AI for Tax Preparers

Updated January 2026 | 8 min read

It's February. You're preparing returns for 180 clients. The retired teacher asks the same question about pension taxation she asked last year. The contractor forgot to track mileage again. The landlord has rental income from three properties but can't remember which expenses go with which property. The small business owner hands you a shoebox of receipts and expects you to categorize everything.

You know most of this because you've done their taxes for five years. But the AI doesn't. Every question you ask it starts from zero, even though you're dealing with the same situations you handled last tax season.

Why Tax Prep Breaks Standard AI

Tax preparation is pattern recognition across time. Same client, different year, similar issues. Income changes slightly, deductions shift, but the core situation stays consistent. The value isn't just calculating taxes, it's catching changes that matter and knowing which questions to ask based on what happened last year.

You ask ChatGPT about home office deduction rules for a client. It explains the criteria and calculation methods. Accurate, but generic. You spend 15 minutes explaining that this client claimed it last year for 200 square feet, works from home full-time, and the IRS questioned it during a 2022 audit but accepted the documentation.

The AI gives you specific guidance for this situation. Next year, same client, same question, and the AI has forgotten the audit, the accepted documentation approach, and what you claimed previously.

What Tax Preparers Need AI to Remember

Tax work is continuity work. This year's return builds on last year's. Deduction patterns matter. Changes trigger questions. Client situations evolve slowly, not radically.

You need AI that knows:

• Client tax situations from prior years
• Recurring deductions and typical documentation
• Changes from year to year that affect planning
• IRS correspondence and resolution history
• Client communication patterns and common mistakes
• State-specific rules for multi-state filers

When AI remembers this context, it stops giving you tax code citations and starts helping you manage client relationships and catch filing errors before they happen.

How Memory Changes Tax Prep AI

Claude Code reads one markdown file at the start of every session. That file contains everything about your clients and their tax situations that the AI needs to know.

You write in your memory file:

• Client profiles with tax situation summary and filing history
• Prior year deductions and supporting documentation types
• Year-over-year changes that affect current filing
• IRS notice history and how issues were resolved
• Client-specific notes about documentation habits and common gaps
• Multi-state filing requirements and apportionment methods

The AI reads this before every conversation. It knows your clients' tax patterns, not just the tax code.

Client Questions That Reference History

Your client calls: "Can I still deduct my home office like last year?"

Standard AI tells you the general rules. AI with memory knows:

• Client claimed $5,000 simplified method last year
• Works from home three days a week (not full-time)
• You used actual expense method in 2023, switched to simplified in 2024
• Employer confirmed no reimbursement policy

You tell Claude Code: "Home office question for the Martinez account."

The AI responds: "Martinez claimed $5,000 simplified method last year (300 sq ft). Still works from home three days a week per last year's notes. If situation unchanged, use same method and amount. If square footage changed or work schedule shifted, recalculate. Employer policy confirmed no reimbursement last year — verify still true."

You're not searching through last year's return. You're having a conversation that assumes continuity.

Deduction Tracking That Catches Changes

A self-employed graphic designer has consistent deductions: software subscriptions, equipment, home office, professional development. This year, she hands you receipts and you start categorizing.

Your memory file contains:

• "Chen — self-employed designer. Typical deductions: Adobe CC ($600), equipment ($2-3K), home office (200 sq ft simplified method), conferences ($1-2K). 2024 note: purchased new laptop in Q4, claimed Section 179. 2025 watch: if equipment expenses low, verify laptop depreciation vs. expensing choice."

You tell the AI: "I'm categorizing Chen's receipts. Equipment total is $800."

The AI flags: "Chen typically has $2-3K in equipment. Last year purchased laptop and used Section 179. Verify if lower expense is accurate or if receipts are missing. If accurate, ensure laptop isn't being depreciated if it was expensed last year."

You catch that she forgot to include her new monitor and camera. Pattern recognition prevents missed deductions.

Multi-Year Strategy That Builds Over Time

A client sold rental property this year. You need to calculate depreciation recapture, but that requires knowing the original basis, accumulated depreciation, and improvement costs from when they bought it eight years ago.

Your memory file tracks long-term positions:

• "Rodriguez rental property (123 Oak St) — purchased 2017 for $280K (land $60K, building $220K), 27.5-year depreciation. Improvements: roof 2019 ($15K depreciated separately), HVAC 2021 ($8K depreciated separately). Accumulated depreciation through 2024: $74,181. Sale year: 2025."

You tell the AI: "Rodriguez sold the Oak Street rental. Sale price $420K."

The AI calculates:

• Adjusted basis: $280K - $74,181 + $23K improvements = $228,819
• Gain: $420K - $228,819 = $191,181
• Depreciation recapture: $74,181 (taxed at 25%)
• Remaining gain: $117,000 (long-term capital gains rates)

You're not hunting through eight years of files. The AI tracked the basis adjustments as they happened.

IRS Notice Response That References Resolution History

A client receives a CP2000 notice. The IRS says they have unreported income. You've dealt with these for this client before.

Your memory file contains:

• "Thompson — received CP2000 in 2023 for unreported 1099-MISC from side consulting work. Responded with documentation showing income was reported under business Schedule C. IRS accepted response, no additional tax. 2024: same issue, same resolution approach worked. Pattern: client's business name differs from personal name, IRS matching algorithm flags mismatch."

You tell the AI: "Thompson got another CP2000 for unreported 1099-MISC."

The AI responds: "Same issue as 2023 and 2024. Income likely reported under business Schedule C but IRS matching flagged name mismatch. Draft response referencing Schedule C line items, include business name explanation, attach Schedule C page. Previous responses using this approach were accepted without additional tax."

You draft the response in 15 minutes using the approach that already worked twice.

Client Communication That Prevents Repeat Mistakes

Some clients forget the same things every year. One forgets to send K-1s until you ask. Another always miscalculates estimated payments. A third never tracks charitable donations properly.

Your memory file includes client behavioral patterns:

• "Davis — partner in law firm, always late sending K-1. Send reminder in early February, follow up mid-February. K-1 typically arrives late March. Plan to file extension."
• "Kumar — estimated payments always undercalculated. 2024 owed $2,800 penalty. Set up automatic quarterly reminders with calculation worksheet. Verify Q4 payment was made before preparing return."

It's early February. You tell the AI: "Generate this week's client reminders."

The AI drafts: "Davis — K-1 reminder (law firm partnership, typically late). Kumar — Q4 estimated payment verification (history of underpayment)."

You send proactive reminders that prevent last-minute problems.

Real Use: A Solo Tax Practice

A CPA running a solo practice with 220 individual and small business clients built a memory file with:

• Client profiles with tax situation summaries
• Prior year deduction patterns and amounts
• Multi-year depreciation and basis tracking for rentals and businesses
• IRS notice history and successful resolution approaches
• Client communication patterns and common mistakes

She uses Claude Code for:

• Client question responses that reference prior years
• Deduction review flagging when patterns change
• Multi-year calculation verification (basis, depreciation, carryovers)
• IRS notice responses using proven approaches
• Proactive client communication preventing repeat errors

Her time per return dropped from 3.2 hours to 2.1 hours. Client calls during tax season dropped 35% because proactive reminders caught issues early. The number of amended returns filed dropped from 12 per season to 3.

The AI doesn't replace her tax expertise. It remembers client situations so she can focus on judgment calls and planning instead of searching through prior year files.

Setup Front-Loads the Work, Then Saves 30+ Hours a Season

Building the memory file takes time. You document:

• Client tax situations and filing patterns
• Recurring deductions and typical amounts
• Long-term tracking items (basis, depreciation, carryovers)
• IRS correspondence and resolution history
• Client-specific behavioral notes

You build it during slow season or while preparing returns. After that, every client conversation starts with full context. No file searching. No re-explaining situations you handled last year.