AI for Accounting Firms and CPAs: Client Work Without Risk

Sort tasks into three groups

Low risk: general client reminders, internal checklists, rewriting a confusing paragraph, training notes. Still reviewed, but no client data needed.
Review required: notice summaries, extracting figures from a W-2 or 1099, workpaper explanations, research support. Only in approved systems.
Never AI's call: signing engagement letters or returns, filing, submitting forms, authorizing payments, final client advice or approving entries.

Treat AI like a fast junior assistant that is sometimes wrong. The firm still owns the work.

Data security comes first

A staff member pasting client documents into a public chatbot can put confidential tax data out of the firm's control. The IRS's Publication 4557, Safeguarding Taxpayer Data, explains that federal rules require tax preparers to have a written data security plan, and Publication 5708 shows how to write one.

Add AI to that written information security plan:

Which AI tools are approved, and under what vendor terms?
Can names, SSNs, EINs, bank details, payroll records or tax forms be entered?
Does the tool train on firm inputs?
Who reviews output before it reaches a client or a taxing authority?
Where are AI drafts and summaries saved in the client file?
What happens if someone uses an unapproved tool by mistake?

If a task works with anonymized facts, use anonymized facts. Our list of what never to paste into ChatGPT is a good staff handout.

Document intake and organizer follow-up

Chasing W-2s, K-1s, 1099s, mortgage statements and signed organizers eats hours. In an approved tool, AI can compare the organizer checklist with what arrived and draft a missing-items email grouped by income, deductions, business records and signatures. The preparer reviews the list before it goes out, especially when a missing item affects filing status, credits or basis.

Extraction is not verification

AI can pull payers, dates and amounts from documents, but scans blur, forms get corrected and brokerage totals span pages.

Extract into a table with source page, label, amount and reviewer notes.
Tie every important figure back to the original document.
Mark unclear or inconsistent figures for follow-up; never guess.
Keep the source file and record who reviewed it.

Client emails and engagement letters

AI is good at plain-English explanations of extensions, estimated payments and next steps after a notice. Engagement letters are different: start from firm-approved templates and have the owner or partner review them. AI must not invent scope limits, fee terms or privacy promises. Consider telling clients plainly that the firm uses secure AI-assisted tools for drafting and organizing, while people remain responsible for judgment and deliverables.

IRS and state notices

Notices are deadline-driven. Have AI produce a structured brief for staff, then check every line against the notice itself:

Agency, notice number, taxpayer and tax period.
The issue and the amounts shown.
Response deadline and appeal rights.
Documents needed from the client.
Proposed response and items for the reviewer to confirm.

Research: AI points, official sources decide

AI can frame the question and list issues and authorities to check, and it can be confidently wrong. Practitioners' due diligence duties, such as those in Treasury Circular 230, still apply.

State the question without unnecessary personal details.
Ask AI for issues, search terms and sources to check.
Read the Code, regulations, IRS guidance or state source yourself.
Confirm dates, thresholds, definitions and exceptions.
Save citations in the workpaper.
Have the responsible CPA, EA or reviewer approve the answer.

Our guide to AI hallucinations explains why invented citations happen.

Workpapers and practice management

After the work is done, AI can turn reviewer notes into a clean memo listing documents reviewed, key figures and their sources, open questions and the conclusion. It can also draft SOPs and checklists for 1040 intake, monthly close, payroll quarter-end and extension follow-up. Completion dates, fees, new engagements and scope changes still come from people.

A rollout plan for a small firm

Name one person who owns AI use.
Update the written security plan with approved tools and banned uses.
Start with two workflows, such as missing-document emails and workpaper summaries.
Write approved prompts and templates so staff do not improvise with client data.
Train the team: confidentiality, review and "AI never signs or files".
Review output for a set period, then add notices, research or extraction.

Where ASL fits

Our AI training and AI consulting help firms set the rules before busy season, and our custom software development team builds intake and document tools that keep client data inside the firm's own systems. Contact us to talk it through.

Frequently asked questions

How can a small accounting firm use AI safely?

For document intake, missing-item emails, data extraction with verification, notice summaries, research support and workpaper drafts, in approved tools with a person reviewing every output.

Can tax preparers put client data into ChatGPT?

Not into public or personal accounts. Use only firm-approved tools covered by your written information security plan, and anonymize facts when you can.

What is IRS Publication 4557?

The IRS guide Safeguarding Taxpayer Data. It explains the security steps expected of tax professionals, including a written data security plan.

Can AI do tax research?

It can suggest issues and sources to check, but answers must be verified against the Code, regulations and official IRS or state guidance by a qualified preparer.

Should we tell clients we use AI?

Being plain about it builds trust: explain that AI-assisted tools help with drafting and organizing, while people review and sign off on all work.

Related guides

Explore the complete product

Read features, current access and limitations

AI for Small Accounting Firms and CPAs: Client Work Without the Risk