Why Unsourced Answers Are a Liability in Tax Work
Questions arrive faster than senior staff can research them, consumer chatbots invent citations, and client facts pasted into outside tools can breach Section 7216 and confidentiality duties.
Research Persona: Tax Practice Leader
Autonomy: Assist · System drafts, human drives
Tax Research
AI for tax research searches the Internal Revenue Code, Treasury regulations, IRS rulings, court cases and the firm's own memos, then drafts a research memo that states the facts and assumptions, cites and quotes each authority it relies on and marks where the law is unsettled. A CPA or tax attorney checks every citation, tests the reasoning and signs the advice. The agent advises no client and files nothing.
At a glance
Trigger: A client or internal stakeholder asks a federal tax question that needs a documented answer. Owner: Tax Practice Leader. Primary output: Draft research memo citing and quoting each authority relied on. Consequential actions require approval.
Assess your workflow Professional ServicesFinancial ServicesEnterprise
Verifiable Citations, Professional Sign-Off
VDF AI agents research inside the firm's environment over a curated library of primary authority and past memos, draft memos with checkable citations and leave every conclusion to the reviewer.
Agent Workflow
How the Agent Network Works
- 01
Facts Agent
Restates the client facts and assumptions and lists missing facts that could change the answer.
- 02
Authority Agent
Retrieves Code sections, regulations, rulings, cases and prior firm memos on the issue.
- 03
Citation Agent
Confirms each authority exists, quotes the passage relied on and flags superseded items.
- 04
Drafting Agent
Writes the memo in the firm's format, separating settled points from open questions.
- 05
Review Agent
Routes the draft to the assigned reviewer and records edits and sign-off.
Data and evidence
What Tax Research Needs to Operate
Each tax research source has a defined purpose, freshness expectation, quality gate, and sensitivity boundary.
Code, Treasury regulations, IRS guidance and court decisions
Purpose: Ground every proposition in authority.
Freshness: Updated as new law and guidance are published.
Quality: Each document carries its citation, effective date and status.
Sensitivity: Public, with licensed content used within licence terms.
Firm research memos and knowledge library
Purpose: Reuse the firm's signed positions on recurring issues.
Freshness: Added when a memo is signed off.
Quality: Memo date, reviewer and law-as-of date recorded.
Sensitivity: Confidential work product with matter-level access.
Client facts from the engagement file
Purpose: Frame the question and its assumptions.
Freshness: Captured per research request.
Quality: Facts confirmed by the engagement team.
Sensitivity: Tax return information under Section 7216.
Measurement plan
How to Evaluate Tax Research
Primary measure: Reviewer hours from question to signed research memo. Record hours and review corrections on twenty recent memos across common issue types.
Illustrative model Value hypothesis and full cost
Illustrative: memos per year × hours saved × loaded cost, minus library curation, review, model and infrastructure costs.
Cost inputs to include
- Curating and updating the authority library
- Reviewer time per memo
- Licences for research content used in retrieval
Validation Supporting measures and review cadence
Spot-check citations in a monthly memo sample, and re-run a fixed question set after model or library changes.
- Citations corrected or removed at review
- Memos returned for missing facts or authorities
Decision guide
Tax Research: Operating Model and Implementation
When Tax Research is appropriate
Tax research suits retrieval because the authority is finite and citable. Treasury’s list in 26 CFR 1.6662-4(d)(3)(iii) names what counts toward substantial authority, from the Code, regulations and revenue rulings to cases, committee reports, the Blue Book and private letter rulings. It excludes conclusions in treatises, periodicals and professional opinions, so an AI answer is worth only the sources it points to.
Use it for recurring federal questions, not novel transactions with little authority to retrieve.
Designing the operating workflow
- Frame. Facts, assumptions and gaps are written down first.
- Retrieve. Primary authority and earlier firm memos are gathered.
- Verify. Every citation is matched to a library document and quoted.
- Draft. Settled and open points are kept apart.
- Review and sign. The written-advice standard in 31 CFR 10.37 applies to the professional who signs.
Data, integration, and evidence
Under the Section 7216 regulations, a preparer may share return information with another US preparer without consent only when the recipient makes no substantive determination, meaning analysis, interpretation or application of the law. Keep client facts and the authority library in separate stores, so most research runs with no client data at all.
How VDF.AI supports this use case
VDF AI Agents run on the firm’s own servers, so client facts used in a question are not disclosed to a model vendor. The legal research agent shows the pattern: jurisdiction and date fixed first, every proposition cited. The citation verifier checks that each quote supports its claim, and the VDF AI Router can keep these questions on local models.
The guide to AI for accounting firms covers Section 7216 consent routes. Related workflows include legal research with cited authorities and the wider finance use cases.
Risk and control register
Controls Required for Tax Research
A fabricated or superseded citation reaches a client memo.
Control: Each citation links to a library document, and the reviewer opens every one before signing.
Accountable owner: Reviewing CPA or tax attorney
Return information is disclosed outside Section 7216 rules.
Control: Models run on firm infrastructure, and client facts are limited to what the question needs.
Accountable owner: Firm risk partner
The law changes and an old position is reused.
Control: Each memo records its law-as-of date, and reuse prompts a check for later guidance.
Accountable owner: Tax knowledge manager
Where this workflow should not operate
- It signs no advice, takes no return position and files nothing.
- Answers are only as current as the curated library.
- State and foreign tax questions need their own sources.
Controlled rollout
Pilot and Scale Criteria
Federal income tax questions in one practice group, with reviewers checking every citation and staff researching in parallel.
Prerequisites
- Curate the authority library and the firm's signed memos with dates and status.
- Agree the memo template and Section 7216 approach with risk and counsel.
Approval gates
- The practice leader approves the memo template and review rules.
- Risk management approves the handling of tax return information.
Scale criteria
- No unverified citation reaches a signed memo during the pilot.
- Reviewer time per memo falls against the baseline.
Tooling
VDF AI Tools That Power This Use Case
Assign these prebuilt tools to the bounded agents in Tax Research, or browse all VDF AI tools.
Related Use Cases
Explore Adjacent Workflows
Evidence
Authoritative Sources and Implementation References
These sources inform the governance and evaluation approach for Tax Research. They do not certify a specific deployment.
- 26 CFR 1.6662-4: Substantial understatement of income tax — Legal Information Institute, Cornell Law School
- 31 CFR 10.37: Requirements for written advice (Circular 230) — Legal Information Institute, Cornell Law School
- 26 CFR 301.7216-2: Permissible disclosures or uses without consent of the taxpayer — Legal Information Institute, Cornell Law School
Written by VDF AI Editorial Team. Last reviewed 23 October 2026.
FAQ
Frequently Asked Questions
Answers for CPA firm and corporate tax professionals evaluating this workflow's data, controls, measures, and operating boundaries.
Talk to an expert01 Can an AI research memo support a return position?
Not on its own. Substantial authority is weighed using sources Treasury lists, such as the Code, regulations, rulings, cases and committee reports, and the regulation says professional opinions and treatises are not authority. An AI memo is worth only the authorities it points to, which a professional must read.
02 Does Circular 230 apply to AI-assisted tax advice?
Circular 230 does not mention AI, but section 10.37 binds whoever gives the written advice: reasonable assumptions, reasonable efforts to find the facts, law related to those facts and no reliance on the chance of escaping audit. Adopting an unchecked draft does not shift that duty.
03 Can client data go into AI tax research under Section 7216?
Research on public law needs no client data. Once return information enters a prompt, sharing it with an outside party that analyses or applies tax law generally needs the taxpayer's consent. Running models on firm servers avoids that disclosure; confirm the route with counsel.
Build This Use Case with VDF AI
Start building it free in the cloud, or describe your Tax Research workflow and we will help map the appropriate governed agent network for your environment.