On this page
A CPA evaluating AI research tools has a practical problem: the category is new, the marketing language is interchangeable, and a demo always looks good. What you need is a way to tell the tools apart on the dimensions that matter in practice. This guide gives you that. It does not rank competitors, because we would be describing products from their marketing pages rather than from use, and you deserve better than that. It gives you the criteria, a test, and our own answers.
01Three kinds of tool
- General-purpose AI assistants.Fast, fluent, and trained on the open internet. They answer tax questions from what they absorbed in training, which has no particular reason to be current, complete, or the Code. Useful for drafting and brainstorming; not a research source on their own.
- Traditional research platforms with an AI layer.The established editorial libraries, with a conversational front end added on top. Strong on editorial explanation and history; evaluate how the AI layer cites, what it searches, and whether state coverage is as deep as federal.
- Purpose-built AI tax research tools.Built from the start to answer from a defined corpus of primary authority with citations attached. The category Partnertax.ai belongs to. The criteria below are the ones that separate the good ones from the rest.
02Ten criteria that matter
| Criterion | The question to ask | What a good answer looks like |
|---|---|---|
| Source corpus | What does it answer from, and can I see the list? | A published inventory of primary authority with refresh dates, not "trained on tax law." |
| Citations | Does every claim carry a citation I can open? | A link to the section itself, not a footnote naming it. |
| State coverage | Which states, and from the state's own authority? | All states, with state answers citing state statutes and guidance rather than a federal analogue. |
| Currency | How quickly do new laws, regulations, and rulings arrive? | A visible refresh cadence and a way to see what changed. |
| Documents | Can it answer against my client's K-1 or operating agreement? | Upload and ask, with the document kept out of any training. |
| Settled law versus current thinking | What happens when the IRS has not addressed the question? | A clearly labeled mode that searches the live web, never blended silently with primary-law answers. |
| Multi-step research | Can it handle a long, multi-part question with scenarios? | A deliberate research mode that works through the parts and shows its sources for each. |
| Data handling | Where does client data go, and is it used to train anything? | A plain-language answer, in writing, that you can put in your own engagement letter. |
| Workflow fit | Does it produce anything I can use beyond a chat answer? | Comparison tables, filing templates, and integration with the tools you already work in. |
| Pricing | Is the price published, and can I try it before paying? | A public pricing page and a free trial without a credit card. |
03A twenty-minute test you can run on any tool
Ask five questions whose answers you already know, chosen to probe different weaknesses. Then grade the tool on whether the answer was right, whether it cited the authority, and whether the citation checked out when you opened it.
- A threshold question."When is a basis adjustment mandatory on a transfer of a partnership interest without a §754 election?" The answer should name §743(d), the $250,000 substantial built-in loss test, and the second test added in 2017. A tool working from memory often gets one of the three.
- A state deadline.Pick a state your firm files in and ask for the pass-through entity tax election deadline. The answer should cite the state's statute or form instructions, not a federal provision, and should say which tax year it applies to.
- A recently changed rule."Are domestic research expenditures deductible or amortized for a tax year beginning in 2025?" The answer should know about §174A and the 2025 legislation. A stale corpus will describe five-year amortization as current law.
- A conformity question."Does [state] follow the federal treatment of [a recently changed provision]?" The answer should distinguish rolling from fixed-date conformity and cite the state's conformity statute.
- A question with no guidance yet.Ask about an issue where you know the IRS has not spoken. A good tool says so, and if it offers what practitioners are currently saying, it labels that clearly as something other than primary law.
04How Partnertax.ai answers each criterion
- Source corpus.The AI Tax Assistant answers from a curated library of the Internal Revenue Code, Treasury Regulations, the Internal Revenue Bulletin and Manual, IRS written determinations, Tax Court opinions, income tax treaties, and the statutes, regulations, and forms of all 50 states plus the District of Columbia. The full inventory, with its sources and the date of the current manifest, is published on our corpus page.
- Citations.Every claim in an answer carries a citation chip that opens the authority behind it, linking to the official publisher (eCFR, irs.gov, the state's own site) wherever one exists.
- State coverage.All 50 states and the District of Columbia, with state answers drawn from state authority. The State Tax Matrix compares filing, rate, due-date, withholding, and PTE attributes across states side by side, each cell cited.
- Currency.The corpus page shows the date of the current library manifest. The 2025 federal legislation and the guidance that followed it, including §174A and Rev. Proc. 2025-28, are in the library.
- Documents.Upload a K-1, operating agreement, or notice and ask questions against it. Uploaded documents are used to answer your question, not to train anything.
- Settled law versus current thinking.Web Search Mode is a separate, labeled mode for questions the IRS has not addressed. It is never mixed silently into primary-law answers.
- Multi-step research.Deep Research Mode works through multi-part questions and what-if scenarios, reporting sources for each part.
- Data handling.Client data is never used to train models, and sensitive identifiers are masked before documents are processed. The details are on our security and privacy pages.
- Workflow fit.Beyond the chat: the State Tax Matrix for comparisons, free Taxflows templates for elections and letters, and an MCP server that brings the same cited library into Claude and other MCP-capable tools.
- Pricing.Plans and prices are published on the pricing page, and the free trial requires no credit card.
Run the twenty-minute test on us the same way you would on anyone else. Start with the AI Tax Assistant, open the corpus page to see what it answers from, and compare a few states in the State Tax Matrix. The citations will either check out or they will not, and that is the only ranking that matters.
Try it free at partnertax.ai, no credit card required.
