AI Gave You an Answer. Now Review It Like an Accountant with the Right Prompts.

By LaMichelle Hecht, CPA

Last month, I wrote about one of the biggest risks of artificial intelligence: AI can be confidently wrong.

But knowing AI can make mistakes is only the beginning. The next question is more practical: What should an accounting professional actually do with an AI-generated answer?

I use AI every day and have used it regularly since 2022. It can help me research, analyze information, summarize documents, draft communications, organize ideas and work more efficiently. I am not interested in avoiding AI; I’m interested in using it better.

One of the biggest skills accountants need to develop is not just how to prompt AI.It’s how to review AI.

A polished answer is not necessarily a correct answer. A detailed answer is not necessarily well-supported, and a confident answer is certainly not the same thing as authoritative guidance.

Fortunately, accountants already have many of the skills necessary to use AI responsibly. We know how to question information, trace a number back to its source, and ask for support. We know that if something doesn’t make sense, we should keep digging.

The same habits should apply to AI.

Do Not Review AI Like an Editor. Review It Like an Accountant.

When AI produces a memo, email, analysis or technical response, it’s easy to focus first on how the answer sounds. Is it written professionally? Does the grammar look right? Does the explanation flow? Those things matter, but they are not where I start.

I want to know: What is AI actually claiming?

If the answer says an expense is allowable, a transaction should be recorded a certain way, a filing is required, or an accounting standard applies, those statements deserve attention. Instead of simply asking AI for an answer, ask it to identify exactly what it knows and assumes.

This prompt can completely change the quality of a response. You may discover the AI assumed the organization uses accrual accounting, that a grant is unconditional, that a transaction occurred in the current year, or that federal rules apply when the issue is actually governed by state law.

The answer may still be useful. But now you know what to verify.

Ask AI to Argue Against Itself.

One of my favorite ways to use AI is to make it challenge its own answer.

People naturally look for information that supports the conclusion they already believe. AI can reinforce that tendency when we continue prompting it in the same direction. You can guard against this by changing direction. If AI gives you a conclusion, ask why that conclusion might be wrong.

This does not mean the opposite answer is correct — you are simply forcing AI to expose weaknesses in its original reasoning. For accountants, that should feel familiar. It’s professional skepticism applied to technology.

Make AI Tell You What’s Missing.

Generative AI tools are designed to respond, which can become a problem when the correct answer should be, "I do not have enough information yet."

Imagine asking: "Is this grant expense allowable?" The question sounds simple, but the answer might depend on the grant agreement, funding source, approved budget, period of performance, cost allocation method, prior approvals and other facts.

AI may still give you an answer even when much of that information is missing. You can reduce that risk by telling it not to fill in the blanks.

This prompt is especially useful for technical accounting, tax, audit and grant compliance questions. Sometimes a better AI response is not an answer. It is a list of better questions.

Ask AI to Show You the Authority.

This is where accounting professionals need to be especially careful. AI can summarize technical guidance effectively, but it can also cite the wrong guidance, misinterpret guidance, or occasionally reference something that does not exist.

So, if the answer involves technical accounting, tax, auditing or regulatory guidance, ask AI to identify the authority supporting the conclusion.

Then comes the most important part: Actually check it. Do not treat the citation as verified simply because AI produced one. Go to the source. For tax issues, review the applicable IRS guidance or other primary authority. For accounting issues, review the applicable accounting standards. For audit questions, review the professional standards. For government awards, review the award documents and applicable regulations.

AI can help you locate information faster. It should not be the authority itself.

Use AI as a Second Reviewer.

AI becomes even more useful when you stop asking it only to create things and start asking it to review things. Imagine you have already drafted an accounting memo. Instead of asking AI to rewrite it, ask AI to review it the way a manager might.

That last sentence matters. If you immediately ask AI to rewrite the memo, it may fix the language without helping you understand what was wrong. Reviewing the comments forces you to examine the underlying issues first.

Make AI Look for What You Missed.

Sometimes I do not want AI to tell me whether my conclusion is correct. I want it to identify what I forgot to consider. That is a different question.

You can use this when reviewing financial statement issues, grant compliance, internal controls, tax positions, audit findings, contracts, policies, client explanations, etc. The goal is not to outsource judgment, it’s to expand the number of things you consider before exercising judgment.

Turn the Answer into a Review Checklist.

A key benefit of AI is its ability to reorganize information. If you receive a long analysis, do not leave it as five pages of paragraphs. Make it easier to review.

Now you have something more like a workpaper than a chatbot response, which also makes the output easier for another accountant to review.

The Final Question: Would You Defend It Without Mentioning AI?

Before relying on an important AI-generated answer, I like one simple test:

Would I be comfortable defending this conclusion if I could not say, “AI told me so?”

Because ultimately, our clients are not hiring artificial intelligence, they’re hiring us.The audit committee is relying on us. The board is relying on us. The taxing authority, regulator, granting agency or lender will expect us to explain the conclusion.

AI does not change that responsibility. It changes how quickly we may be able to reach a conclusion, which is why the accountants who benefit most from AI will not necessarily be the people with the fanciest prompts. They will be the people who know when to stop, question the answer, check the source and ask: What am I missing?

AI can make us faster. Professional skepticism is what keeps faster from becoming careless. And sometimes the best prompt you can give AI is not "Give me the answer."

It is “Show me why I should trust it.”