Blog
ai for tax researchai tax research toolchatgpt for tax questions

How to Use AI for Tax Research (and Check Its Work)

Use AI for tax research without trusting wrong answers. Steps, copy-ready prompts and a verification checklist for finance, accounting and tax teams.

Portrait of Luis Chavez-Mattos
Edited by Luis Chavez-Mattos, Director, Product··RSS

For Finance, Accounting and Managers

Aerial view of a tiny woman in a beanie and puffer jacket beside a giant open laptop showing empty chat bubbles, on a vast empty flat surface with one long thin line leading from her feet to the frame edge.

How do you use AI for tax research?

Use AI to frame the question, find the likely rules, and draft a summary. Then verify every conclusion against primary sources (the statute, regulations, rulings, case law, or the tax authority's own guidance) before anyone relies on it. AI is fast at orienting you and weak at being right on the details: dates, thresholds, state-by-state differences, and whether a rule still applies. Treat its output as a research lead from a fast but overconfident junior colleague, never as the answer.

The rest of this post gives you the workflow, the prompts to paste, and a checklist for catching wrong answers.

TL;DR

  • AI is a starting point for tax research, not a source. It can tell you where to look, but the authority you cite must be something you opened and read yourself.
  • Ask for the question to be broken down first. A good prompt makes the AI state its assumptions (jurisdiction, tax year, entity type) so you can correct them before it answers.
  • Require citations, then check every one. AI tools can invent code sections, rulings and court cases that look real. A citation you have not opened is not a citation.
  • Check the date and the jurisdiction. Tax law changes yearly and differs by state and country. These are the two most common ways an AI answer is quietly wrong.
  • Keep confidential data out of general chat tools. Strip names, tax IDs and client details, and follow your company's AI policy.
  • Log what you asked, what AI said, and what you verified. A short record protects you in an audit and makes review fast for your manager.

What is AI actually good and bad at in tax research?

Tax research has a structure: identify the issue, find the authority, read it, apply it to the facts, document the conclusion. AI helps with some of those steps and is dangerous at others.

TaskUse AI?Why
Turning a vague business question into a precise research questionYesIt is good at structuring and asking clarifying questions
Listing the areas of law that might applyYes, as a leadGives you search terms and code sections to look up
Explaining a dense rule in plain languageYes, after you have read the ruleEasy to check against the text
Finding the actual authorityCarefullyIt may name real sources or invent plausible ones
Stating current rates, thresholds, deadlinesNoOften outdated or mixed up across years and jurisdictions
State, local and foreign rulesCarefullyCoverage is thinner and errors are more common
Reaching the final conclusionNoA qualified person owns the judgment
Drafting the memo from your verified notesYesWriting is where it saves the most time

The pattern: AI is strongest at language work (structuring, explaining, drafting) and weakest at being a reliable lookup of facts that change.

How do you do tax research with AI, step by step?

1. Write down the facts and the question before you open any tool

Tax answers depend on facts. Write a short fact pattern: who the taxpayer is (entity type, not the name), where they operate, which tax year, what happened, and what you need to decide. If you cannot write this in five or six sentences, the AI will fill the gaps with guesses.

2. Remove confidential information

Replace company names, employee names, taxpayer IDs, account numbers and unique deal details with generic labels ("Company A," "an employee in State X"). Check your company's AI policy first. Many large companies allow only approved tools for anything touching financial or tax data. If you are unsure, ask before pasting.

3. Have AI restate the issue and list its assumptions

Do not ask for the answer yet. Ask it to identify the issues and say what it is assuming. This is where you catch the wrong tax year or the wrong jurisdiction before they poison the whole answer.

4. Ask for leads, not conclusions

Request the areas of law, the likely code sections or regulations, and the search terms to use in a research database. Your goal is a reading list.

5. Open the primary sources yourself

Go to the statute, the regulations, the tax authority's website, or your firm's research platform. Read the actual text. Confirm it exists, says what the AI claimed, and applies to your tax year.

6. Have AI help you apply the rule to your facts

Now paste in the exact text you verified and ask AI to apply it. This is much safer than asking from memory, because the AI is working from text you have confirmed.

7. Run a second-pass check

Ask AI to argue the opposite position and to list exceptions, limitations and recent changes. Then verify any new authority it names.

8. Document and get review

Record the question, sources, conclusion and open uncertainties. A qualified reviewer (your tax director, an outside advisor) signs off on anything that drives a filing or a material decision.

What prompts should you paste?

Prompt 1: Frame the question

I am doing tax research. I will give you a fact pattern with identifying 
details removed. Do not answer yet.

Facts: [2 to 6 sentences: entity type, jurisdiction, tax year, what happened]
Question: [what I need to decide]

1. Restate the tax issue(s) in one or two sentences each.
2. List every assumption you are making (jurisdiction, tax year, entity type, 
   residency, anything else).
3. List any facts you would need that I have not given.
4. Tell me which of these you are unsure about.

Correct its assumptions, then move on.

Prompt 2: Get a reading list

Based on the corrected facts below, give me a research plan, not a conclusion.

Facts and corrections: [paste]

For each issue, list:
- the type of authority I should check (statute, regulation, ruling, 
  case law, state guidance)
- the specific section or document you believe applies, labeled 
  "confident" or "unsure"
- search terms I can use in a tax research database
- how the rule might differ by tax year or jurisdiction

Do not invent citations. If you are not certain a citation exists, 
say so and describe the topic instead.

The last instruction does not eliminate fabricated citations, but it gives you honest "unsure" labels to prioritize your checking.

Prompt 3: Apply verified text

Below is the exact text of the rule I have verified, with its source and 
effective date. Use ONLY this text and my facts. Do not rely on anything 
else you know.

Rule text: [paste]
Source and effective date: [paste]
Facts: [paste]

1. Apply the rule step by step to the facts.
2. Quote the specific words of the rule you rely on at each step.
3. Flag any fact that is missing or ambiguous.
4. State your conclusion and how confident the text makes it.

Requiring quotes lets you spot-check quickly against the source.

Prompt 4: Challenge the answer

Act as a skeptical reviewer. Here is a conclusion I reached: [paste].

1. What are the strongest arguments against it?
2. What exceptions, limits or phase-outs might apply?
3. What would change the answer (different year, state, entity type, facts)?
4. What authority would I need to find to be sure? 
   Mark anything you are not certain exists.

What does a worked example look like?

Say a finance manager at a US company wants to know whether the company can deduct gifts it sends to clients at year end.

Fact pattern (cleaned): "A US C corporation sends holiday gift baskets to clients, roughly $60 each, in the current tax year. Are these deductible?"

Prompt 1 output (typical): The AI restates the issue as deductibility of business gifts, and assumes federal income tax, a calendar-year taxpayer, and that the gifts are not de minimis items. It asks whether the recipients are employees, whether the gifts are food or entertainment, and whether the company's name is on the item.

Prompt 2 output (typical): It points you to the Internal Revenue Code's business gift rules, around section 274, and to the general business expense provision. It also suggests checking the IRS's guidance on gifts and entertainment.

Your verification: You open the statute text yourself. You find that the deduction for business gifts is limited per recipient per year, and you read the exact dollar limit and the exceptions (for example, for incidental costs like packaging, or items with the company's name that are widely distributed and cost very little). You confirm the language applies to the current tax year, not just a previous one. You also check whether your state follows the federal rule.

Prompt 3: You paste the verified text and your facts. The AI walks through it: at $60 per recipient, the per-recipient limit means only $25 of the cost is deductible, and the company should track recipients to apply the limit across multiple gifts to the same person.

What the example shows: the AI did not know the answer. It produced the path. The accurate dollar limit came from the statute you opened, and the conclusion rests on that text. Had you asked the AI for the limit directly, you would have been trusting memory on a number that you could check in thirty seconds.

How do you check an AI tax answer?

Run every AI-assisted answer through this checklist before it goes into a memo, an email to a client, or a filing position.

  1. Does each citation exist? Search for the exact section, ruling or case name in an official or licensed source. If you cannot find it, treat it as invented.
  2. Does the source say what the AI claims? Read the passage. AI often cites a real document for a statement it does not contain.
  3. Is it the right tax year? Check effective dates, sunset provisions and amendments. A rule that applied last year may have been changed or expired.
  4. Is it the right jurisdiction? Federal rules do not automatically apply at state or local level, and other countries differ entirely. States may conform to federal law, partially conform, or ignore it.
  5. Is it the right level of authority? Statutes and regulations carry more weight than agency FAQs, publications or commentary. Informal guidance explains the tax authority's view but is not the same as the law itself.
  6. Are the numbers right? Recheck every rate, threshold, percentage and deadline against the source. Do the arithmetic yourself.
  7. Did it miss an exception? Ask: what would make this not apply? Then look for it in the text.
  8. Are the facts the AI used the facts you have? Reread the assumptions it listed.
  9. Does a qualified person agree? For anything material, get review from someone with tax credentials or from outside counsel.

A useful habit: if an AI answer sounds smooth and certain, that tells you nothing about whether it is correct. Fluent wrong answers are the main risk.

Where should you not trust AI on tax?

  • Recent changes. AI tools may not know about legislation, rulings or guidance published after their training, unless they search the web, and even then they may pull from low-quality pages.
  • Citations from memory. Fabricated sources are a known failure. Lawyers have been sanctioned for submitting court filings with cases that an AI tool made up. Tax work has the same exposure.
  • Thresholds and deadlines. Round numbers and common figures get mixed up across years.
  • Local and international rules. Less training material means more mistakes.
  • Anything involving your private data in an unapproved tool. That is a confidentiality and policy problem separate from accuracy.
  • Final positions. Whether to take a filing position is a professional judgment with accountability attached. AI cannot hold that accountability.

Which AI tools should you use for tax research?

There are three categories, and the right choice depends on your company's rules.

  • General AI assistants. Good for framing, explaining and drafting. Weakest on citations and currency. Use only with sanitized facts and where policy allows.
  • AI features inside tax research platforms. The major tax research vendors have added AI assistants to their subscription products. These are generally designed to answer from the vendor's own library of authority, which can make sources easier to trace. You still open and read the cited material, since the summary can be wrong even when the source is real.
  • Your company's approved internal AI. If your company provides a governed tool, it is often the only one permitted for tax data. Ask your IT or compliance team what is approved.

Whatever you use, the verification steps do not change.

How do you keep a record of AI-assisted research?

Reviewers and auditors want to know how you reached a position. For each question, keep:

  • the question and date
  • the cleaned fact pattern
  • the tool and prompts used
  • the authority you read, with links or citations and effective dates
  • the conclusion and your confidence
  • open uncertainties and who reviewed it

A shared document works, but it gets messy across a team. Questions repeat, answers go stale, and nobody can tell which memos were verified.

Could you build a tax research log for your team?

If your team handles recurring tax questions, a small internal tool can replace the scattered documents: one place where each question is logged with its fact pattern, the AI draft, the sources checked, a verified or unverified status, a reviewer sign-off, and a flag when the tax year rolls over and the answer needs rechecking. Remy can build that kind of app from a plain-English description, with a database, sign-in and roles (for example, preparer and reviewer) included, and deploy it to a live URL. Your team would own it rather than rent another subscription. It supports your process; it does not replace the verification itself.

Frequently Asked Questions

5 questions

Can I use ChatGPT or another general AI for tax research?

You can use it to structure a question, learn the vocabulary, and draft a summary, if your company's policy allows it and you remove confidential details. Do not rely on it for citations, current rates or final conclusions. Verify everything against primary sources.

Can AI give tax advice?

AI can produce text that reads like advice, but it is not a licensed professional and carries no accountability for the result. For decisions that affect filings, penalties or large amounts of money, a qualified tax professional should review the answer.

How do I know if an AI-cited tax source is real?

Search for it directly in an official source, such as the government's own website for statutes and regulations, or in your licensed research platform. Match the exact section number, title and year. If you cannot find it, assume it does not exist.

Is it safe to paste company financial data into an AI tool?

Only if your company has approved that tool for that kind of data. Otherwise, remove names, tax IDs, account numbers and anything that identifies a person or transaction, and ask your IT or compliance team if you are unsure.

How much time does AI save on tax research?

It saves the most time on framing the question, summarizing rules you have already read, and drafting the memo. It saves little on verification, which should stay a fixed part of the process. Expect the gain to come from faster starts and cleaner write-ups, not from skipping the reading.

How we source, write, and correct our articles. Editorial standards

Related articles

Straight to your inbox.
New pieces as they publish. No noise.