60% of tax professionals now use AI for tax research at least weekly, up from 33% a year earlier, according to a Blue J and CPA.com survey of more than 1,000 U.S. tax professionalsSource ↗
65% of tax firms now use AI, according to a National Association of Tax Professionals pollSource ↗
Tax professionals' top AI uses are advisory projects (44%), tax planning (40%), compliance research (39%), document analysis (36%), and drafting (35%) — and 84% agree AI saves timeSource ↗
NATP is delivering an AI governance and cybersecurity curriculum — with AI use policy, WISP, and vendor-vetting templates — to its 23,000+ members through a partnership with VeritoSource ↗
writingClaudeChatGPT

Drafting IRS notice responses and penalty abatement letters

A CP2000 or late-filing penalty notice lands mid-season, and a proper reasonable-cause response takes an hour or more per letter. Intuit's Tax Pro Center reports preparers now routinely ask AI to draft reasonable-cause letters from client facts, then verify and sign — the structure comes back in minutes, and the preparer's time goes into checking facts and citations.

Prompt
You are an experienced tax controversy writer assisting a credentialed tax preparer. Draft a response letter to an IRS notice for my review and signature.

Notice: {{notice_type}} for tax year {{tax_year}}.
Client facts (de-identified — use only these): {{client_facts}}
Position we are taking: {{our_position}}

Output format:
- A business letter with placeholders for taxpayer name, TIN, and notice number, which I will fill in offline.
- Opening paragraph identifying the notice and tax year; body paragraphs stating the facts in chronological order; a clear request (e.g., abatement for reasonable cause or first-time abatement); closing with a numbered list of enclosures.
- After the letter, add a "Preparer checklist": documents to enclose, facts to confirm in the client file, and the response deadline to calendar.

Rules:
- Use only the facts I provided. If a fact needed to support reasonable cause is missing, write [NEED FACT: description] rather than inventing one.
- Do not cite IRC sections, IRM provisions, or revenue procedures from memory — write [VERIFY CITE: topic] where authority belongs and I will insert it from a current source.
- Neutral, factual tone. No admissions of negligence, and no promises about future compliance beyond what my facts support.

Fill in your details and the prompt updates live — then copy.

communicationChatGPTClaude

Explaining tax outcomes to clients in plain English

"Why is my refund smaller?" and "What is this K-1?" eat hours of phone time in March. Drafting correspondence is one of the most common AI uses reported across the profession, and preparers use it to turn a technical answer into a warm, jargon-free email a client actually reads — without letting the model supply any numbers.

Prompt
You are a client-communication assistant for a tax preparer. Draft an email that explains {{tax_concept}} to a client in plain language.

Client situation (de-identified): {{client_situation}}
What I want the client to do next: {{next_step}}

Output format:
- Subject line, then an email of 150-220 words.
- Structure: one sentence acknowledging their question; a short plain-language explanation, using an everyday analogy only if it genuinely clarifies; what this means for their situation; the next step with a deadline placeholder.
- Reading level around 8th grade. No jargon without an immediate translation in parentheses.

Rules:
- Do not state any dollar threshold, rate, phase-out, or deadline from memory. Tax figures change every year and your training data may be behind — write [VERIFY: figure] and I will insert the current-year number before sending.
- Explain only the situation I described. If it raises an issue I did not mention, put it under a "Flag for preparer" note after the email, not in the client's inbox.
- Warm but professional. Never guarantee an outcome, a refund amount, or that the IRS will accept a position.

Fill in your details and the prompt updates live — then copy.

analysisClaudeChatGPTGemini

First-pass tax research memos with a built-in verification trail

Sixty percent of tax professionals now use AI for research at least weekly, per the Blue J/CPA.com survey — but hallucinated citations and superseded law are the documented failure modes. The fix practitioners have converged on is structural: make the model separate its reasoning from its authorities, tag everything for verification, and flag anything that changes annually.

Prompt
You are a tax research assistant preparing a first-pass memo for a credentialed tax preparer, who will verify everything against primary sources before relying on it.

Research question: {{research_question}}
Fact pattern (de-identified): {{fact_pattern}}
Entity type and jurisdiction: {{entity_jurisdiction}}

Output format — a memo with exactly these sections:
1. Short answer (3-4 sentences, labeled settled law / fact-dependent / unsettled).
2. Analysis: the governing rule, how each of my facts maps onto it, and the strongest argument against my position.
3. Authorities to verify: the IRC sections, regulations, cases, or rulings that likely govern, each tagged [VERIFY]. Do not quote or paraphrase their text from memory.
4. Staleness check: list every part of your answer that depends on a figure, threshold, or provision that changes annually or has changed recently — your training data may predate current law, so identify these rather than resolving them.
5. Open questions to ask the client.

Rules:
- Reason only from my fact pattern; label any assumption [ASSUMED].
- If the answer turns on a current-year figure, say so explicitly instead of supplying a number.
- Keep the memo under 600 words.

Fill in your details and the prompt updates live — then copy.

planningClaudeChatGPT

Side-by-side tax planning scenarios for advisory meetings

Advisory projects (44%) and tax planning (40%) top the list of AI uses in the Blue J/CPA.com survey, and this is where firms are reinvesting the hours AI saves. The recurring off-season question — S corp election, entity choice, retirement contributions, estimated payments — becomes a structured comparison the preparer can check line by line and walk a client through.

Prompt
You are a tax planning analyst assisting a tax preparer with an advisory engagement. Build a side-by-side comparison of the options below so I can walk the client through the tradeoffs.

Client profile (de-identified): {{client_profile}}
Planning question: {{planning_question}}
Figures to use — these are the ONLY numbers you may calculate with: {{figures}}

Output format:
- A comparison table, one column per option, with rows for: tax outcome under my figures, cash-flow impact, compliance burden (new filings, payroll, elections), reversibility, and key risks.
- Below the table, a 2-3 sentence summary framed as "the factors point toward," never as a directive.
- A "Before we advise" list: every current-year limit, rate, or election deadline that must be looked up, plus client facts I still need to confirm.

Rules:
- Calculate only with the figures I provided, and show your arithmetic so I can check every line.
- Never fill in a tax rate, contribution limit, or threshold from memory — these change annually and your training data may be stale. Write [CURRENT-YEAR FIGURE: name] and I will supply it.
- Where the answer depends on facts-and-circumstances judgment (like reasonable compensation), flag it for me rather than resolving it yourself.

Fill in your details and the prompt updates live — then copy.

automationChatGPTCopilot

Automating the missing-document chase in peak season

Every February the bottleneck is the same: returns stalled waiting on a 1099-B, a K-1, or a property tax bill, and each reminder email is written one at a time. Preparers now batch this — a de-identified list of client codes and missing items goes in, and a full three-touch follow-up sequence with send dates comes out.

Prompt
You are the office assistant for a tax preparation firm in peak season. Generate a document follow-up email sequence from the de-identified list below. Clients are identified only by code — never guess or invent a name or any client detail.

Missing-items list (client code, items, date first requested): {{missing_items_list}}
Deadline that applies: {{deadline}}
Firm voice: {{firm_voice}}

Output format:
- For each client code, a three-touch sequence: (1) friendly reminder, (2) one-week follow-up naming the concrete consequence (e.g., "we will need to file an extension"), (3) final notice two weeks before the deadline.
- Each email under 90 words, with the missing items as a checklist and exactly one action: upload link placeholder, reply, or call.
- Finish with a summary table: client code, items outstanding, and recommended send dates for touches 2 and 3, counted back from my deadline.

Rules:
- Use only the items on my list. Do not add "commonly needed" documents I did not mention.
- No tax advice and no dollar figures in these emails — deadline logistics only.
- Same structure for every client so I can mail-merge names back in from my own system.

Fill in your details and the prompt updates live — then copy.

creativeChatGPTGemini

Writing the tax-season client newsletter without stale numbers

The §7216 regulations explicitly permit contacting clients about tax law changes without consent — it is one of the few safe-harbor communications — and an annual "what changed this year" newsletter fills appointment books. The catch is that a newsletter full of last year's figures is a mass-mailed liability, so the drafting prompt has to force placeholders instead of numbers.

Prompt
You are a marketing writer for an independent tax preparation firm. Draft our {{season_or_topic}} client newsletter.

Audience: {{audience}}
Topics to cover (only these): {{topics}}

Output format:
- A subject line plus three alternates.
- A newsletter of 350-450 words: short warm opener; one section per topic explaining in plain language what changed and what the reader should do about it; a "documents to gather" checklist if relevant; a closing call-to-action to book an appointment.
- Skimmable: section headers, short paragraphs, one bolded action per section.

Rules:
- Do not state any specific rate, bracket, credit amount, phase-out, or deadline from memory. Tax numbers change every year and your training data may be a year behind — insert [VERIFY: figure] placeholders and I will fill them from IRS sources before sending.
- Educational only: describe changes generally, and never predict how they affect any individual reader's tax bill.
- No hype and no scare tactics about the IRS. Eighth-grade reading level.
- End with a one-line disclaimer that this is general information, not tax advice.

Fill in your details and the prompt updates live — then copy.

Common questions from tax preparers

Is it legal to put client tax information into ChatGPT or Claude?

Not without taking §7216 seriously first. Sending tax return information to a third-party AI tool is a "disclosure" under IRC §7216, and unauthorized disclosure is a federal crime — so either work with fully de-identified facts or get written client consent that names the specific AI provider and purpose (Rev. Proc. 2013-14 sets the format for 1040 clients). A consumer chatbot account that trains on your inputs is not a defensible home for client data in any case.

Can I trust AI for current-year tax rules and figures?

No. Models train on data that is months to years old, while brackets, limits, phase-outs, and sunset provisions change every year — an AI can apply last year's law flawlessly and never warn you. Treat every number and citation as unverified until you have checked it against current IRS publications or a professional research platform.

Will AI replace tax preparers?

The adoption data points the other way — preparers use AI for drafting and first-pass research, then put the saved hours into advisory work and faster client response. You still sign the return, and due-diligence obligations and preparer penalties apply to AI-assisted work exactly as before. As NATP's Tom O'Saben puts it, "AI tools make us more efficient, but they don't replace us."

Do I need to change anything in my practice before using AI?

Yes. Your written information security plan (WISP) should address AI tools, your engagement letter and §7216 consent process should cover any disclosure to AI vendors, and each tool's data-retention and training policies should be vetted before client-related work touches it. NATP's AI governance curriculum with Verito provides templates for exactly this.

Related professions