When Clients Use AI, What Changes for the Accountant?

When a client uses AI, the accountant needs to know where it was used, what information went into it, which records support the answer, and who checked it. If that answer will affect the books, a filing, a forecast, or a major decision, it needs a closer check.

AI use is already common enough to appear in client work. The U.S. Census Business Trends and Outlook Survey estimated that 23.2% of the businesses covered by the survey used AI in a business function during August 10–23, 2026. In a separate Federal Reserve survey, 46% of employer firms said their business or employees used AI.

The surveys asked different questions and used different methods, so the percentages are not directly comparable. Both still point to the same practical issue: accountants will see more work that AI helped create.

‍

Accountants already need to understand where a number came from. AI can make that trail harder to follow.

A client may bring an AI-generated cash forecast, expense classification, tax explanation, or summary of business performance. It may look finished even when the records are incomplete, the assumptions are weak, or the cited sources do not exist.

Start with one question:

What records, assumptions, and instructions produced this result?

A draft used to prepare meeting questions does not need the same review as a forecast used to approve hiring or an answer included in a tax filing.

AICPA’s 2026 guidance on technology output says accountants still need to decide whether a tool’s answer is fit for the job. For federal tax work, the IRS Office of Professional Responsibility tells practitioners to check the facts, citations, and calculations in AI-created work.

AI may write the answer, but the evidence is still the invoice, contract, ledger, report, calculation, or official rule.

The review is easier when the client and accountant can see the same source records.

‍

Sometimes clients will tell you they used AI. They may share a chatbot conversation or say that a tool prepared a forecast. Other times, AI may be built into their software or used by an employee without a company-wide process.

A 2026 Census study of business AI use found that employees sometimes used AI for work even when their company did not report using AI. Simply asking, “Does the business use AI?” may therefore miss part of the picture.

A more useful question for onboarding and regular client reviews is:

Has anyone used an AI tool to prepare, classify, summarize, forecast, or explain information for this work?

If the answer is yes, ask what happened next. Did the result stay as a rough draft, or did it affect the books, a filing, a report, a recommendation, or a business decision?

‍

Accuracy was the most common challenge reported by firms using AI in the Federal Reserve survey. An answer needs a closer review when it will affect the books, a filing, a forecast, or a business decision.

Five questions can guide that review:

1. Where was AI used? Name the tool and the task it performed.
2. What data went into it? Note any financial, tax, payroll, customer, or other private information.
3. What did it produce? Separate brainstorming and drafting from classifications, calculations, forecasts, or a proposed entry in the books.
4. What records support the result? Return to the original documents, reports, calculations, or official guidance.
5. Who checked it? Make sure someone is responsible before the result affects a filing, report, recommendation, or important decision.

The exact check depends on the work. For an expense classification, review the transaction and the firm’s accounting policy. For a forecast, check the starting cash balance, time period, expected receipts, committed payments, formulas, and assumptions. For a tax or accounting rule, open the official source instead of relying on an AI citation.

Imagine that a client brings an eight-week cash forecast from a general-purpose chatbot. They cannot say which prompt or receivables report was used, whether payroll and tax payments were included, or who checked the formulas.

Start again with the current cash balance, expected payment dates, bills, payroll, taxes, and other committed payments. Recalculate the forecast with clear assumptions, then explain what it can and cannot support.

‍

When clients use AI, private information can end up in places they did not expect. An owner might paste a profit-and-loss statement into an unapproved chatbot. An employee might upload invoices, payroll details, customer records, or tax documents without knowing who can access them.

AICPA & CIMA’s 2026 ethics guidance recommends understanding a tool’s limits, where it stores data, and how it protects private information.

A firm does not need to review every app a client uses. But it should set clear rules for handling client information in AI tools. Clients and staff should be able to answer these questions:

• Was the tool approved for this type of information?
• What information must never be uploaded?
• Who can access the output and conversation history?
• What record of the prompt, sources, assumptions, and review must be kept?

The same privacy and confidentiality rules the firm already follows should apply here.

‍

Take a common client question: “Sales were up this month, so why is cash still tight?”

With a general chatbot, the client may paste one report and get a confident answer. But that answer may miss late invoices, upcoming bills, refunds, or other records the client did not include. The accountant then has to work backward to find out what the AI saw and what it missed.

In Veltrix, the client can ask the cash question using the records already connected to their workspace. The accountant sees the same answer and the reports behind it.

Veltrix can read the connected records but cannot post, edit, or delete anything in the client’s tools. The accountant can open the reports behind the answer, check the numbers, and add the business context that software may not know.

The accountant can review the answer with the client before it is treated as final. There is no need to reconstruct it later.

‍

Before the next client review, the accountant can run the free Client Business Health Check. In about two minutes, it flags results that may need attention and shows the numbers behind them.

Suppose the Health Check shows that sales increased but cash did not. The accountant can open that result, check the numbers behind it, and see whether slow customer payments, higher costs, or another change may explain the gap.

The next client conversation can now start with something specific:

Sales improved this month, but cash did not move with them. I checked the numbers and customer payments appear to be taking longer. Has anything changed in how invoices are being followed up?

After checking the numbers, the accountant can decide whether the finding matters and what to ask the client next. That gives the client a specific issue to discuss instead of a general review of the month.

If the review uncovers work outside the current engagement, the accountant can show the client the supporting numbers before discussing a change in scope or fee.

Start with one existing client and one finding, then decide whether the process belongs in more reviews.

If you want to try the same process with more clients, explore Veltrix for accountants and the Partner Firm Program.