9 tips to write more effective AI prompts

Artificial intelligence (AI) can transform accounting workflows if you can translate what you need done into commands AI can understand.  

“If you learn how to prompt an AI engine, you get radically better results,” said Randy Johnston, CEO and founder of Network Management Group, Inc., a Kansas-based IT consultancy focused on public accounting.

According to Johnston, effective prompt writing is a must-have skill for CPAs navigating everything from tax research to client communications. By following a simple structure, being clear and concise, and continuously refining your approach, CPAs can dramatically improve the quality of AI-generated outputs.

Johnston will talk about prompt-writing in a session June 10 at AICPA ENGAGE 2026. Here are his nine most practical, field-tested tips:

1. Start with a simple, repeatable approach

Despite the complexity of modern AI systems, Johnston emphasizes simplicity. His recommended approach boils down to four core components:

• Task: What exactly do you want the AI to do?
• Context: What background or data should it consider?
• Expectations: What specific details or criteria should it meet?
• Output: What format should the response take?

For example, instead of prompting, “Analyze this data,” try instead, “Analyze 2024 sales data for Southeast region clients, identify trends in revenue growth by industry, and present the findings in a bullet-point summary.” That level of clarity dramatically improves results.

“If you do those four things, you’re going to be way down the path most of the time,” Johnston said.

2. Use ‘uber prompts’ for better efficiency

Beginners often submit each step in a request separately to the generative AI engine. Once you’re comfortable using AI, Johnston recommends switching to combining everything into a single, well-structured prompt.

“I believe that you’re better off with an uber prompt that has those four components than going one step at a time,” he said.

This approach reduces back-and-forth and produces more-complete outputs in a single interaction, which is especially valuable in time-sensitive environments like audit workflows or client deliverables.

3. Be specific and keep it concise

A common mistake CPAs make is being either too vague or too wordy. Johnston stresses the importance of precision. Being specific when you’re issuing a prompt is critical, but you should use the minimum number of words that you can.

Overly long prompts can dilute intent and vague prompts lead to generic outputs. The goal is a concise, targeted instruction that leaves little room for ambiguity.

4. Provide examples to improve output quality

If you want AI to match your tone, style, or structure, show it what “good” looks like. Johnston frequently uses example-based prompting.

“When I want to have something written, I often will provide an example of my writing and say: Write about this in my style,” he said.

This technique, often called “few-shot prompting,” is especially useful for client communications, internal memos, and thought leadership content.

By anchoring AI in real examples, you can generate outputs that align more closely with firm standards and personal voice.

5. Refine prompts through iteration

Even strong prompts may not deliver perfect results on the first try. Johnston encourages a continuous refinement process.

“Generally, with the refinement process, you look at the result and then figure out if there is a different way you can say it,” he said.

If something goes wrong, revisit the four components:

• Is the task unclear?
• Is the context incomplete?
• Are the expectations too vague?
• Is the output format underspecified?

Small adjustments in any of these areas can significantly improve results.

6. Start fresh when needed

Johnston said that AI performance can degrade over extended interactions. “These AI engines do tend to get tired, so it often helps to start a new session,” he said.

If outputs become inconsistent or irrelevant, resetting the conversation can restore quality and save time and frustration, he said.

7. Build a library of high-quality prompts

One of the most powerful ways to scale AI productivity across a firm is to capture and reuse successful prompts.

“If you can store those prompts for everybody else to use, you’ll get the productivity of your team up in a hurry,” Johnston said.

He recommends storing prompts in shared tools like spreadsheets or collaborative lists, allowing teams to standardize workflows, reduce duplication of effort, and share best practices across departments. Leading AI platforms can store projects to supply base information on your firm, approach, and personality to the AI engine.

Over time, prompt libraries can become a valuable internal knowledge base of proven AI interactions.

8. Match the tool to the task

Different AI tools excel in different areas, Johnston said. Some are stronger in tax research, others in financial analysis or forecasting, and still others in writing and communication.

Privacy and security considerations should influence tool selection, he said. For example, many firms prefer more controlled environments for sensitive data.

He recommends adopting a “1-2-3 strategy”:

1. A primary AI platform
2. Specialized tools for specific tasks
3. Agent-based solutions for automation

This multi-layered approach allows firms to balance performance, accuracy, and security.

9. Be aware of accuracy limitations

AI can be incredibly useful, but Johnston cautions against overreliance, especially in high-stakes areas like tax. The problems, he said, come with the large language models (LLMs) that power generative AI.  “Even though firms are trying to use the LLMs, the accuracy and precision of AI engines for a defensible position is not very good,” he said.

For general questions, AI can “get you in the ballpark.” But for nuanced, jurisdiction-specific guidance, errors are still common. That’s why many firms continue to rely on specialized, authoritative tools for final answers.

“Make sure you use your professional skepticism regarding the results,” he said.

AI outputs should never be accepted at face value. CPAs remain ultimately responsible for the accuracy and integrity of their work.

“It can’t leave your hands wrong,” Johnston said. “I see things produced all the time by AI and they’re wrong, but they slipped through the vetting process.”

The growing divide in AI skills

As AI adoption accelerates, Johnston predicts a widening gap between professionals.

“Accountants that are using AI in a sophisticated way will radically outperform accountants that are using AI in an unsophisticated way,” he said.

Those who master prompting will unlock significant efficiency gains, while those who don’t risk falling behind. The good news: You don’t need to be an expert to benefit.

“You can get a whole lot of leverage without the super skills,” Johnston said. “But developing those skills can elevate not just individual performance, but entire teams.”

Learn more tips from Johnston during his session at AICPA ENGAGE 2026: AI – Better Prompts, Better Results, on Wednesday, June 10, from 9–9:50 a.m. PT (Noon–12:50 p.m. ET) in Las Vegas and live online.

— Hannah Pitstick is a content writer for the AICPA and CIMA. To comment on this article, contact Jeff Drew at Jeff.Drew@aicpa-cima.com.