AI for accountants: what it does well and where to be careful

Most accounting work is a mix of judgment and handling. The judgment part, deciding how a transaction is treated or what a client should do, is yours. The handling part, which includes reformatting exports, chasing documents, writing the same explanation for the tenth time and assembling packages, is where AI earns its keep. This page covers where that line falls as of September 2026.

Where AI helps accountants

Cleaning data before you reconcile. Bank exports, payroll files and ledger dumps arrive with dates as text, inconsistent vendor names and amounts with stray symbols. An assistant can standardize these in a new column and list the rows it could not resolve. The Excel guide walks through the habits that make this safe: work on a copy, add columns rather than overwriting, and check the row count and a control total afterward.

Drafting client communication. Requests for missing documents, engagement-letter follow-ups and explanations of a balance are formulaic. Give the assistant the facts and your tone and it produces a draft in seconds. You still read every one before it goes out.

First-pass commentary. Variance analysis is a good fit. Provide the two periods and ask for the five largest movements with possible explanations to verify. The explanations are hypotheses, not findings.

Summarizing long documents. Leases, loan agreements and vendor contracts hide the terms that affect the books. Ask for a list of dates, amounts, renewal terms and obligations, with the page or clause of each, so you can confirm quickly.

Repeatable reporting. The largest gain is building a process once so that next month’s export produces the same package. The weekly reports question covers the basic pattern.

Where AI does not help

• Arithmetic and totals. Language models can miscalculate or invent figures. Ask them to write formulas or run calculations in a spreadsheet, then verify with a control total. Never accept a total that only appears in prose.
• Tax and accounting positions. An assistant may cite a rule that is outdated, wrong for the jurisdiction or does not exist. This is a hallucination. Use it to structure your research questions, not to answer them.
• Anything you cannot check. If you could not verify the output yourself, do not send it.

Risks specific to accounting

Client confidentiality. Financial records, tax IDs and payroll details are sensitive, and your engagement letters may restrict where they go. Before pasting anything, read whether it is safe to put company data in ChatGPT and check your firm’s policy. A practical rule is to remove names, account numbers and tax IDs when the task does not need them.

Data defaults differ by plan. Consumer and business plans handle data differently. Anthropic’s documentation says consumer Claude conversations are used for model improvement only if you allow it, while business products fall under separate commercial terms. Check the current setting for whatever tool you use and record your choice.

Professional responsibility. You sign the work, so the review step cannot be delegated. Keep a note of which steps used AI and how you checked them.

Starter prompts

Data cleanup:

Here is a bank export with columns Date, Description, Amount.
1. List every formatting problem you see (mixed date formats, text
   stored as numbers, duplicate rows). Do not change anything yet.
2. Then propose a cleaned version in new columns. Flag any row
   where you were not sure, instead of guessing.

Variance commentary:

Below are two months of trial balance data. Identify the five largest
changes by dollar amount. For each, give two or three possible
explanations I should check with the client, and state what
document would confirm each. Do not assert a cause as fact.

Your first week

• Day 1: Read your firm’s AI policy, or draft a one-page version using this question as a starting point. Decide what data is off limits.
• Day 2: Take a de-identified export and try the cleanup prompt. Compare the result against your own tie-out.
• Day 3: Use it to draft three routine client emails. Time yourself against writing them from scratch.
• Day 4: Summarize one lease or loan agreement and verify each item against the source.
• Day 5: List your three most repetitive monthly tasks and pick one to turn into a repeatable workflow.

The point of the week is calibration. You learn where the tool is reliable for your kind of work, and where you always need to check.

A worked example: month-end bank rec prep

Suppose a client sends a bank export with 400 rows every month. Your current process is to sort, fix dates, match to the ledger and chase the leftovers. With an assistant, the sequence changes. First, work on a copy and record the row count and the sum of the Amount column. Second, ask for a list of formatting problems without changes. Third, ask for cleaned values in new columns. Fourth, ask it to propose matches to the ledger by amount and date window, and to list everything unmatched. Fifth, re-check the row count and total.

The assistant does not post anything. You review the proposed matches, and you resolve the leftovers with the client. The time saved is in the mechanical passes, and the accountability stays where it was.

Common mistakes to avoid

• Trusting a total in prose. Ask for the formula or the calculation, not just the answer.
• Pasting identifiers you did not need to. If the task is about amounts and dates, strip names and account numbers first.
• Skipping notes on what was AI-assisted. If a reviewer or regulator asks how a figure was produced, you want to be able to say.
• Automating before you understand the manual version. Do the task by hand once, then let the tool repeat what you can already verify.

What the sources say

Figures from the Bureau of Labor Statistics, the PCAOB (the US audit regulator) and the IRS on accounting work, AI and client data.

• The Bureau of Labor Statistics projects employment of accountants and auditors to grow 5 percent from 2025 to 2035, faster than the average for all occupations.
• The Bureau of Labor Statistics says the automation of routine tasks, such as data entry, is not expected to reduce overall demand for accountants.
• Staff of the Public Company Accounting Oversight Board wrote that integration of generative AI in audits at the firms it spoke to appears to be focused primarily on administrative and research activities.
• Staff of the Public Company Accounting Oversight Board report that firms view human involvement as essential for auditors and preparers, including to review generative AI output.
• The Internal Revenue Service says the definition of financial institutions includes professional tax preparers.
• The Internal Revenue Service says the Safeguards Rule requires companies to develop a written information security plan that describes their program to protect customer information.

If you want help building these

Learn AI Fast is a two-day in-person class in Boston where non-developers practice spreadsheet, email and report automation on provided case files.

Sources

• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Accountants and Auditors: “projected to grow 5 percent from 2025 to 2035, faster than the average for all occupations”
• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Accountants and Auditors: “this change is not expected to reduce overall demand”
• PCAOB, Spotlight: Staff Update on Outreach Activities Related to the Integration of Generative AI in Audits and Financial Reporting (July 2024): “integration of GenAI in audits conducted by the firms we spoke to appears to be focused primarily on administrative and research activities”
• PCAOB, Spotlight: Staff Update on Outreach Activities Related to the Integration of Generative AI in Audits and Financial Reporting (July 2024): “human involvement remains essential for auditors and preparers and is needed to review the output”
• Internal Revenue Service, Publication 4557: Safeguarding Taxpayer Data: “definition includes professional tax preparers”
• Internal Revenue Service, Publication 4557: Safeguarding Taxpayer Data: “The Safeguards Rule requires companies to develop a written information security plan that describes their program to protect customer information”