Have you tried using ChatGPT for communications at your accounting firm yet? If you have, you might have noticed that sometimes its outputs aren’t quite what you were looking for. But fortunately, there might be a solution: better prompt engineering.

Prompt engineering aims to get the most out of generative AI by ensuring that all prompts contain the right information, context, and direction for more targeted outputs.

Below, we’ll cover the state of prompt engineering today, plus 7 tips for creating better prompts.

The state of prompt engineering in 2024

Last year, the World Economic Forum declared prompt engineering one of the top jobs of the future. Around the same time, OpenAI’s Andrej Karpathy wrote on Twitter: “The hottest new programming language is English.”

A year later, there’s more skepticism around prompt engineering as a career path than there was before. AI models are becoming more adaptive, and better at autonomously generating context they need to deliver an appropriate response. There will still be skill involved in creating good prompts, but skeptics argue it will be a commonplace skill that will quickly become “table stakes.”

However, that’s good news: it means you don’t necessarily have to hire a prompt engineer to get better output from generative AI. Below, we’ll share 7 tips for prompt engineering that accountants can use today.

7 prompt engineering tips for accountants

If you give ChatGPT a vague instruction, you might notice that it will tell you to be more specific. 

For example, when you ask it “write an email to collect on an overdue bill from a client” it will respond with:

“Absolutely, I can help you draft that email. To make it as effective and appropriate as possible, could you please provide me with a few details? Such as the nature of the service or product provided, the duration for which the bill has been overdue, and any previous communication you've had with the client regarding this matter. Additionally, let me know if you'd like the tone to be firm, polite, or a mix of both.”

So to some extent, there are already guardrails in place to help guide you to write better prompts. But AI can also summarize, extract data, help you plan, brainstorm, and assist with high-level strategic tasks. Each type of task will require you to take a different approach to prompts to help you get the most out of generative AI.

Below are 7 general guidelines to follow:

1. Be specific and detailed

The one overarching rule in prompt engineering is to be specific and detailed, to help generative AI understand what you’re really asking for.

To get into the right mindset for prompt engineering, think of how you’d describe the task if you were assigning it to a stranger who didn’t know your business well. What details would you give them?

2. Context is important

The more context you give ChatGPT, the more the output will match your expectations. For example, start with a hypothetical role for it to play so it understands what tone and stance to adopt.

If you’re writing client-facing communications, you could also give it a short background on your brand and style guides. For example, a firm focused on tax filing for individuals might have a less formal tone than a B2B accounting firm.

3. Clarify your “why”

This one is especially relevant if you’re using ChatGPT for higher-level strategic tasks like brainstorming and business planning. When you write your prompt, you can use the “Five Whys” technique to drill down into the root cause of the problem.

The Five Whys method can help you steer clear of assumptions and identify processes you’d like to explore to solve your problem, which will help you create clearer prompts.

4. Ask one thing at a time

ChatGPT can handle multi-part requests, but specifying one task will usually give you more precise and focused outputs. If you can, break multiple tasks into their own prompts. For example, if you want ChatGPT to write an outbound email and then personalize it for a few different client segments, ask it to write the general version first. 

5. Specify the format

When you’re writing a communication, internal training document, or other specific communication, getting the format right is key.

For prompts, that means not only specifying what the document is (job offer, engagement letter, landing page etc) but the format you’d like the text in. Include details about the relative length and structure, such as where to place lists, summaries, and headers.

6. Update with feedback

If an initial response isn’t what you’re looking for, ask for the modifications you want. For example, you might get an email template that’s way longer than what you had in mind. Asking ChatGPT to shorten it to two paragraphs will fix the issue and help you remember what to ask for up front next time.

7. Keep up to date on new developments

In September of last year, OpenAI announced ChatGPT’s new multimodal capabilities – text-to-image and speech recognition features that broadened the horizons of what it could accomplish.

Keeping up to date on developments like these can help you get the most out of new features and stay on top of your generative AI game.

A few final points about prompt engineering

As HBR notes, prompt engineering is really about identifying and framing a problem correctly in order to find a solution. It’s up to you to understand the dynamics of a task in order to guide the AI correctly.

Also note that it’s important to include hypotheticals rather than real data in your prompts to avoid any security breaches.

And finally, remember that once you find a good prompt, you can save it for later! If you’re using ProCharted, you can save relevant prompts as documents or notes in client and work files so that your communications team can access them whenever they need to.

So go ahead and start generating! The more you create great prompts, the more value you’ll be able to get from your favourite generative AI tools.