Understanding AI in the Tax Context

What AI Actually Is (and Is Not)

If you have used ChatGPT, Claude, or Microsoft Copilot in the past year, you have interacted with a large language model. If you have not, some of your colleagues have, and some of your clients definitely have. Either way, AI is now part of the tax practice landscape, and understanding what it can and cannot do is no longer optional.

A large language model (LLM) is a software system trained on vast quantities of text. It learns patterns in language: how sentences are structured, how arguments are built, how technical documents are organized. When you ask it a question, it generates a response by predicting the most likely sequence of words given your prompt. This is a fundamentally different process from looking up an answer in a database.

This distinction matters enormously for tax practitioners. When you search for IRC Section 162 in CCH AnswerConnect, the system retrieves the actual statutory text. When you ask an LLM about Section 162, it generates text that looks like an authoritative answer based on patterns it learned during training. Sometimes that generated text is accurate. Sometimes it is not. The LLM does not know the difference.

How LLMs Differ from Traditional Tax Software

Traditional tax research platforms like Thomson Reuters Checkpoint, CCH AnswerConnect, and Bloomberg Tax Law are retrieval systems. They store actual source documents -- the Internal Revenue Code, Treasury Regulations, Revenue Rulings, court opinions -- and let you search them. When Checkpoint shows you Reg. Section 1.162-5, you are reading the actual regulation.

Tax preparation software like Lacerte, UltraTax, and ProSeries performs calculations. It applies tax rules algorithmically. If you enter a $10,000 charitable contribution for a taxpayer with $200,000 AGI, it applies the correct percentage limitation because a programmer coded that rule explicitly.

AI tools do neither of these things inherently. An LLM does not store the tax code, and it does not calculate taxes. What it does is process natural language with remarkable fluency. You can describe a complex fact pattern in plain English, and it will generate an analysis that reads like it was written by a knowledgeable practitioner. That fluency is both its greatest strength and its most dangerous quality.

The Current State of AI in Professional Tax Practice

As of 2025-2026, AI is appearing in tax practice through several channels:

Embedded AI features. Thomson Reuters has integrated AI into its research platforms. Intuit has added AI-assisted categorization and anomaly detection to ProConnect Tax. These are relatively controlled implementations where the AI operates within guardrails set by the software vendor.

General-purpose AI tools. Many practitioners are using ChatGPT, Claude, Google Gemini, or Microsoft Copilot for research drafts, client letter writing, and brainstorming. These tools have no tax-specific guardrails unless the practitioner provides them.

Specialized legal and tax AI. CaseText (now part of Thomson Reuters) and similar platforms use AI specifically trained on legal and regulatory text. These tend to be more reliable for citations but still require verification.

Productivity tools. Microsoft Copilot integrated into Excel and Word helps with data manipulation, document drafting, and formatting -- tasks adjacent to tax work but not tax-specific.

The AICPA has published guidance acknowledging that AI will be part of practice going forward, while emphasizing that professional standards -- including competence, due diligence, and supervisory responsibilities -- apply fully to AI-assisted work. The question is no longer whether practitioners will use AI. The question is whether they will use it competently.

The Job Replacement Question

Let us address the elephant in the room. Many practitioners worry that AI will replace them. The short answer: it will not -- but it will change what you do.

AI cannot sign a return. AI cannot represent a client before the IRS. AI cannot exercise the professional judgment required to assess whether a position has substantial authority. AI cannot build the trust relationship that keeps clients coming back year after year. These are the core of your practice, and they are not automatable.

What AI will change is the time you spend on tasks that are not the core of your practice: initial research scoping, data entry, form letter drafting, and routine review tasks. Practitioners who learn to delegate these tasks to AI -- while maintaining rigorous oversight -- will be able to serve more clients, take on more complex engagements, or simply reclaim time during busy season.

The real risk is not that AI replaces competent practitioners. The risk is that practitioners who do not understand AI tools either avoid them entirely (and fall behind on efficiency) or adopt them uncritically (and create liability exposure). This course is designed to put you in neither camp.

AI-Assisted Research vs. AI-Generated Advice

There is a critical distinction between using AI as a research assistant and treating AI output as professional advice. Consider two scenarios:

Scenario A: A practitioner asks Claude, "What are the requirements for deducting home office expenses under Section 280A?" The AI provides a well-organized summary. The practitioner then verifies every element against the actual statutory text and relevant regulations before incorporating any of it into client advice. This is AI-assisted research.

Scenario B: A practitioner copies a client's question into ChatGPT, copies the response, and sends it to the client on firm letterhead. This is AI-generated advice, and it is a professional responsibility disaster waiting to happen.

The difference is not the tool. The difference is the practitioner's process. AI-assisted research enhances your workflow. AI-generated advice replaces your judgment. One is a productivity tool. The other is a liability.