Published on

ChatGPT Prompts for Accountants: 18 Prompts and What to Avoid

Authors
  • avatar
    Name
    PromptShelf Editorial
    Twitter

This article is for informational purposes only and does not constitute professional accounting, tax, audit, or financial advice. Always consult a qualified CPA, CMA, or licensed tax professional for your specific situation. Nothing here should be relied on for filings, audits, or client deliverables without independent verification.

Most accounting work is not the math. The math is fine. The work is everything around the math: explaining variances to a sales VP who skipped finance class, drafting the polite-but-firm email asking a client for missing receipts, writing the SOP a new hire will read twice and then ignore, summarizing what changed in the latest FASB update for the partner meeting on Monday.

This is the part of the job ChatGPT is good for. Not the math, not the audit, not the advice. The writing, the structuring, the first-draft work that eats your Tuesday afternoon.

This guide has 18 ChatGPT prompts for accountants, split across the six places AI actually helps in finance work, plus a frank section on the work where you should keep AI out completely. Every prompt is copy-paste ready. One has been tested on free ChatGPT and the response is reproduced verbatim below.

What ChatGPT can and can't do for finance work

You are not handing your trial balance to a chatbot. You are using a writing tool that happens to know what a deferred revenue waterfall is. The mental model that works:

ChatGPT is good at the language layer of accounting. It can structure a memo, soften a client email, translate jargon for a non-finance audience, draft a process doc, summarize regulatory text, and rough out a workpaper template. It can explain a concept like a textbook would.

ChatGPT is bad at the precision layer. It will hallucinate citations to ASC sections, invent IRS publication numbers, get effective dates wrong on tax law, and confidently tell you a transaction is non-taxable when it is. It cannot do reliable math on long lists of numbers. It cannot give you advice. It does not know your client's facts.

The split is simple. Use it for words. Don't use it for facts you'll sign your name to.

If you want a longer treatment of how to structure prompts so they actually return useful output, our guide to how to write ChatGPT prompts that work breaks down the four-part structure every prompt below uses.

Category 1: Client emails and follow-ups

These are the highest-volume, lowest-risk use case. You write dozens of these emails a week. ChatGPT drafts a clean first version, you spend two minutes editing for tone and accuracy.

Prompt 1. Polite request for missing client documents

Prompt: "You are a senior tax accountant writing a follow-up email to a small business owner who has not sent the documents you requested two weeks ago. Tone: warm, professional, slightly firm but not annoyed. The deadline pressure is real because their extension expires in 21 days. List the missing items as bullets so they can scan it. End with a clear single ask: reply with a date you can have these to me. Output: a 130-word email with a subject line. Missing items: [list, e.g., '2025 1099-NEC for contractor X, year-end bank statements from Chase, mileage log, home office utility bills']."

Prompt 2. Client onboarding welcome

Prompt: "You are an account manager at a mid-sized accounting firm. Write a welcome email for a new bookkeeping client. Cover: confirming the engagement scope, the portal login process, who their main point of contact is, what we'll need from them in the first week, and a friendly close. Output: 200 words, conversational but professional, no marketing language. Client info: [paste, e.g., 'Pacific Bay Café, $1.2M annual revenue, monthly bookkeeping, QBO file already shared']."

Prompt 3. Late-payment reminder for a long-term client

Prompt: "You are a partner at a small accounting firm sending a third late-payment reminder to a 5-year retainer client. Tone: respectful of the relationship, but the invoice is now 60 days past due and we need it resolved this week. Avoid hedging. State the amount, the original date, suggest two paths (pay in full or set up a 3-month plan), and end with a deadline for their reply. Output: 140 words, includes subject line."

Prompt 4. Apology email after a missed deadline

Prompt: "You are an accountant writing to a client to apologize for missing the agreed-upon deadline for their quarterly financial package. Brief, sincere, not over-apologetic. Acknowledge the impact, state the corrected delivery date, explain the safeguard you're putting in place so it doesn't recur. No excuses. Output: 100 words, includes subject line. Reason for delay: [paste, e.g., 'two team members out sick the same week we received your bank reconciliation files']."

Category 2: Internal memos and explainer documents

This is where the time sink lives. Variance explanations, board memo summaries, partner-meeting prep notes, fluctuation analysis writeups for the file. ChatGPT is unusually good at translating numbers into plain language.

Prompt 5. Variance explanation for non-finance executives

Prompt: "You are a senior staff accountant writing a one-page month-end variance memo for a non-finance executive audience. Summarize the following variances in plain English, lead with the bottom line, group similar drivers together, and end with one sentence on what we expect next month. Variances: [paste, e.g., 'Revenue 1.4Mactualvs1.4M actual vs 1.5M budget (-7%, due to 2 customer renewals slipping into next month); Gross margin 38% actual vs 41% budget (-3pts, due to higher freight costs on rush orders); Operating expenses 620kactualvs620k actual vs 580k budget (+7%, due to a one-time legal settlement of $45k)']. Output: a 250-word memo in three short paragraphs, no jargon, no bullet points."

What this prompt actually returns (tested 2026-05-05)

We ran Prompt 5 on free ChatGPT with the variances baked in ('Revenue $1.4M actual vs $1.5M budget (-7%, due to 2 customer renewals slipping into next month); Gross margin 38% actual vs 41% budget (-3pts, due to higher freight costs on rush orders); Operating expenses $620k actual vs $580k budget (+7%, due to a one-time legal settlement of $45k)'). Reproduced verbatim:

We finished the month slightly below plan, mainly due to timing and one-off costs rather than a change in underlying demand. Revenue came in at 1.4Mversusa1.4M versus a 1.5M budget, about 7% lower. This shortfall was driven by two customer renewals that were expected this month but are now set to close next month. On the cost side, expenses were higher than planned, which also put pressure on overall results.

Looking at profitability, gross margin was 38% compared to a 41% target. The gap is primarily tied to higher freight costs, as we fulfilled several rush orders that required expedited shipping. These orders supported customer needs and revenue, but at a higher delivery cost than usual. Operating expenses totaled 620kversusa620k versus a 580k budget, about 7% over plan. This increase is largely explained by a one-time legal settlement of $45k, which was not included in the original budget and is not expected to repeat.

Overall, the main drivers this month were timing-related revenue delays, higher shipping costs tied to urgent orders, and a one-time legal expense. The business itself remains stable, and none of these items point to a structural issue. Next month, we expect revenue to rebound as the delayed renewals close, margins to normalize as freight costs return to typical levels, and operating expenses to come back in line without the one-time legal cost.

Editor's notes. ChatGPT got the structure right: bottom-line lead, three short paragraphs, a single forward-looking sentence at the end, no bullet points, no jargon. The response runs about 270 words, slightly long against the 250 ask but acceptable. Three things worth fixing before sending. First, the memo doesn't quantify the absolute dollar miss on operating income, which is the line a non-finance CEO actually wants. Second, paragraph three partly repeats paragraphs one and two and could be cut to two sentences without losing anything. Third, the "remains stable" and "no structural issue" framing is an editorial call you should verify against your real picture before forwarding. The model also opted for "we finished the month slightly below plan" rather than a sharper number-led opener, so if your house style leads with the number, you'll want to rewrite the first sentence.

Prompt 6. Plain-English explanation of a complex accounting concept

Prompt: "You are a CPA explaining a technical accounting concept to a smart, non-accountant CEO. The audience knows business but not GAAP. Use everyday analogies. Avoid jargon. If you must use a term, define it once in parentheses. End with one sentence on why this matters for their business specifically. Concept: [paste, e.g., 'why we recognize revenue ratably over the contract term instead of all at signing for our 12-month SaaS subscriptions']. Output: 200 words, conversational tone."

Prompt 7. Audit committee meeting summary

Prompt: "You are an audit manager preparing a one-page summary of an audit committee meeting for partners who weren't able to attend. Pull out the three most important decisions or discussion points. State each one in a single sentence, followed by 2-3 sentences of context. End with a list of action items with owners and due dates. Output: 350 words. Meeting notes: [paste raw notes]."

Prompt 8. Fluctuation analysis writeup

Prompt: "You are a senior auditor documenting a fluctuation analysis for the workpapers. The client's professional services revenue increased 22% year-over-year. Explain the increase in the standard format: identify the change, identify the drivers (volume, price, mix), assess whether the change is consistent with the auditor's expectations, conclude on whether further procedures are needed. Use the language and structure typical of a Big 4 fluctuation memo. Output: 300 words. Client context: [paste, e.g., 'Mid-sized B2B consulting firm; added two enterprise clients in Q3; raised standard rates 8% in January']."

Category 3: Process and workflow documentation

SOPs that nobody reads are still SOPs that have to exist. ChatGPT is good at writing them in the first place, and even better at converting messy notes into a clean process doc.

Prompt 9. Standard operating procedure from rough notes

Prompt: "You are an accounting operations manager turning a senior bookkeeper's rough notes into a polished SOP that a new hire can follow without supervision. Use a numbered step format. Include: prerequisites at the top, the step list, expected outputs at each step, and 'common mistakes' callouts where the senior bookkeeper flagged them. Be specific. Avoid generic phrases like 'be careful'. Output: a clean SOP, around 600 words. Rough notes: [paste]."

Prompt 10. Month-end close checklist

Prompt: "You are a controller designing a month-end close checklist for a small accounting team (3 people, $20M-revenue services company on QuickBooks Online). Group tasks into: pre-close, day 1-3 close, day 4-5 close, post-close review. For each task, list the owner role (Bookkeeper, Senior, Controller), expected duration, and dependency on other tasks. Don't include audit-specific tasks. Output: a structured checklist, ~30 line items."

Prompt 11. Engagement letter scope section

Prompt: "You are a tax partner drafting the scope-of-services section of an engagement letter for a new individual tax client. Be specific about what's included (federal and one state return, K-1 income from two pass-through entities, foreign asset reporting if applicable) and what's explicitly excluded (audit support, IRS correspondence after filing, amended returns, FBAR is not part of scope unless added in writing). Output: 250 words, formal legal style. Client situation: [paste, e.g., 'High-earning individual, ownership stake in two LLCs, brokerage accounts in Switzerland and Singapore']."

Category 4: Research starters

Use ChatGPT to get oriented, never to get the final answer. The model gets dates, citations, and effective-date language wrong often enough that you cannot rely on it. But it's a fine first scan to figure out what to research properly.

Prompt 12. Topic primer for unfamiliar regulation

Prompt: "You are a tax research assistant. I need to get oriented on a topic I haven't worked with in years. Give me a primer that covers: what the rule does, who it applies to, what the most common practical questions are, what publications or code sections I should pull up to verify, and what's likely to have changed in the last two years that I should double-check. Be explicit that you are likely to have outdated or incorrect specifics, and flag the items most worth verifying. Topic: [paste, e.g., 'IRC Section 174 R&D capitalization rules for software companies']."

Prompt 13. Comparison of accounting treatments

Prompt: "You are a senior accountant explaining the difference between two accounting treatments to a junior staff accountant. Cover: what each method does, when GAAP requires or permits each, the typical pros and cons, and a practical example showing how the financial statements differ under each. Be honest if the difference is small or rarely matters. Output: 400 words, plain language. Treatments: [paste, e.g., 'completed-contract method vs percentage-of-completion for long-term construction contracts']."

Prompt 14. List of red-flag transactions for diligence

Prompt: "You are a forensic accountant helping a junior team member prepare for a buy-side financial diligence engagement on a small private company. List the top 10 red-flag transaction patterns to look for, organized by category (revenue recognition, cash management, related-party, expense classification, working capital). For each, describe what the pattern looks like in a general ledger or QuickBooks export, and the question to ask management about it. Output: 600 words. Industry: [paste]."

Category 5: Talent and team management

Accounting is a people problem dressed up as a numbers problem. The same prompt patterns we wrote about in our HR and recruiting prompts post apply here, with finance-specific framing.

Prompt 15. Performance review for a staff accountant

Prompt: "You are a controller writing a mid-year performance review for a staff accountant. Use the standard format: strengths, areas for development, specific goals for the second half of the year, overall rating with rationale. Be specific. Avoid HR-cliches like 'great team player'. If feedback is critical, frame it as actionable. Output: 400 words. Staff member's recent work: [paste 4-5 bullet points of what they did well and what struggled]."

Prompt 16. Job description for a senior tax associate

Prompt: "You are a recruiting manager at a small CPA firm writing a job description for a Senior Tax Associate role (3-5 years experience, partnership and S-corp focus). Sections: about the firm, what you'll do (specific tasks, not buzzwords), what we're looking for (specific skills and credentials), what we offer (compensation range, learning, partner-track if applicable). Avoid corporate cliches like 'fast-paced environment' and 'rockstar'. Output: 500 words. Firm details: [paste]."

Category 6: Career and business development

Prompt 17. LinkedIn post about a recent client win or industry change

Prompt: "You are a CPA writing a LinkedIn post for your professional network. Topic: a recent change in tax or accounting standards that small business owners should know about. Tone: knowledgeable but accessible, not preachy. Open with a real-world hook (a question your clients have actually asked you, or a misconception you keep correcting). Make one specific, useful point. End with a soft call to action. Output: 200 words. Topic: [paste]."

Prompt 18. Email pitch for a referral partnership

Prompt: "You are a tax partner reaching out to a financial advisor at a wealth management firm to propose a mutual referral relationship. The two of you met briefly at an industry event last month. Tone: professional, warm, not pushy. Cover: the connection, why our two practices fit (specific, not generic), proposed first step (a 30-minute call or coffee), and a clear ask. Output: 180 words, includes subject line."

When NOT to use ChatGPT for finance work

This is the part most "ChatGPT for accountants" guides skip. Here's where you should keep AI completely out of the workflow, even with a paid plan.

Anything you'll sign or attest to. If your name goes on it, the work needs to be yours. That covers tax returns, financial statements, audit reports, attest engagements, compilation reports. ChatGPT can help you draft the email that goes around the deliverable. The deliverable itself should not pass through it.

Math on long lists of numbers. This is the most common mistake. People paste a 200-row trial balance and ask ChatGPT to total a column or check the math. It will produce a confident answer that is wrong roughly a third of the time. Tools that do math reliably (Excel, your GL system) exist for a reason. Use them.

Citations to specific code sections, ASC paragraphs, IRS publications, or court cases. The model will fabricate. It will give you ASC 606-10-25-31 when the right citation is 606-10-25-23, or invent an IRS publication number that does not exist. Always pull the source yourself before relying on a citation in a memo or in correspondence with a client.

Effective dates and version-current rules. Tax law changes. GAAP gets amended. ChatGPT's training data is fixed at a point in time, and the model does not know reliably whether what it's telling you is current. Anything date-sensitive needs to be verified against the actual current source.

Client-specific advice. Two clients in the same industry have different facts. ChatGPT does not know the facts. It can tell you the general framework for evaluating, say, whether a transaction qualifies for Section 1031 treatment. It cannot tell you whether your client's specific transaction qualifies. The distinction is not subtle and your professional judgment is what the client is paying for.

Anything containing client confidential information on the free tier. Free ChatGPT trains on conversations by default. Even with the training opt-out toggled, you should treat the input as if it could leave the platform. Do not paste a client's general ledger, their bank statements, their employees' W-2s, or any PII. If you want to use AI on actual client data, you need a paid enterprise plan with a signed data processing agreement, your firm's IT signoff, and ideally a deployed-on-your-tenant solution. That is a real conversation to have with your firm leadership, not something to hack together on a Tuesday afternoon.

The honest version. ChatGPT saves you maybe 4-6 hours a week on the writing layer of accounting work. That is a meaningful productivity gain. The temptation, once you see how good it is at the writing, is to push it into the precision layer too. That is where firms get into trouble. Hold the line.

Best practices for ChatGPT prompts for accountants

Three habits separate the people getting real value from ChatGPT from the people getting a confident-sounding mess.

Give the model a role. "You are a senior tax accountant" produces meaningfully different output than no role assignment. The model's writing register shifts. Vocabulary becomes more accurate. The output reads like it came from an accountant.

Provide the actual data. The biggest difference between a useful response and a generic one is whether you pasted in the real variances, the real client situation, the real rough notes. The bracketed [paste, e.g., ...] examples in every prompt above are not optional. They are the prompt.

State the output you want. Word count, format, paragraph structure, what to leave out. Most "ChatGPT didn't give me what I wanted" complaints are actually "I didn't tell ChatGPT what I wanted". The output spec is the cheapest way to get a usable first draft.

For a deeper look at the same pattern applied across other professions, our small business owners' guide to ChatGPT prompts walks through the same role-task-context-output pattern with examples from operations and money admin.

FAQ

Is it OK to use ChatGPT for tax research?

For getting oriented, yes. As a starting point to figure out what to read and which code sections to pull, ChatGPT is a fast way in. As the final answer that you cite to a client or in a memo, no. The model fabricates citations and gets effective dates wrong often enough that you cannot rely on it without independent verification against the actual primary source.

Can ChatGPT prepare a tax return?

No. Not because it lacks the math (which it also lacks), but because tax preparation requires applying judgment to a specific client's specific facts, and getting the answer wrong has real legal and financial consequences for the client and for you. ChatGPT can help you draft the cover letter that goes with the return. It cannot prepare the return itself.

What about CPA exam prep?

This is one of the genuinely strong use cases. ChatGPT is a patient tutor who can re-explain concepts in different ways, generate practice questions, walk you through how it would approach a simulation, and quiz you on weak areas. Treat it as a study buddy, not as the answer key. Always verify against the actual study materials and AICPA blueprints.

Will ChatGPT replace accountants?

The writing layer of the job will get faster. The judgment layer, the relationship layer, and the attest function are protected by professional standards and by genuine difficulty. Junior staff who learn to use AI well will produce more output per hour and become useful faster. Senior staff who refuse to engage with it will fall behind on speed but not on judgment.

Is paid ChatGPT worth it for accountants?

If you use it daily for the writing-layer work in this guide, the $20/month plan pays for itself in roughly the first hour of the first day. The bigger benefit of the paid plan is not the model quality, it's the data handling. Plus and Team plans give you better controls over training opt-out, and Team adds an admin console. For client confidential work, you want at least Team and ideally an enterprise plan with a signed data processing agreement.

What to do next

Pick the one prompt above that matches the task on your desk right now. Paste it into ChatGPT with your real bracketed inputs. Edit the output. Send it. Then come back and pick a second one. The point is not to have 18 prompts ready. The point is to use 3 of them this week and keep the ones that earn their place.

If you want to write your own prompts in this style, our PRSO framework guide breaks down the four-part structure every prompt above uses.

Related: more prompts by profession