When Your NoteTaker Rats You Out

What Every CPA Needs to Know About AI, Documentation, Discovery, and Privilege

By: Megan J. Whitlock, CPA

For decades, we taught clients and internal staff one lesson above all the others.

Document. Document. Document.

If it isn’t documented, it didn’t happen.

I’ve spent more than 20 years practicing public accounting including time in national tax offices and extensive advisory work — partnership transactions, M&A, valuations, tax controversies and partner disputes. The kind of work where millions of dollars can turn on a single fact or circumstance. In nearly every one of those situations, documentation was the answer.

Need to establish business purpose? Document it. Support a valuation conclusion? Document it. Defend a partnership allocation? Document it. Survive an IRS examination? Document it.

Then artificial intelligence walked into the room and changed the discussion. Today the technology records every meeting, transcribes every word, summarizes every discussion, drafts the action items, and files all of it in a searchable database. Whether AI can do that was never the interesting question. The interesting question is whether it should. For the first time in my career, I’m asking something new:

What happens when your note taker rats you out?

The Documentation Paradox

Tax lives in a world of facts and circumstances. The IRS evaluates facts, circumstances, and intent. The Tax Court evaluates facts, circumstances, and intent. Valuation professionals evaluate facts and circumstances. Reasonable compensation, economic substance, business purpose, disguised sales, research credits, partnership and debt allocations — they all turn on the story the documentation tells.  I recently taught a course on disguised sales in partnership transactions, an area that weighs heavily on the reliance of a good fact pattern to support a positive taxpayer outcome and support transaction intent. For the first time I asked participants to consider the specific client and whether they should include their AI notetaker in these discussions. 

For most of my career, the fear was having too little of that documentation. Today, a growing number of firms have the opposite problem.

They have too much.

That sounds strange coming from a CPA who has spent 20 years preaching the gospel of the workpaper, but AI has quietly changed what documentation even is. For decades, we documented conclusions through written notes from a phone call or email thread. Today, we are preserving conversations. Those are not the same thing — and the difference is where the risk lives.

Traditional Notes vs. AI Transcripts

One of the biggest misconceptions I see is that an AI meeting transcript is just a modern set of professional notes. It isn’t. Traditional notes are an act of judgment: the accountant listens, weighs what matters, discards what doesn’t, and records the facts that support the engagement. An AI transcript does none of that. It captures everything.

Traditional Professional Notes	AI Transcript
Prepared using professional judgment	Captured automatically
Focus on the relevant facts	Capture every statement
Document conclusions and rationale	Preserve the entire discussion
Reflect what matters	Reflect everything
Support a position	Capture a process

The distinction matters. Picture a planning discussion on a complex partnership transaction. The team may talk through five different approaches before landing on the right recommendation. A traditional memo records the facts, the analysis, and the conclusion. An AI transcript records every question, every concern, every rejected idea, every half-formed thought, and every comment made before all the facts were on the table.

The issue isn’t whether the transcript is accurate. The issue is whether a verbatim recording reflects the same professional judgment we would ordinarily capture in a well-prepared workpaper. 

Documentation vs. Preservation

Our profession was built around documentation. AI is built around preservation. Documentation requires judgment. Preservation just keeps everything. That single distinction may turn out to be one of the most important risk-management issues facing CPA firms over the next decade — because what AI preserves includes:

• Preliminary assumptions later proven wrong

• Hypothetical planning ideas both internally and with clients

• Rejected alternatives

• Informal, off-the-cuff observations

• Speculation, humor, and frustration

• Statements made before all the facts were known

None of these things are inherently a problem. But every one of them can become evidence. And evidence has a way of taking on a life of its own years later.

Can a Court Demand Your AI Notes?

A lot of professionals assume AI-generated transcripts carry some special protection. They don’t. From a discovery standpoint, an AI transcript may be just another business record. If it’s relevant and not otherwise shielded by privilege or the work-product doctrine, it can be discoverable. The question is not whether AI created the document. The question is whether the document exists.

Assume that transcripts, summaries, recordings, prompts, and everything around them may eventually be fair game for examination, discovery requests, subpoenas, or litigation.

When AI Becomes Part of the Evidence

A recent federal decision should be a wake-up call for anyone using AI in their workflow. In Conservation Law Foundation, Inc. v. Shell Oil Co.,^[1] the court ordered production of AI-related chat logs and materials used by an expert witness. The party resisting production argued the materials weren’t discoverable. The court disagreed. The reasoning was simple. If the AI tool was part of the expert’s methodology, opposing counsel was entitled to understand how the analysis was actually performed.

That is a meaningful shift. We have always focused on the final report. Shell suggests courts may increasingly look at the process behind the report. What prompts were entered? What searches were run? What information was considered — and what was left out? Those questions matter now. The digital breadcrumbs have become part of the evidentiary trail.

What This Means for Tax Professionals

Picture a disguised-sale analysis under IRC Section 707. The planning team works through several alternatives before reaching a defensible conclusion. The final memo is technically sound. The transaction is properly documented. The position is supportable.

Now picture an AI transcript that captured comments like:

“We need to get cash out without triggering gain.”

“How do we structure this so it doesn’t look like a sale?”

Maybe that was brainstorming. Maybe those ideas were rejected. Maybe they were completely irrelevant to the final recommendation. It won’t matter. Years later, they become part of the story — and once something is part of the story, someone has to stand up and explain it.

The Privilege Problem Nobody Is Talking About

Discovery is only half the issue. The bigger one may be confidentiality. One of the most significant AI decisions this year came in United States v. Heppner,^[2] where the court reviewed materials generated through Anthropic’s Claude platform and concluded that the communications were not protected by attorney-client privilege.

Why? Because the information had been shared with a third-party AI platform rather than communicated directly between attorney and client in a way that preserved the privilege.

Whether Heppner becomes the prevailing view remains to be seen. But it raises a question every CPA, attorney, and client should be asking: what happens when confidential information is typed into an AI system? For years we asked whether information was confidential. Now we also have to ask:

• Who owns the AI subscription?

• Is the platform a public AI tool or a private enterprise environment?

• Does the provider retain prompts, transcripts, or uploaded documents?

• Is the information used to train future models?

• Who controls access to the data?

• What happens if the subscription is terminated?

• Does the law firm, CPA firm, or client own the information and outputs?

• Can the information be exported, searched, or subpoenaed?

• What contractual protections exist between the firm and the AI provider?

  
  

Kovel in the AI Era

For accountants working under Kovel arrangements, the stakes climb higher. For decades we’ve served as an extension of legal counsel in controversy matters, helping attorneys make sense of complex financial and tax issues. The privilege framework in those engagements is usually built with care. AI introduces brand-new uncertainty.

If privileged communications are uploaded into an external AI platform, does the privilege survive? Heppner hints at the answer: the court suggested that had counsel directed the defendant to use the tool, the AI “might arguably” have functioned as a lawyer’s agent under the Kovel doctrine — which means attorney-directed use may rest on very different ground than a client going it alone. The law is still developing. What’s already clear is that firms can no longer assume information entered into an AI platform receives the same treatment as information kept inside traditional attorney-client channels.

Five Questions Every Firm Should Be Asking

Before you let the software record every meeting by default, answer these:

1.        What exactly is being retained — video, audio, transcripts, summaries, prompts, metadata?

2.        Where is it stored — internal servers, third-party vendors, or a public cloud?

3.        How long is it kept — thirty days, a year, or forever?

4.        Who has access — the engagement team, firm administrators, or the software vendor?

5.        Should this meeting be recorded at all?

That last one may be the most important question of all.

When Should the Note Taker Be Turned Off?

Let me be clear: I am not anti-AI. Quite the opposite. Used well, this technology makes advisors more effective, more efficient, and more valuable than ever. But every technology brings new risk, and our job as professionals is to understand that risk before it becomes a problem. There are conversations where automatic recording deserves real scrutiny:

• Tax controversy discussions

• Attorney-directed and Kovel engagements

• Litigation support and valuation disputes

• Sensitive negotiations

• Internal personnel matters

• Strategy sessions involving privileged communications

In a lot of those situations, a carefully prepared memorandum — facts, assumptions, analysis, conclusions — is better risk management than a verbatim recording of every word spoken in the room.

The WhitTalks Takeaway

For over 20 years I’ve told clients that documentation matters. I still believe it. What has changed is our understanding of what documentation is.

The profession spent decades worrying about whether we documented enough. The AI era forces a different question: have we preserved too much? The next IRS examination, malpractice claim, controversy, or discovery request may not stop at the memo, the workpaper, or the expert report. It may ask for the transcript. The recording. The prompt history. The chat log. The whole digital trail that led to the conclusion.

In a world where AI remembers everything, professional judgment may no longer be about deciding what to document. It may be about deciding what should never have been recorded in the first place.

^[1] Conservation Law Found., Inc. v. Shell Oil Co., No. 3:21-cv-00933 (VDO) (D. Conn. May 18, 2026) (order granting motion to compel, ECF No. 970). The order has been stayed pending the district court’s review of the plaintiff’s objection.

^[2] United States v. Heppner, No. 25 Cr. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026) (Rakoff, J.).