Can I Use AI to Defend Myself in an IRS, EDD, or Sales Tax Audit?
Author
John Milikowsky, Esq. | Founder | John Milikowsky represents U.S. and foreign businesses and individuals in sophisticated business transactions involving U.S. tax matters. Relentlessly defending each client in federal and state audits and criminal investigations to protect their civil rights and provide financial security.
A business owner gets an audit notice from the EDD, CDTFA, or IRS, opens a laptop that same afternoon, and starts typing questions into ChatGPT. What does this letter mean? What records do they want? What should the response say? Within an hour, there’s a drafted letter, a list of documents, and a plan.
It feels productive. In a lot of ways, it is. AI is genuinely useful for understanding a notice, organizing a response, and putting language on paper faster than most people can do it alone.
What it cannot do is sit across from an auditor, read the room, and decide in real time what to share, what to hold back, and when. That part of an audit has always been about people, and it still is.
What AI Can Actually Do in a Tax Audit
AI tools are strong at a specific set of tasks. They can summarize a dense notice into plain language. They can help organize receipts, payroll records, or sales data into something more presentable. They can draft a first pass at a response letter, clean up grammar, and suggest a structure for a written explanation.
For a business owner who is overwhelmed and unsure where to start, that’s real value. It lowers the anxiety of staring at a government letter with no idea what to do next.
The limits show up quickly once the audit moves past paperwork and into conversation. An audit is not a form to fill out once and submit. It’s a back-and-forth that plays out over weeks or months, often with the same auditor, and the direction it takes depends heavily on how that relationship develops.
Why Audits Are Personal, Not Just Procedural
The Auditor Is a Person With Discretion
Every audit has a process behind it. There are statutes, procedures, and documentation standards each agency follows. But inside that process, individual auditors make constant judgment calls. They decide how closely to scrutinize a category of expenses. They decide whether an inconsistency looks like an honest mistake or a pattern worth digging into. They decide whether to expand the scope of the audit or keep it narrow.
We often see two audits with nearly identical facts land in very different places, and the difference usually traces back to how the case was handled with the person conducting it, not just what the numbers showed.
How Rapport Affects Audit Outcomes
Auditors are not adversaries by default. Most of them approach a case wanting a clean, defensible file they can close without controversy. When a business owner or their representative communicates clearly, responds on time, and treats the auditor as a professional doing a job, the tone of the audit tends to stay calm and cooperative.
When communication feels evasive, disorganized, or combative, even unintentionally, auditors tend to dig deeper. A generic AI-drafted letter, however well-written, cannot read an auditor’s tone in an email, sense when a phone call would land better than a written response, or adjust an approach because a particular auditor has a reputation for being thorough on a specific issue. That kind of read comes from having sat across from hundreds of auditors, not from a prompt.
The Real Skill: Knowing What to Send and When
Why Sending Everything Backfires
One of the most common mistakes we see is business owners who want to be maximally cooperative and end up sending far more than the auditor asked for. A single spreadsheet meant to answer one question can open the door to five new ones if it includes columns or line items outside the original scope.
An EDD audit letter or a CDTFA information request usually defines a specific scope. Staying inside that scope, and knowing which documents actually answer the question being asked, is a skill built on experience with how these audits typically expand. AI has no way to know which document in a client’s files is going to raise a flag that was never part of the original inquiry.
Timing Shapes the Scope of an Audit
Timing matters just as much as content. Responding too quickly, before records have been fully reviewed internally, can lock a business into an explanation that doesn’t hold up once more information surfaces. Responding too slowly invites penalties or a default assessment based on the auditor’s own estimates.
There’s also a sequencing question that comes up constantly. Sometimes it makes sense to request a short extension before sending anything. Sometimes a phone call to clarify the request saves weeks of back-and-forth over documents that were never going to be relevant. These are decisions that depend on where the audit is in its lifecycle and how it has gone so far, not something a chatbot can calculate from a single prompt.
Auditor Personalities and How They Shape a Case
Business owners are often surprised to hear that auditor personality plays a real role in how a case unfolds. It does, and recognizing the type of auditor on a file early changes the strategy.
The By-the-Book Auditor
Some auditors work strictly within a checklist. They want specific documents, in a specific format, and they’re not inclined to accept explanations that fall outside their standard framework. With this type, precision matters more than persuasion. Responses need to map directly to what’s being requested, formatted the way the auditor expects to receive it.
The Auditor Looking to Close the File
Other auditors are managing a heavy caseload and are motivated to reach a defensible resolution efficiently. With this type, a well-organized, complete response early in the process can move a case toward closure faster. Dragging out the process or submitting incomplete information tends to frustrate this type of auditor and can turn a routine case into a longer one.
The Auditor Building a Broader Case
Occasionally, an audit reveals something that shifts an auditor’s posture, and they start looking beyond the original scope. This is where discernment matters most. Recognizing the shift, understanding what triggered it, and adjusting the response strategy accordingly is often the difference between a contained issue and a multi-year expansion. This is a real-time read that develops through direct experience with how CDTFA audits typically expand once a discrepancy surfaces, and it’s not something a general-purpose AI tool is positioned to catch.
Where AI Falls Short in an Actual Audit
AI tools don’t have access to the auditor’s history, the agency’s current enforcement priorities, or the specific patterns that tend to trigger a deeper look in a given industry. They also don’t carry attorney-client privilege, which means anything typed into a public AI tool while preparing an audit response isn’t protected the way a conversation with an attorney would be.
There’s also a practical risk in relying on AI-generated explanations for complex tax positions. A chatbot can produce language that sounds confident and well-reasoned without it being accurate for a specific business’s facts. An auditor who has seen thousands of cases can often tell when a written explanation doesn’t match the underlying documentation, and that mismatch tends to invite more questions rather than fewer.
How Experienced Representation Uses Judgment AI Cannot Replicate
Every audit follows a general path, whether it’s an IRS examination, an EDD worker classification review, or a CDTFA sales tax audit. Understanding that path in advance, including where audits tend to slow down or accelerate, shapes how a response gets built from the first letter forward. The step-by-step lifecycle of an EDD audit looks fairly consistent on paper, but the way it plays out depends heavily on the specific auditor, the industry, and how the business responds along the way.
Attorneys who handle these cases regularly build a working relationship with the auditor over the life of the case. That relationship, paired with a clear read on what documentation actually needs to go out and when, is what keeps a routine audit routine. AI can support the drafting and organizing that happens behind the scenes. The judgment calls that determine how the audit actually goes still belong to the person managing the relationship with the agency.
Business owners who want the efficiency AI offers without losing the strategic judgment an audit requires tend to get the best results by using both together, with an experienced attorney directing how AI-assisted drafts get used rather than sending them straight to an auditor.
Frequently Asked Questions
Can I use ChatGPT to respond to an IRS, EDD, or CDTFA audit notice?
AI tools can help draft language and organize documents, but a response should be reviewed by someone familiar with how that specific agency evaluates audit responses before it gets sent. An AI-drafted response that misreads the scope of the request or includes unnecessary detail can extend the audit rather than resolve it.
Will an auditor know if I used AI to write my response?
Auditors are generally focused on whether the content is accurate and responsive to their request, not on how it was drafted. The bigger risk isn’t detection. It’s submitting a well-written response that doesn’t actually address what the auditor needs, or that opens up new questions the business wasn’t prepared to answer.
Can AI negotiate with an EDD or CDTFA auditor on my behalf?
No. AI tools cannot communicate directly with an agency, cannot read an auditor’s tone or intent, and cannot make real-time decisions about what to disclose during a call or meeting. Negotiation and scope management require a person who understands both the law and the specific auditor’s approach.
What should I never send an auditor without review first?
Full financial statements, complete bank records, or broad data exports should generally be reviewed and narrowed before submission. Sending more than what was requested is one of the most common ways an audit expands beyond its original scope.
Is it risky to use AI tools to prepare tax audit documents?
The main risk is confidentiality. Information entered into a public AI tool isn’t protected by attorney-client privilege the way it would be in a conversation with a tax attorney. Business owners should be cautious about entering sensitive financial or legal details into general AI platforms while an audit is active.
An audit rarely goes wrong because of one document or one missed deadline. It goes wrong because of a string of small decisions about what to share, how to say it, and when to say it, made without a clear read on the person evaluating the case. That read comes from experience, and it’s the part of audit defense that technology hasn’t replaced.
