Claude AI for Tax Planning: A CPA's Complete Guide (2026)

How CPAs Are Using Claude for Tax Work

Most CPAs have tried asking ChatGPT or Claude a tax question and gotten a reasonable-sounding answer that was technically wrong on something important. The problem isn't the model — it's the prompt. Without structure, AI gives you a general answer. With a structured workflow that encodes the actual professional process (the four-part R&D credit test, the OECD BEPS transfer pricing framework, the ASC 606 five-step model), you get analysis you can actually use.

That's what ClaudeFinanceLab's Tax & Transfer Pricing templates are built for. Each SKILL.md template loads into Claude's Projects as a system prompt, telling it exactly how to approach a specific tax workflow. The result is less "here's some general guidance" and more "here's the completed memo."

Year-End Tax Planning

Q4 is when the real planning happens — there's still time to act before December 31. The challenge is running multiple scenarios quickly enough to have a real conversation with the client, not just hand them a summary after the year closes. Claude handles the scenario math; you handle the judgment call on what actually makes sense for the client's situation.

• "Individual tax planning for a married couple: W-2 income $380K, RSU vesting $95K (all included in W-2), long-term capital gains $42K from stock sales, rental income $28K net. Estimated marginal rate: 32% federal, 9.3% California. Analyze: (1) Is the couple in the NIIT zone? (2) Roth conversion opportunity: they have $340K in traditional IRA — what's the optimal conversion amount to fill the 32% bracket? (3) QBI deduction: do they have any passthrough income? (4) Charitable strategy: donor-advised fund vs direct giving. Quantify each strategy's tax savings."
• "S-corp owner tax planning: net profit before owner compensation $420K. Owner is currently taking $180K salary. Model the optimal salary level: (1) what is the SE tax savings from keeping salary at current level vs. $120K vs. $200K? (2) At each salary level, what is the QBI deduction? (3) Compute the total tax (SE + income) at each salary scenario and find the optimal salary. Assume: married filing jointly, no other income, standard deduction."
• "Year-end capital loss harvesting analysis: client has $120K in unrealized long-term losses in a taxable brokerage account. They have $35K in realized gains from earlier in the year. Model: (1) tax benefit from harvesting all $120K in losses — how much tax is saved? (2) Wash sale risk — which holdings should they wait 31 days before repurchasing? (3) Should they use losses to offset ordinary income (up to $3K/year) or save carryforwards for future gain years? Assume 20% LTCG rate + 3.8% NIIT."

Transfer Pricing Documentation

Transfer pricing is both high-risk and high-billing-rate work — which also makes it the area where partners are most skeptical about AI. The concern is valid: bad TP analysis is worse than no analysis. But the parts of a TP study that take the most time are also the most structural: the functional analysis (FAR), the benchmarking rationale, the arm's length range narrative. Claude handles those scaffolded sections well when given the right framework. Counsel still needs to review the conclusions.

• "Transfer pricing functional analysis (FAR) for intercompany services: Parent company (US) provides IT, finance, and HR shared services to European subsidiaries for a charge of $4.2M. Functional analysis: Parent performs: system development, maintenance, governance, risk management. Subsidiaries: routine users, limited risk. Asset ownership: Parent owns all IP and systems. Risk: Parent bears development risk, market risk. Using TNMM with operating margin as PLI — what comparable search would I run and what is a reasonable arm's length margin for routine IT services in Europe? Benchmark: technology services companies, net operating margin 8–15%."
• "Transfer pricing for IP licensing: US parent owns software IP and licenses it to German subsidiary (distributor). Royalty rate currently 8% of the subsidiary's net revenue. How do I defend this rate? Describe: (1) the CUP method analysis using third-party software license rates, (2) the profit split method as a secondary check, (3) how the OECD's 'DEMPE' functions apply here, (4) what documentation is required under German local file rules (§90 AO) and the OECD master file. Subsidiary revenue €38M, operating margin before royalty: 14%, after royalty: 6%."

ASC 842 Lease Accounting

You'd think ASC 842 would be solved by now, given how long it's been effective. But companies keep acquiring new entities with messy lease portfolios, modifying existing leases, and discovering embedded leases in service contracts that nobody noticed during implementation. The classification test and ROU/lease liability math is deterministic — Claude runs it correctly every time, which is more than can be said for the spreadsheet models floating around most accounting departments.

• "ASC 842 classification: 5-year office lease, monthly rent $28,500, with two 5-year renewal options that management is NOT reasonably certain to exercise. IBR at commencement: 5.8%. No purchase option. Fair value of building: approximately $9M (building is 3 floors of a 30-floor tower — the leased space is not substantially all). Classify as operating or finance lease. Calculate: (1) right-of-use asset, (2) lease liability at commencement, (3) quarterly amortization for the first 4 quarters, (4) the journal entries at commencement and at each quarter-end."
• "Lease modification analysis: existing 7-year office lease, 4 years remaining, book value of lease liability $485K. Landlord is offering: (1) early termination at 2 years for a $60K payment, (2) extension of the lease for 3 additional years at the same rent with a 3-month free rent period, (3) downsizing to 60% of space (a decrease in scope — this is a partial termination requiring separate accounting). Walk through the ASC 842 accounting for each option and indicate which requires remeasurement and how."

ASC 606 Revenue Recognition

The five-step model under ASC 606 sounds straightforward until you're staring at a multi-element software contract with variable consideration, material rights, and a consulting component that may or may not be a distinct performance obligation. Claude works through each step in sequence and produces the accounting memo — including the SSP allocation and the quarter-by-quarter recognition schedule — that your auditors will ask for anyway.

• "ASC 606 analysis for a SaaS contract: $240K total contract value — $180K for a 3-year software subscription, $40K for implementation services, $20K for annual training. Term: 36 months. Start date: January 1, 2026. Is implementation a separate performance obligation? What are the SSPs? Allocate the TCV. When does each obligation recognize revenue? Produce the revenue recognition schedule by quarter for all 12 quarters."
• "Variable consideration — most likely amount vs. expected value: Our company has a $2M base contract with a $400K performance bonus paid if we achieve >98% uptime for 12 months. Historical data: we've achieved >98% uptime in 8 of the last 10 years. Apply ASC 606-10-32 constraint analysis: should we include the performance bonus in the transaction price? Use both the 'most likely amount' method and 'expected value' method. Apply the constraint: is it highly probable the inclusion won't result in a significant reversal?"

R&D Tax Credit Analysis

The Section 41 R&D credit is probably the most commonly under-claimed incentive in the tax code. Part of the problem is that clients genuinely don't know what qualifies — they assume R&D means scientists in lab coats, not software engineers testing architectural approaches or process engineers eliminating production uncertainty. Claude applies the four-part test to specific project descriptions and flags which activities qualify, which don't, and why. That's the hardest part of the analysis to document quickly.

• "R&D tax credit four-part test for a fintech company: Our engineering team spent 60% of their time on the following projects this year: (1) Developing a new machine learning model to detect fraudulent transactions — no existing solution; multiple architecture approaches tested; (2) Building a new API integration with a third-party payment processor — used documented API specs — required some customization but generally followed instructions; (3) Maintaining and patching existing production software — routine updates. Analyze each activity against the four-part test (permitted purpose, technological in nature, elimination of uncertainty, process of experimentation). Which qualify?"
• "R&D credit calculation using the Alternative Simplified Credit (ASC) method: Current year QREs $1.8M. Prior 3 years: FY23 $1.1M, FY24 $1.35M, FY25 $1.6M. Compute: (1) 3-year average QREs; (2) 50% of that average; (3) excess QREs above that; (4) ASC credit at 14%; (5) effective credit rate. Also compute what the credit would be if this is the first year claiming (6% of all QREs). Compare to the regular credit (RRC) method assuming fixed base percentage of 4.2% and 4-year average gross receipts of $22M."

Multi-State Tax Nexus Analysis

Post-Wayfair, the nexus exposure conversation happens earlier in the client relationship — often because a CFO just got a notice from California and realized they've never filed there. The analysis isn't complex in principle, but it's tedious: economic nexus thresholds vary by state, SaaS taxability varies by state, and voluntary disclosure timelines and look-back periods vary by state. Claude maps the full exposure picture in one pass.

• "Multi-state nexus exposure assessment: SaaS company based in Delaware, primarily selling to businesses. We have remote employees in: California (2 sales reps), Texas (1 developer), New York (1 support rep). Revenue by state (top 5): CA $2.1M, NY $1.8M, TX $1.4M, FL $0.9M, IL $0.7M. Physical presence: office in Delaware only. Assess: (1) income tax nexus in each of the top 5 revenue states; (2) sales tax nexus — is SaaS taxable in each? (3) Economic nexus thresholds for each state; (4) Estimated back-tax exposure and VDA recommendation."

Tax Research Memos

A well-structured research memo is the deliverable that protects both the client and the preparer. It documents the question, the authorities, the analysis, and the conclusion with a confidence level that holds up to scrutiny. The issue is that writing a good memo takes time — time you often don't have when a client calls with an urgent question. Claude drafts the structure, cites the relevant IRC sections and regulations, and gets you 70% of the way there. You review and finalize. That's a meaningful time saving on work that has to get done regardless.

• "Tax research memo on the Section 1202 Qualified Small Business Stock exclusion for a founder selling stock after 7 years: Company was a C corporation since founding. I held 500,000 shares, originally issued at $0.001/share. I'm selling at $18/share — total proceeds $9M. Issue: What is the amount excluded from capital gains? Are there limits? What are the issuer requirements the company must have met? Is there an AMT concern? Cite: IRC § 1202, Rev. Proc. 2013-13 (no — wrong cross-reference — cite relevant PLRs if any). Confidence level for the exclusion: should?"

Estate and Gift Tax Planning

High-net-worth client engagements often require running gift tax scenarios quickly before year-end — annual exclusion gifting, 529 superfunding, GRAT modeling, and charitable strategies. Claude structures these correctly under IRC § 2503, § 2522, and § 2642 (GST) and produces a memo the estate attorney can review.

• "Gift tax planning for a couple with $14M in total assets. 2026 annual exclusion: $19,000 per person per donee. They have 3 adult children and 6 grandchildren. Analyze: (1) maximum annual exclusion gifting this year to all 15 donees; (2) 529 superfunding for 4 grandchildren — how does the 5-year election work and what are the gift tax consequences?; (3) GRAT strategy: $2M asset expected to grow at 9% per year — calculate the annuity payment using the 7520 rate of 4.8% and the taxable gift if any; (4) What's left of their lifetime exclusion if they've previously used $3.8M?"
• "Charitable lead annuity trust (CLAT) analysis: Client wants to fund a CLAT with $3M in appreciated stock (FMV $3M, basis $400K). Terms: 10-year annuity payout to donor-advised fund, remainder to children. Assume IRS 7520 rate 4.8%. Calculate: (1) required annuity payment for a zeroed-out CLAT (no taxable gift to children); (2) income tax treatment of CLAT distributions to the charity; (3) capital gains treatment on the transfer of appreciated stock; (4) what the children realistically receive at the end of the 10-year term if the assets grow at 7% per year."

Where to Start

The Tax & Transfer Pricing category has seven templates. If you're a CPA in public practice, start with the Tax Research Memo Writer — it's the most universally useful and takes about 30 seconds to set up in a Claude Project. If you do a lot of international work, the Transfer Pricing Documentation Generator is worth loading next. For accounting teams at corporates, the ASC 842 and ASC 606 templates save the most time. Each one is a system prompt you paste into Claude's Projects — no software to install, no API key required.

Related guides: IFRS 17 CSM Roll-Forward for insurers reporting under IFRS, Budget & Forecasting AI for FP&A teams, and AML & Financial Crime for compliance-adjacent tax work.

Frequently Asked Questions

Can I use Claude AI for actual tax advice?

Claude is a drafting and analysis tool, not a licensed tax advisor. CPAs use it to draft research memos, model scenarios, structure documentation, and produce first drafts of deliverables they then review and finalize. The CPA takes professional responsibility for all conclusions. Claude should not be used as the final authority on tax law — always verify citations against current IRC text, Treasury regulations, and relevant case law.

Does Claude know current tax law?

Claude's training includes the IRC, Treasury regulations, and major tax court decisions, but it has a knowledge cutoff and is not updated in real time. For 2026 tax law changes, recent IRS guidance (Rev. Procs., PLRs, Chief Counsel memos issued after training), and state tax law changes that postdate training, verify Claude's output against primary sources. Claude performs most reliably on structural questions (how does the ASC 842 lease classification test work, what are the four parts of the R&D credit test) rather than on rapidly-changing guidance.

How should CPAs set up Claude for tax work?

The most effective CPA setup is a Claude.ai Pro account with a Project for each client type (public company, private business, high-net-worth individual). Add context to each Project: the client's business, relevant tax positions, and the specific SKILL.md template from ClaudeFinanceLab's Tax & Transfer Pricing category. Every conversation in that Project inherits the context, so you don't re-explain the client situation each time. Claude.ai Pro costs $20/month; no API key, no IT approval required.

Is transfer pricing documentation a good use case for Claude?

Yes — transfer pricing documentation is one of Claude's strongest CPA use cases. The functional analysis (FAR), benchmarking rationale, and arm's length range narrative are structural and time-consuming to write from scratch. Claude drafts these sections correctly when given the intercompany transaction details, entity functions, asset ownership, and risk allocation. The tax counsel or transfer pricing specialist reviews and finalizes. This workflow is particularly valuable for mid-market TP studies where full Big 4 fees aren't justified but proper documentation is still required.