Tax season broke me in 2024.
I was doing 70-hour weeks from late January through April 15. Sleeping five hours a night. Living on coffee and the kind of takeout you order at 11 PM because you're still at the office. By March I was making mistakes I should have caught. By April I was questioning whether I wanted to stay in the profession.
Tax season 2026 was different. Same client load. Same deadlines. Same complexity. But I finished April 15 at a reasonable hour, with my health intact and my judgment sharp through the last return.
The difference was AI. Not as a replacement for my work — as a multiplier for the parts that don't require my brain.
Here's the complete playbook. Every prompt I actually use during tax season, organized by the phase of the return process where it saves the most time.
Document collection is the single biggest time sink in tax season that nobody talks about. You send an email. You get half the documents. You send a follow-up. You get more documents but not all. You chase. You escalate. By the time you have everything, you've spent 45 minutes on communication for a 2-hour return.
AI doesn't fix clients who don't respond. But it makes your side of the communication dramatically faster.
"Draft a professional document request email for a [INDIVIDUAL / BUSINESS] tax client for [YEAR] returns.
Client type: [W-2 ONLY / SELF-EMPLOYED / S-CORP OWNER / INVESTMENT INCOME / MULTI-STATE / RENTAL PROPERTIES]
Include:
1. Numbered checklist of documents needed (organized by category)
2. Instructions for secure file sharing
3. Deadline with specific date
4. What happens if they miss the deadline
5. Friendly but professional tone
6. One line that addresses the most common question clients ask at this stage
Format for easy reading on mobile."
This produces a clean, comprehensive email in 30 seconds. You personalize the opening, verify the checklist covers this specific client's situation, and send.
For clients who haven't responded after 7-10 days, the follow-up email has a specific tone problem: friendly enough to preserve the relationship, firm enough to actually get results.
"Draft a follow-up email to a tax client who hasn't submitted their documents yet. Original request sent [DATE]. Deadline is [DATE].
Tone: warmer than transactional, firmer than passive. Reference that we're trying to help them meet the deadline without penalty. Offer a brief 15-minute call if they're stuck on what's needed.
Length: 4-5 sentences max. Subject line should create urgency without being dramatic."
This is the kind of email that used to take me 10 minutes to write because I'd second-guess the tone. Now it takes 90 seconds.
Once documents arrive, you still have to organize them, read them, extract the data you need, and flag anything unusual. This is grunt work that AI is excellent at accelerating.
"Extract the following information from the tax documents below and format as a structured table:
- Document type (W-2, 1099-NEC, 1099-MISC, 1099-INT, 1099-DIV, etc.)
- Payer name and EIN (last 4 digits only for privacy)
- Recipient name (anonymized as "Client")
- Tax year
- Box-by-box amounts
- Any unusual codes or flags
If any documents appear incomplete or have inconsistencies, flag them.
Documents: [PASTE TEXT CONTENT OR USE IMAGE UPLOAD]"
Important: remove full SSNs and full EINs before pasting. The 5-step anonymization workflow applies to tax documents too — strip identifying data, keep the amounts, run the prompt.
One of the highest-value things you can do for a client is spot what changed from last year. AI makes this almost automatic.
"Compare this year's and last year's tax data for [CLIENT TYPE] and identify:
1. Significant changes in income (any source over [10%] variance)
2. New income sources this year
3. Income sources that disappeared
4. Significant changes in deductions
5. Changes in filing status indicators
6. Items that might trigger different treatment this year
For each change, note: the amount, the percentage change, and one possible explanation worth asking the client about.
Prior year: [PASTE KEY FIGURES]
Current year: [PASTE KEY FIGURES]"
This one prompt has caught more client situations worth asking about than any manual review I used to do. Tired-brain me missed things. AI doesn't get tired.
For any return with actual complexity, you'll have research questions. AI is excellent at the first pass — mapping the territory, finding relevant sections, identifying planning opportunities.
The rule from my earlier article still applies: verify every citation before using it. AI fabricates tax authorities. If you haven't opened the actual IRC section, don't trust what AI said about it.
This is where AI creates real value during tax season — not by doing the research, but by making sure you don't miss planning opportunities.
"Based on this client's tax profile, identify potential tax planning opportunities I should consider and discuss with them before filing.
Client profile:
- Filing status: [STATUS]
- Income level: [RANGE]
- Income sources: [W-2, 1099, BUSINESS, INVESTMENT, etc.]
- Retirement accounts: [401K / IRA / NONE / SEP / SOLO 401K]
- Business entity: [NONE / SOLE PROP / S-CORP / LLC]
- State: [STATE]
- Children/dependents: [NUMBER]
- Home ownership: [YES/NO]
- Notable items from intake: [ANYTHING UNUSUAL]
For each opportunity, provide:
1. The strategy
2. Estimated tax savings (rough range)
3. Deadline to implement (before year-end, before filing, etc.)
4. Requirements/restrictions I should verify
5. Whether this should be done this year or planned for next year
Rank by potential savings. Flag any that are aggressive positions requiring careful documentation.
IMPORTANT: I will verify all rules independently. If you are not certain about any rule, explicitly say so."
For discrete research questions that come up during preparation, use a structured prompt that forces AI to hedge where it should:
"Research this tax question: [QUESTION]
Client facts: [FACTS]
Jurisdiction: [STATE/FEDERAL]
Provide:
1. Brief answer (yes/no/depends, with the core logic)
2. Relevant IRC sections and Treasury Regulations
3. Recent IRS guidance (Rev Rulings, Notices, Procedures)
4. Relevant court cases
5. Common positions I should consider
6. Key facts that could change the answer
7. Audit risk factors for this position
I will verify all citations. If you're uncertain about any specific citation, explicitly say so. Do not fabricate."
Then verify every citation before using anything from the response.
AI doesn't prepare returns for you. Tax software does that. But AI accelerates the decision-making and judgment that tax software can't automate.
For specific schedules that require judgment, I have prompts for each one. Here are two examples:
Schedule C review:
"Review this Schedule C data for reasonableness and flag anything that needs attention:
- Gross receipts: [AMOUNT]
- By category of expense: [LIST CATEGORIES AND AMOUNTS]
- Net profit: [AMOUNT]
- Profession: [TYPE]
Flag:
1. Categories that are unusually high for this profession
2. Categories that are unusually low or missing
3. Expense categories that are common audit triggers (meals, vehicle, home office)
4. Ratios that look off (COGS to revenue, expense to revenue)
5. Anything that might indicate misclassification between business and personal
For each flag, suggest what to verify or ask the client about."
Schedule E review:
"Review this rental property data and flag common issues:
Property details: [NUMBER OF UNITS, TYPE, LOCATION]
Gross rents: [AMOUNT]
Expenses by category: [LIST]
Depreciation: [AMOUNT]
Year acquired: [YEAR]
Flag:
1. Missing depreciation or recapture issues
2. Expense categories that might indicate improvements vs. repairs
3. Loss deduction limitations that may apply (passive activity, AGI phase-out)
4. Short-term vs long-term rental indicators
5. Personal use days that might trigger vacation home rules
Note anything that requires the client's confirmation before I can complete the return."
The last step before e-filing is the review. This is where AI adds a second set of eyes that doesn't get tired at 9 PM during the April 10 crunch.
"Review this [FORM TYPE, e.g., 1040 / 1120S / 1065] tax return data for common errors:
Return data: [PASTE KEY FIGURES — total income, deductions, credits, tax, payments, balance]
Prior year comparison: [PASTE LAST YEAR'S SAME FIGURES]
Check for:
1. Math errors or obvious calculation mistakes
2. Significant changes from prior year that need explanation
3. Deductions or credits that seem unusual for this return type
4. Missing schedules based on income types shown
5. Signs of misclassified income
6. Red flags for audit risk
7. State filing considerations I might be missing
Rate overall return completeness on a 1-10 scale and note anything that should be double-checked before filing."
I run this prompt on every return before filing. It catches things I would have missed at least 30% of the time.
For clients who want to understand their return (not all do, but the ones who do really want to), AI writes the explanation better than I do.
"Summarize the key outcomes of this tax return for the client in plain English:
Return type: [TYPE]
Total tax: [AMOUNT]
Total payments/withholding: [AMOUNT]
Refund or amount due: [AMOUNT]
Effective tax rate: [PERCENT]
Prior year tax: [AMOUNT]
Key factors this year: [NOTEWORTHY ITEMS]
Include:
1. Bottom line in one sentence
2. What drove the change from last year
3. 2-3 planning items to think about for next year
4. What they need to do next (pay, review, sign)
Tone: clear, professional, no jargon. Write for someone who doesn't understand tax."
Tax returns are full of the most sensitive client data you handle. SSNs, full financial pictures, family information. All of this is a hard "NEVER paste into AI" category.
For any prompt involving client tax data, run through this checklist before pasting:
Better yet: use ChatGPT Team or Claude Team for tax work. These plans contractually do not train on your data. For tax season, the $25-30/user/month is a reasonable cost for the confidentiality protection.
Here's what using AI during tax season 2026 did for me:
Net effect: about 3-4 hours saved per return. For an accountant doing 200 returns during tax season, that's 600-800 hours. That's the difference between finishing April 15 depleted vs. finishing April 15 still functional.
Same quality. Same accuracy. Same professional standards. Just not destroying myself to get there.
That's what AI is actually for during tax season. Not replacing the professional judgment that returns require. Just making sure the administrative weight doesn't crush you while you're trying to apply it.
The complete tax research prompts (Section 3 of the free PDF) include verification checklists and data privacy reminders built in. Download the PDF here.
Get 50 copy-paste prompts delivered to your inbox — free.
Get the Free PDF →