ChatGPT Prompts + Tips for Accountants

Your CFO has tapped you to lead the charge in bringing AI to the accounting team, but there’s a catch – there’s no budget for new software. As the controller, you’re pretty comfortable working with limited resources, but you have little familiarity with existing AI tools, especially free ones. 

So, what solutions remain?

Enter ChatGPT. With its meteoric rise in popularity, you’ve likely heard of the product; maybe you’ve even used it for one-off use cases like writing an email or sourcing suggestions for new reading material. While ChatGPT excels at accomplishing these general tasks, it also performs well in completing a number of qualitative and quantitative functions within accounting workflows. 

With just a little bit of know-how, you can turn ChatGPT from an amusing chatbot into a highly useful accounting sidekick. 

What is ChatGPT? 

Developed by OpenAI and initially released in late 2022, ChatGPT is a large language model (LLM) that has been trained on billions of pages of data in order to generate human-like content. 

While several competing AI models exist (we covered them all for you in our AI explainer), ChatGPT is easily the most popular and readily accessible—that being said, the tips in this guide are fair to apply to all chatbots.

How to put ChatGPT into accounting mode 

People interact with ChatGPT through “prompts”, which are questions or requests that the system interprets and responds to. For example, if you enter prompts such as “Give me a recipe for how to bake a chocolate cake,” or “When did the first Star Wars movie come out?”, ChatGPT will answer questions like these with ease. While these prompts and responses are useful, they are not very different from a casual Google Search. 

ChatGPT strongly adheres to the classic maxim: you get out what you put in. 

Many people might not realize just how much information they can provide to the chatbot to guide the model’s decision making and narrow the response they hope to receive. This technique is called priming – adding context to a prompt to receive a more targeted response.

For instance, let’s say you wanted to ask ChatGPT a simple technical accounting question. You ask, 

It returns the following:

Yikes. That’s not a useful response. But no need to give up on ChatGPT (or AI more broadly) yet.

Instead, a better developed prompt would inform ChatGPT of the writing style, tone, and perhaps even identity that the response should reflect. Prompts with good priming tend to begin by directly informing ChatGPT of what role it should adopt – ex. “I want you to act as an experienced marketer.” From there, it’s best to specify what output you want to receive – article, bullet points, table of data, etc. – and if possible, even drop in an example response of what the model should try to produce. Look at this revamped prompt below:

Here’s the response:

Huge difference. Not only is the current response relevant, but it also addresses the question in the concise, well-informed manner requested. Leaning into this specificity allows large language models to deliver the impressive content they are capable of producing.
‍

Purpose-Built Accounting AI

ChatGPT is great for quick references and as a general sidekick. But it pulls from the entirety of the internet for responses and requires strong priming.

For future accounting questions, you can avoid priming altogether by using Numeric’s technical accounting AI. Trained on GAAP guidance, public company filings, and more, the accounting AI bot can provides answers specific  for any standard (US GAAP, IFRS, UK GAAP) and for guidance from any Big 4 audit firm. 

From policy memo generation to quick ASC 606 clarification, accountants leverage Numeric’s technical accounting AI to get pointed in the right direction immediately.

How Can Accountants Use ChatGPT Better? 

Trust but verify – people and AI

Nobody’s perfect. With people, that’s easy to see. Computers? Less so. 

ChatGPT’s radical innovation derives from its speed and ability to cull loads of information on even seemingly niche topics. Unfortunately, this doesn’t always prevent the model from generating false information amid what otherwise might be perfectly suitable content. 

Developers refer to this phenomenon as hallucination, and while this shouldn’t completely undermine one’s trust in artificial intelligence or specifically, LLMs, people should acknowledge that chatbots are still improving at recognizing their own knowledge gaps. 

It’s no different than humans: if even subject matter experts can get things wrong from time to time, it’s reasonable that computers would too. 

Reduce hallucination with narrow source material 

ChatGPT, while powerful, is susceptible to getting a bit over its skis when it comes to providing accurate information at scale.The risk of such inaccuracies increases when your topic is too broad or when you provide too many parameters for the model to consider in forming its response.

To mitigate this risk, it's advisable to limit the scope of your prompt. One effective strategy is to upload specific documents, tables, or images that you want ChatGPT to reference directly. For instance, if you need assistance drafting a policy for lease accounting, consider uploading a PDF of an existing policy from your company. By doing so, you provide ChatGPT with precise, relevant information to work with. This approach typically yields more accurate and tailored responses compared to relying on the model's training data, which may include outdated or unreliable information from various internet sources.

Use the model to read, write, and format table data   

If accountants can consider themselves to be spreadsheet connoisseurs, then ChatGPT is a kindred spirit. 

When performing data-heavy work, one neat tip for accountants is that you can ask ChatGPT to format any data outputs as a table – a function well-suited for merging or transferring data across software. Similarly, if providing a table, you can ask ChatGPT to run analysis on the dataset in the listed cells: this feature can streamline accounting processes like flux analysis by writing plausible first-draft explanations that summarize the variance within an account. Similarly, it can help to get a headstart on forecasting for future months. 

Exercise ChatGPT’s coding & formatting abilities    

While ChatGPT’s content prowess is highly evident, its coding & mathematical capabilities are equally impressive. Between generating code across coding languages to performing nearly-instant data analysis, developers get just as much mileage out of chatbots as more non-technical roles. 

For calculations, both simple and complex, ChatGPT is a far superior option to breaking out the calculator to perform manual calculations. 

Save everything for audit trail 

With audits, you can never be overly prepared – any and all documentation helps a team’s cause. When using ChatGPT, the same principle applies: teams should document any prompts, scripts, and responses used to generate content that informs accounting tasks & processes. 

By maintaining a comprehensive audit trail, your team can ensure compliance, facilitate reviews, and provide clear evidence during audits.

What are the best ChatGPT prompts for accountants?

Here’s a list of sample prompts that accountants can use in their month-end close. As one can see, ChatGPT can serve a number of accounting practices, such as categorizing invoices by account, analyzing financial reports, drafting emails for client communication, combining accounting data from different sources, and plenty more. 

Before asking any question, consider using some priming such as this,

"I want you to act as an expert accountant who is also an [elite copywriter / experienced bookkeeper / IFRS specialist]. You have been hired by [insert company]— a [insert business case] to help their accounting department to [craft internal emails / draft journal entries / create new policies]. Your writing style is to be short and to the point, and you should produce responses [in bullet point format / in paragraphs]. Here is your first task"

1. Checklist Creator 
   • Ex: Create a checklist for the account reconciliation process for two people for both the month-end close and for ongoing reconciliations.
2. Adjusting Entries 
   • Ex: Identify and create adjusting entries based on my trial balance and [insert workpaper] for [month/quarter/year]. 
3. Excel Assistant 
   • Ex. For these cells below, combine text from multiple cells into one cell and render your output as a table. 
4. Data Extraction & Bookkeeping 
   • Ex. With the invoice information provided below, extract invoice details and create journal entries to add to [insert accounts]. 
5. Email Writing 
   • Ex. Write a first draft of an email to the marketing team requesting additional information about [insert transaction]. Keep your response under 500 words. 
6. Tax Insights
   • Ex. Analyze the impact of recent tax law changes on the company’s tax liability and suggest potential tax planning strategies. 
7. Financial Data Analysis 
   • Ex. Can you prepare a detailed analysis of the company’s income statement for the last fiscal year, highlighting key revenue streams and major expense categories? 
8. Policy Writer/Reviewer
   • Ex. Our legal team just updated this contract, do we need to make any changes to our 606 policy? 
9. JE Writer
   • Ex. If I buy equipment for $8,000,000 to be paid over 3 years with an interest rate of about 4%, the PV is $7,600,000 and the total cash that is paid including interest is $8,300,000, what is the journal entry to record the assets and the liability?
10. GAAP/IFRS Discovery
    • Ex. Which ASC number deals with the treatment of unbilled receivables? 
11. Accounting Theory/Hypotheticals 
    • Ex. Is there a scenario where I can expense goodwill and other assets as a part of small acquisition? 

What’s the best ChatGPT model for my team? 

3.5? 4? 4o? 4o Mini? 

As OpenAI continues to revise their ChatGPT models, they develop newer versions that vary in computing power, language understanding, and accessibility. If you’re unsure which option is best for your accounting team, don't worry.

For the average user or accountant, the differences between models are minimal and will not significantly affect ChatGPT's performance. Currently, free users typically interact with ChatGPT 3.5, while ChatGPT Plus subscribers access GPT-4. Both models are fully capable of handling most tasks.

Subscribers to ChatGPT Plus, OpenAI’s $20/month premium offering, gain access to GPT-4, which is considered the most advanced model. This plan is generally recommended for writers and researchers who need high-volume, daily interaction with ChatGPT or require enhanced generative AI capabilities for more complex tasks.

And for purpose-built accounting AI, take Numeric’s Technical Accounting AI for a spin to get responses trained narrowly on accounting source material.