Best Claude prompts for accountants in 2026

Twelve Claude prompts working accountants actually use — covering worker classification, reconciliation hunts, tax notices, depreciation comparisons, R&D credit triage, audit prep, KPI commentary, fixed-asset memos, sales-tax nexus, and AR escalations. Each is structured to force stepwise reasoning, not guessing. Strip client identifiers before pasting, treat every output as a first draft, and route material work past a licensed CPA. Cited to AICPA, IRS, FASB, and Anthropic documentation.

→ **Generate a custom accounting prompt with our builder** (free, no signup)

---

Marketing prompts reward creativity. Accounting prompts reward the opposite — explicit rule citation, stepwise reasoning, and a clear refusal to invent facts. According to the AICPA's 2024 guidance on responsible AI use in accounting practice, professional skepticism still owns the final answer. Claude is a drafting partner, not the signing CPA.

A few ground rules sit underneath every prompt below:

- **Strip client identifiers before pasting.** Replace names, EINs, SSNs, and account numbers with placeholders like `[CLIENT_A]` or `[EIN]`. Anthropic's usage policy and trust documentation explicitly allows business use, but client confidentiality remains your obligation under AICPA Code of Professional Conduct §1.700. - **Force citations, not summaries.** Ask Claude to name the IRC section, ASC topic, or Rev. Proc. it is relying on. If it cannot name a source, treat the output as opinion, not authority. - **Use Claude for reasoning paths, humans for sign-off.** The Journal of Accountancy's 2025 AI adoption survey found firms that treat AI as a "second-set-of-eyes drafter" report 4x fewer review reworks than firms that treat it as an answer machine.

The prompts below assume Claude 3.5 Sonnet or newer. Older models will miss some of the reasoning structure.

---

**When to use:** A client asks whether a new hire can be paid as a contractor, or you're reviewing a misclassification risk during year-end.

``` You are reviewing a worker classification question under U.S. federal rules. Apply the IRS three-category common-law test (behavioral control, financial control, relationship of the parties) and the 20-factor analysis. Do not give a final answer. For each category, list the facts that point toward employee status and the facts that point toward contractor status, then state which direction the weight of evidence falls and why. Worker facts: - [Role and duties] - [Who sets schedule and where work is performed] - [Whether tools/equipment are provided] - [Method of payment and whether benefits are offered] - [Duration of relationship and exclusivity] - [Written agreement terms] End with: (1) the three biggest risk factors if classified as contractor, (2) what documentation would shift the analysis, and (3) whether a Form SS-8 determination request is advisable. ```

**Sample output (excerpt):** _"Behavioral control: company sets hours and reviews work weekly → points toward employee. Worker uses own laptop and software → mild contractor signal but weak under Rev. Rul. 87-41 weighting. Net direction: behavioral control leans employee. Largest contractor risk: schedule control plus exclusivity together create a strong reclassification case under recent DOL economic-realities guidance…"_

Anchor your prompt in the IRS's own framework. The IRS guidance on independent contractor vs employee is the actual source of authority — make Claude apply it, not paraphrase it.

---

**When to use:** A client questions an adjusting entry on their P&L, or you need a plain-English explainer for a board packet.

``` Explain the following journal entry to a small-business owner who does not have an accounting background. Do not use the words "debit" or "credit" in the explanation — translate to plain English ("we recorded an increase to..." or "this reduced..."). After the plain-English explanation, give a one-line summary of why this entry was needed and what would have been wrong if we had not recorded it. Entry: Dr. [Account] $[Amount] Cr. [Account] $[Amount] Context: [Why the entry is being made] ```

**Sample output:** _"This entry recognizes $4,200 of insurance coverage that we have already used up between January and March. We reduced the prepaid insurance asset on the balance sheet by $4,200 because that coverage is no longer 'prepaid' — it's been consumed — and we increased insurance expense on the P&L by the same amount. Why it matters: without this entry, Q1 profit would be overstated by $4,200 and the balance sheet would show an asset you no longer have."_

Claude is unusually good at this because it can switch register. Force the no-jargon constraint or it defaults to textbook language.

---

**When to use:** A bank rec, intercompany rec, or subledger-to-GL rec is off by an amount that doesn't tie to a single transaction.

``` I have a reconciliation difference of $[AMOUNT] between [SOURCE A] and [SOURCE B]. Below are the transactions from both sides for the period. Hunt for the discrepancy by: 1. Checking for sign reversals (a transaction posted opposite direction). 2. Checking for transposition errors (digits swapped) by listing any amounts where the digits could be a swap of the discrepancy. 3. Checking for duplicates (same vendor + amount within 7 days). 4. Checking for timing differences (same amount appearing in different periods on each side). 5. Checking for FX rounding patterns if multi-currency. For each candidate match, show the transactions involved and your reasoning. Do not declare a winner — present the candidates ranked by likelihood. Side A transactions: [paste] Side B transactions: [paste] ```

**Sample output:** _"Most likely candidate (transposition): Side A shows $1,840 to Vendor X on 03/14; Side B shows $1,480 to Vendor X on 03/14. Difference $360 matches your variance exactly. Second candidate (timing): Side A 03/29 deposit $4,200 not on Side B until 04/02…"_

Reconciliation hunting is exactly the kind of patterned drudge work Claude accelerates. Per the Journal of Accountancy's 2024 coverage on AI in audit and accounting workflows, pattern-matching tasks see the largest productivity lift.

---

**When to use:** An IRS or state notice arrives and you need to brief the client before billing review time.

``` A client received the attached tax notice. Write a plain-English summary in this exact structure: 1. What the IRS/state is saying (one sentence, no jargon). 2. Why they're saying it (likely cause — pick the most probable from the notice text). 3. What we believe is actually true (placeholder for our position). 4. What happens if we do nothing. 5. The deadline and what we need from the client to respond. 6. Our recommended next action. Constraints: - No tax code citations in the client-facing summary. - Tone: calm, factual, not alarming. - Maximum 250 words. Notice text: [paste — REMOVE CLIENT NAME, SSN, EIN FIRST] ```

The IRS notice explanation library is your fact-check source — verify Claude's interpretation against the actual notice number before sending. Notices like CP2000 and LT11 have very specific procedural meanings that Claude will sometimes blur.

---

**When to use:** A client is acquiring a fixed asset and wants to understand the cash-flow consequence of MACRS vs straight-line vs Section 179 vs bonus depreciation.

``` Compare depreciation treatment options for a [ASSET TYPE] purchased for $[COST] placed in service [DATE], with a useful life of [YEARS] and a business-use percentage of [%]. The client is a [ENTITY TYPE] in a combined federal+state marginal rate of [%]. Compare across these options: - MACRS (state the property class and recovery period under IRC §168) - Section 179 expensing (note 2026 limits and phase-out) - Bonus depreciation (note the current applicable percentage) - Straight-line book depreciation For each option, show: - Year 1 deduction - Year 1 tax savings - 5-year cumulative deduction - Book vs tax difference and the deferred-tax implication - One reason to choose this option - One reason to avoid this option End with a decision matrix scored 1-5 across: cash-flow front-loading, audit-defensibility, book-tax alignment, future-year flexibility. ```

Authority sources: IRC §168 MACRS rules (Publication 946) and current-year inflation-adjusted limits from Rev. Proc. updates. Always re-confirm Section 179 and bonus-depreciation percentages against the current year's IRS announcement — those numbers change.

---

**When to use:** Monthly close is done, the indirect-method cash flow is built, and you need a one-page narrative for the management report.

``` Write a cash-flow narrative for the management report based on the indirect-method statement of cash flows below. Structure: Paragraph 1 — Headline (two sentences): net cash position change and the single biggest driver. Paragraph 2 — Operating activities: what worked, what didn't, what the working-capital movements actually mean operationally (not just "AR increased" — interpret it). Paragraph 3 — Investing activities: capex pace vs budget, any divestitures. Paragraph 4 — Financing activities: debt movement, distributions, any covenant-relevant items. Paragraph 5 — Forward look: what to watch next month, and one number the operator should track weekly. Tone: confident, factual, no hedge words. Audience: COO and CFO, not the board. Statement of cash flows: [paste] Budget for context: [paste] ```

The narrative quality jumps when you pre-stage the budget alongside the actuals — Claude can then call out variances without you having to label them.

---

**When to use:** Year-end audit is approaching and you need a tailored PBC (prepared-by-client) list for a specific engagement.

``` Generate an audit prep checklist for a [INDUSTRY] [ENTITY TYPE] with $[REVENUE] annual revenue, [HEADCOUNT] employees, and the following material balance-sheet areas: [list account areas with balances]. For each material area, list: - Documents the auditor will request (be specific — not "bank confirmation" but "signed bank confirmation for each account >$[threshold]") - Working papers we should prepare in advance - The most common audit adjustment in this area for this industry - The one question the audit senior will ask first Reference the relevant AU-C section where applicable. Group output by audit area, ordered by likely audit hours. ```

Cross-check Claude's AU-C references against the AICPA's clarified auditing standards index. It occasionally cites legacy SAS numbers; the AU-C re-codification is authoritative.

---

**When to use:** A client mentions software development, product engineering, or process improvement spend and you want a fast triage of whether IRC §41 credit work is worth scoping.

``` Triage whether this activity likely qualifies for the federal R&D credit under IRC §41 by applying the four-part test: 1. Permitted purpose (new or improved business component — function, performance, reliability, or quality). 2. Technological in nature (relies on hard sciences). 3. Elimination of uncertainty (technical uncertainty, not commercial). 4. Process of experimentation (systematic evaluation of alternatives). For each prong, rate the activity Strong / Moderate / Weak / Not Met, and explain in one sentence. If any prong is Not Met, the activity does not qualify — say so explicitly. End with: (a) qualified research expense categories that would likely apply, (b) documentation gaps the client needs to close, (c) whether the §174 capitalization treatment changes the value calculus, and (d) whether this is worth a full study or a §280C statement. Activity description: [client description] ```

The four-part test is from IRC §41 and Treasury Regulation §1.41-4. Post-TCJA §174 capitalization rules materially changed credit economics — make Claude flag that interaction or you'll over-promise on cash benefit.

---

**When to use:** Monthly KPI dashboard is built and the CFO needs interpretive commentary, not just numbers.

``` Write KPI commentary for the metrics below. For each metric: - State the value and the variance vs prior month and vs budget. - Interpret what the variance means operationally — not "revenue up 8%" but "revenue up 8% driven by [the underlying business reason if visible from the data, else: 'driver not yet identified — recommend ops review']." - Flag any metric where the trend is masking an underlying issue (e.g., revenue up but AR aging deteriorating → cash impact coming). - Recommend one action per metric where action is warranted. Constraint: do not invent drivers you cannot see in the data. If you cannot identify a driver, say so and recommend the investigation. KPIs: [paste with current/prior/budget] Supporting data: [paste any sub-detail] ```

The "do not invent drivers" constraint is the most important line in the prompt. Without it, Claude will confidently attribute variances to plausible-sounding causes that may not be true.

---

**When to use:** A client needs a written capitalization policy for internal control purposes or to support an audit.

``` Draft a fixed-asset capitalization policy memo for a [ENTITY TYPE] with approximately $[ANNUAL CAPEX] in annual capital expenditures. Include sections: 1. Purpose and scope 2. Capitalization threshold (recommend a dollar threshold appropriate to entity size — typical SMB ranges $1,000-$5,000) 3. Asset classes and useful lives (table format — align book lives with reasonable economic life, not tax life) 4. Componentization rules (when an asset is broken into components) 5. Repair vs improvement decision rules (with three examples each) 6. Disposal and impairment triggers 7. Approval authority matrix 8. Annual policy review trigger Reference relevant ASC 360 guidance for impairment and ASC 842 for any ROU asset interactions. Output in memo format ready for the client's controller to adapt. ```

Reference: FASB ASC 360 (Property, Plant, and Equipment) and ASC 842 (Leases) for ROU asset interactions. Have Claude name the ASC topics so the reviewer can confirm.

---

**When to use:** A client expanded into new states and you need to scope nexus exposure before recommending a study.

``` Help me frame the sales-tax nexus questions for a [BUSINESS TYPE] that sells [PRODUCT/SERVICE TYPE] and recently expanded into [STATES]. Output: A. For each state listed, name the economic-nexus threshold (sales dollar amount and/or transaction count) post-Wayfair, and the lookback period. If a threshold has changed in the last 24 months, flag it. B. List the data the client needs to pull from their order/billing system to answer the nexus question definitively (be specific — fields, date range, filters). C. Identify which states in the list have marketplace facilitator rules that may shift the collection obligation away from the seller. D. Identify which states tax the specific product/service type differently (e.g., SaaS treated as taxable tangible property in some states, services exempt in others). E. Output the three highest-risk states for this client and why. End with the scope of a nexus study you'd recommend, including which states to include and which to defer. ```

Verify thresholds against state DOR websites — nexus rules shift annually. The Wayfair decision background and current state thresholds tracked by AICPA's state regulation resources is a useful cross-reference.

---

**When to use:** AR is past 60 days and you need to send a graduated collection email that doesn't damage the relationship.

``` Draft an AR collection email to a B2B client that is [DAYS] days past due on invoice #[NUMBER] for $[AMOUNT]. This is the [FIRST/SECOND/THIRD] escalation. Calibrate tone to escalation level: - First: warm, assumes oversight, easy path to payment. - Second: firmer, references prior outreach, asks for a specific commitment date. - Third: direct, references account hold or collections referral, but keeps door open for one final dialogue. Constraints: - Maximum 120 words for first, 150 for second, 180 for third. - Include a single, specific call to action. - Offer one practical payment option (e.g., split payment, ACH link). - No threats. No legal language unless escalation level requires it. Client context: [paste — REMOVE CLIENT NAME before sharing with Claude if confidentiality is a concern; use a placeholder] ```

The graduated-tone constraint is what makes this prompt earn its keep. Without it Claude defaults to a single register that sounds either passive or aggressive depending on its read of the situation.

---

---

The AICPA Code of Professional Conduct §1.700 on confidential client information and the IRS §7216 regulations on disclosure or use of tax return information both apply when you paste client data into any LLM, including Claude.

Practical rules I follow and recommend to firms I consult with:

1. **Strip identifiers before pasting.** Replace names, EINs, SSNs, account numbers, and physical addresses with placeholders. The reasoning quality of Claude's output does not depend on these identifiers. 2. **Use Claude's "no training" setting for business accounts.** Anthropic's data usage documentation for commercial customers confirms enterprise and team plan inputs are not used for model training by default. Verify your team plan settings. 3. **Never paste full tax returns, full GLs, or full client lists.** Paste the section relevant to the question. The output quality is often better with focused input anyway. 4. **Document AI use in your engagement workflow.** Audit standards under AU-C 315 risk assessment increasingly contemplate how AI tooling factors into the control environment. A short engagement-level memo covering "we use Claude for X, Y, Z, with these controls" is a defensible position. 5. **For §7216 tax engagements, get the written consent form on file** before using any AI for tax return preparation or advisory work that processes return information. Sample language is in IRS Rev. Proc. 2013-14.

A simple rule: if you would not paste it into a public forum, prep it before pasting it into Claude.

→ **See our full prompt library for accounting and finance workflows**

---

Each prompt above went through the iteration protocol described in our 7-point prompt grading rubric guide. Specificity and constraint-compliance were the dimensions that lifted accounting prompts most — generic prompts produce textbook outputs; constrained prompts produce client-ready drafts.

The structure draws on Anthropic's published prompt engineering guidance — particularly the patterns around role assignment, structured outputs, and explicit reasoning steps. The accounting authority sources behind the reasoning (IRC sections, ASC topics, AU-C standards, AICPA guidance, IRS publications) were chosen because Claude will sometimes cite legacy numbering when it cannot recall current — naming the authoritative source forces it to either cite correctly or admit uncertainty.

→ **Try our prompt builder to adapt these for your practice**

---

### Are these prompts safe to use with client data?

Only after you strip identifiers and confirm your Claude account's data-handling settings. The prompts themselves are safe; the client data inputs require the confidentiality controls described above. AICPA §1.700 and IRS §7216 obligations do not pause for AI tooling.

### Will Claude replace a CPA?

No. Claude drafts. CPAs decide, sign, and bear professional responsibility. The 2025 Journal of Accountancy AI adoption coverage frames this well — firms treating AI as a drafter outperform firms treating it as an oracle.

### Which Claude model works best for accounting work?

Claude 3.5 Sonnet or newer. Older models miss the structured reasoning steps these prompts ask for, particularly in worker classification (Prompt 1) and R&D credit triage (Prompt 8). For high-stakes outputs, run the prompt twice and compare — divergence is a useful signal that the question is genuinely ambiguous.

### Can I use these for state-specific work?

Yes, with one adjustment — name the state in the prompt and ask Claude to cite the state authority alongside the federal authority. Sales-tax nexus (Prompt 11) is the obvious example; depreciation timing (Prompt 5) is the less obvious one because state conformity to federal bonus depreciation varies.

### How do I know if Claude's output is wrong?

Cite-check it. If Claude names IRC §168, look up §168 and verify the rule it described matches the actual code. If it cites ASC 842, open ASC 842 and confirm. Hallucinated citations are the single biggest risk pattern with LLMs in accounting work, and the cheapest one to catch.

### Do I need to disclose AI use to clients?

Practice varies. The conservative position — and the one I recommend — is yes, in your engagement letter, with language describing how AI is used and what controls apply. Several state boards are moving toward explicit guidance; track AICPA professional ethics updates for changes in your jurisdiction.

### What about IRS Circular 230 obligations?

Circular 230's competence and due-diligence standards apply to AI-assisted work the same as any other. The AI does not relieve the practitioner of §10.22 (due diligence) or §10.35 (competence) obligations. Treat Claude outputs the way you would treat a junior staff member's draft — useful, fast, requires review.

---

- AICPA — Code of Professional Conduct - AICPA — Generative AI guidance for assurance - IRS — Worker classification guidance - IRS — Section 7216 FAQs on tax-information disclosure - IRS — Research credit (§41) overview - IRS — Publication 946 (How to depreciate property) - FASB Accounting Standards Codification - Anthropic — Prompt engineering overview - Anthropic — Commercial terms and data handling - Journal of Accountancy — ongoing AI in accounting coverage

→ **Start with our prompt builder — free, no signup**