Tax Research Fact Sheet, TEMPLATE
What this solves. Tax research is only valuable if it fits your client, but the client's details are exactly what's confidential. This sheet is the bridge: it captures the facts the tax law actually turns on (so the research is applicable) in abstracted form (so no PII/TRI is exposed). Fill it out, then hand the shareable part to your AI tool. Keep the local key on your side only. See Module 2, Tax Research. Not tax advice.
Guiding principle: Tax outcomes turn on structure, relationships, magnitudes, timing, and elections, not identity. Identity is not a tax input, so removing it costs zero applicability.
Part A, LOCAL KEY (keep private; never paste into any AI)
| Token | Real identity | Notes |
|---|---|---|
| Taxpayer A | [real name, KEEP LOCAL] |
|
| Entity X | [real entity / EIN, KEEP LOCAL] |
|
| Property P | [real address/parcel, KEEP LOCAL] |
Store this with the client file, not in the research prompt. You'll re-attach identities locally when you write up the memo.
Part B, SHAREABLE FACT PATTERN (this is what goes to the AI, no identifiers)
1. Issue / question
[The precise tax question. State it before facts so you only gather what's determinative.]
2. Parties (tokenized), type & relationships
[Taxpayer A: individual / MFJ; resident of [STATE]][Entity X: S-corporation; Taxpayer A owns [__]%][Relationships that matter: related party? spouse? >50% owner? controlled group?]
3. Jurisdiction
- Federal:
[yes]· State(s):[STATE, not address]· Other:[multistate / foreign?]
4. Tax year(s) & timing
- Year(s) at issue:
[Year 1 / 20XX] - Key dates & sequence:
[acquisition, election, sale, distribution, order matters] - Holding period / method of accounting:
[___]
5. The transaction / event (abstracted)
[What happened, described by type, "Entity X distributed appreciated real property to Taxpayer A,"
not "ABC LLC deeded 123 Main St to John."]
6. Amounts, exact only if determinative
-
[Use magnitude / threshold-relative figures for legal research: "income ~$50k OVER the §199A threshold," "basis materially less than FMV." Reserve exact $ for Layer 2 in a tool you control.] -
Exact figure genuinely required for the legal question?
[ ] no [ ] yes → why: ___
7. Character / classification
[Capital vs. ordinary, §1231, passive vs. active, QBI eligibility, etc.]
8. Elections, prior positions, methods
[Elections made/available (e.g., §754, §83(b), QSub), prior-year treatment, positions taken.]
9. What's driving the question / desired outcome
[e.g., "minimize current-year gain," "confirm deductibility," "determine filing obligation."]
Part C, SCRUB CHECK (confirm before sending Part B)
- No names (people, businesses, properties), tokens only
- No SSNs or EINs
- No street addresses, account/parcel/policy numbers, DOB, phone, email
- State retained only where it affects the answer
- Exact dollar figures removed unless determinative (then Layer 2, controlled tool)
- If any real identifier must remain, I am using a firm-approved tool and have considered IRC §7216 / AICPA Confidentiality Rule obligations (see Regulatory Foundation)
Part D, completeness check (so the research is actually applicable)
- Every fact the issue turns on is captured (re-read Part B against the issue in #1)
- Relationships and ownership %s are stated (these usually drive the answer)
- Timing/sequence is clear where the law is order-sensitive
- Elections and prior positions are noted
- If unsure which facts are determinative, ask the AI to list the facts it needs for this issue first, then fill them in here (abstracted)
Part E, re-attach (the round trip)
Abstraction is a loop, and it closes on your machine. When the research is done, the AI's answer
will refer to your tokens (Taxpayer A, [Person_1], [ADDRESS_1]…). Bring it back to a usable,
client-specific work product:
-
[ ] Swap every token back to the real identity using Part A (the local key), the Redactor tool's "Bring the answer back" step does this automatically (paste the AI's answer + your key), or find-and-replace by hand.
-
[ ] Treat the re-attached document as confidential again, it now holds real client info, so it should not go back into a general AI tool.
-
[ ] Carry the verified conclusion into the Research Memo with identities restored.
The whole loop: real facts → (Part C scrub) → abstracted facts + local key → research with AI → tokenized answer → (Part E re-attach) → client-specific work product. The client's identity never leaves your desk; the local key is the only bridge.
Template maintained with the AI Lab for Accountants library. Pairs with the
tax-research skill and the
Tax Research Memo template. Not tax/legal advice.