop.

Specificity, on the other hand, is like your stage directions, cueing the AI on what precisely to perform.

When both these elements are in harmony, you're not just asking the AI to generate content; you're commanding it to produce the exact content you desire using its AI models.

When do you know if your AI prompt is any good?

To know if a prompt is rich and specific, evaluate a few key factors related to AI prompt engineering.

First, does the prompt contain descriptive language with emotive words that paint a vivid picture? Rich prompts are packed with details that help AI tools understand the exact tone, style, and content you want.

Second, is the prompt precise in its wording? Specific prompts avoid ambiguity and clearly state what you want the AI to generate.

They label context and examples appropriately so the AI, backed by natural language processing algorithms, understands how to use that input data.

Specific prompts also avoid pronouns like “it” or “this” without properly establishing what those refer to.

Finally, do I have to make inferences or fill in gaps to understand the prompt? If so, the prompt likely isn’t specific enough.

The more assumptions the AI (or a human) has to make, the more opportunities there are for “covert failures” where the output isn’t quite what you intended. This is especially important when you're writing code to automate AI tasks.

The ideal prompt anticipates potential areas of confusion and addresses them to ensure the meaning is crystal clear.

Mastering this art can make you akin to prompt engineers who specialize in crafting optimized queries for AI models.

Should you add examples in your prompts?

Yes, providing examples when using subjective terms in your AI writing prompts can be very helpful.

That's because terms like “catchy,” “high quality,” or “effusive” can mean different things to different people. While AI writing assistants are getting better at understanding nuance and context, they still benefit from clear examples when working as a text to image model or generative AI tool.

Without examples, AI may struggle to determine exactly what you mean by a subjective descriptor.

It has to spend extra effort trying to infer your meaning, which takes away from its ability to actually generate the content you want. Providing examples serves as additional input data for the AI.

Trial and error is part of the prompting process

With practice, prompting AI writing assistants does get easier.

But providing good examples remains an important technique, especially when using subjective language, to help communicate your vision and set the AI up for success.

Examples, combined with rich and specific language, are key to crafting prompts that will yield the results you want.

Anna, the prompt engineer at Copy.ai, notes that when she first started prompting AI systems, the initial prompts she entered often needed more refinement to produce the target response. Now, after much more experience, the prompts she starts with are typically much closer to what will generate a good result.

She has developed an intuition for how to construct prompts that will resonate with various AI models, including text to image models.

For example, she finds prompts that ask the system to “use this to do that” tend to work better than just stating “this is X, do Y.”

Developing a list of adjectives and examples that work well for the system, using consistent language to refer to the same concepts, and framing prompts as instructions for a very literal but intelligent intern can all help generate better results with less tweaking over time.

With regular use, prompting AI systems can become second nature, especially as you become proficient in the art of AI prompt engineering.

Examples of rich and specific prompts

1. Vague Prompt: "Tell me about sales."

Refined Prompt: "Can you explain the primary stages in a B2B sales process and the strategies employed at each stage?"

Why: The original prompt could lead to a generic response about sales in various sectors and contexts.

The refined version narrows the focus to the B2B (Business-to-Business) sales process, providing a more specialized answer and detailing the strategies at each stage.

2. Vague Prompt: "How do I market online?"

Refined Prompt: "What are the key components of an effective digital marketing strategy for a startup in the e-commerce fashion industry?"

Why: "How do I market online?" is broad and can pertain to any industry or objective. By refining it, we focus on a specific sector (e-commerce fashion) and a specific business stage (startup), which provides guidance on how to frame the answer.

3. Vague Prompt: "How do ads work?"

Refined Prompt: "Could you explain the mechanics and strategies behind creating a successful pay-per-click (PPC) ad campaign on Google Ads for a new product launch?"

Why: The original question is too general, as "ads" can span from traditional print ads to modern digital ads.

The refined question targets a specific form of online advertising (PPC on Google Ads) and situates it in the context of a new product launch.

4. Vague Prompt: "Help me with branding."

Refined Prompt: "What are the essential steps to develop a cohesive brand identity for a health and wellness company targeting millennials?"

Why: "Help me with branding" is unclear about the exact need or target audience. The refined prompt specifically seeks guidance on creating a brand identity, specifies the industry (health and wellness), and defines the target demographic (millennials).

5. Vague Prompt: "How do I get leads?"

Refined Prompt: "What are some proven inbound marketing techniques to attract and convert high-quality leads for a software-as-a-service (SaaS) company?"

Why: Simply asking about getting leads can be answered in countless ways, depending on the industry, audience, and method of sales. The refined question emphasizes inbound marketing and further specifies the context (SaaS), making the expected answer more actionable and relevant.

How to refine your prompts when your output isn’t up to par

The way you phrase your question can have a big impact on the quality and relevance of the answer you receive.

Here are some tips on how to refine your prompts when the output isn't what you were hoping for:

1. Be specific

A general or vague prompt may yield a similarly vague answer. Try to be as specific as possible about what you're looking for.

• Vague: Tell me about cars.
• Specific: Can you explain the advantages and disadvantages of electric cars over traditional gasoline cars?

2. Specify the context or criteria

If there are specific conditions or constraints that should be considered in the answer, make sure to include them in the prompt.

• Without Context: How do I fix my computer?
• With Context: How can I troubleshoot a Windows 10 computer that won't start up?

3. Indicate the desired output format

If you have a preference for how the answer is formatted (list, paragraph, step-by-step instructions), mention that in the prompt.

• No Format: Tell me about time management.
• With Format: Can you provide a list of effective time management techniques?

4. Ask follow-up questions

Sometimes, one question may not be enough to get the detailed answer you're looking for. Feel free to ask for clarification or ask follow-up questions to get more depth or breadth.

• Follow-Up: Can you elaborate on the second technique you mentioned?

5. Experiment with different phrasing

The way you word your question can influence the answer. If you're not getting what you want, try asking the same question in a different way.

• Original: How do I make money?
• Alternate: What are some proven strategies for increasing my income?

6. Use keywords

If there are certain terms or jargon related to the subject you're asking about, including them may help you get a more targeted answer.

• Without Keywords: How do I get better at the game?
• With Keywords: What strategies can improve my chess endgame?

7. Limit or broaden the scope

Sometimes your question may be too broad or too narrow, affecting the usefulness of the answer. Adjust the scope of your question as needed.

• Too Broad: Tell me about history.
• Too Narrow: Tell me about the War of Jenkins' Ear.
• Adjusted: Can you summarize the major conflicts in European history during the 18th century?

8. Check for Ambiguity

Make sure your question can't be interpreted in multiple ways, or you may not get the answer you're expecting.

• Ambiguous: How do I get better?
• Clear: How can I improve my writing skills?

By carefully considering how you phrase your prompts, you're more likely to get the output that meets your expectations.

What's the easiest way to improve your prompts?

If you're just getting started with generative AI and working with large language models, you might feel intimidated at times.

That's why Copy.ai came out with the Improve feature.

Here's the prompt before:

And here's the same prompt after clicking Improve:

This allows you to quickly and easily improve any of your prompts at the click of a button. Once you've entered the prompt, simply click "Improve" and watch as Copy.ai makes it more rich and specific.

Plus, you can generate the output in a specific Brand Voice (and with a Pro account, you get unlimited Brand Voices to work with).

This allows you to generate higher-quality output in two ways: by improving the prompt at the source and by guiding the output to use a specific tone that matches your brand.

Ready to get started? Sign up and give Copy.ai a try today!