TaxGPT

Positioning

TaxGPT is not the consumer tax helper the category tagline suggests — it is an AI operating system for tax, accounting and advisory firms, built by tax professionals and aimed squarely at CPAs, enrolled agents, tax attorneys and their staff. The company launched in 2023, is venture-backed (its FAQ names Y Combinator and Mangusta Capital), and claims 80,000+ active users and 650,000+ answered tax questions at the time of my review. The product thesis is one connected workflow rather than a chatbot: research feeds writing, client memory feeds research, and agentic workflows execute preparatory work inside the tax software firms already run. For an individual filer looking for deduction advice, this is the wrong product — and its pricing model confirms it.

What it actually does

Six modules make up the platform as listed on taxgpt.com/pricing. AI Tax Research answers federal, state, local and treaty questions with citations — the FAQ says it researches US and Canadian tax law against a database of thousands of tax laws, publications, case law and regulatory guidance, and operates strictly within that verified content. AI Tax Writer drafts client-ready memos, IRS notice responses and engagement letters in the firm's voice. Client Intelligence keeps a persistent memory of every client's entities, filing history and prior-year positions, so each question starts with context instead of a blank prompt. Agent Andrew is an automated return reviewer for Forms 1040, 1041, 1065, 1120 and 1120-S before sign-off. AI Tax Preparation is an agent that prepares partner-level returns inside the tax software a firm already uses. TaxGPT Cowork is the agentic layer: you assign dedicated agents to recurring workflows — individual and business return preparation, extensions and amended returns, R&D credit studies, bookkeeping, tax planning, entity-structure reviews, K-1 production, projection models — and they execute end to end, handing off to your team for judgement. Cowork agents operate inside Drake, ProConnect, UltraTax and CCH Axcess, among others.

Pricing is quote-based across three plans — Pro (solo practitioners), Pro Plus (growing practices) and For Firms (enterprise). Nothing is published: the pricing page says "book a demo" and the FAQ confirms each plan is scoped to team size, client count and capabilities. A free trial exists. Third-party trackers disagree about what it costs — one review site says pricing starts at $49/month, another directory lists $17–$25/month tiers that do not match anything on the vendor's own site — and TaxGPT's own comparison content says plainly: not published; demo required. I treat every public number as unverified and say so.

Security posture is stated explicitly across plans: SOC 2 Type II certified, automatic PII redaction, encryption, AWS and Azure hosting with US-based data processing, role-based access controls, activity logs and audit trails, and a commitment never to train on client data.

My testing notes

The first thing I checked was the grounding claim, because accuracy is the whole ballgame in tax. The FAQ's answer is the right shape: the model operates strictly within a verified corpus of statutes, publications and case law with "multiple guardrails," and answers carry sources. What I look for and did not find is third-party accuracy evaluation — published hit rates against a benchmark like CPA exam questions or a blind firm trial. The company's own numbers (research and response times cut 90%, per the FAQ) are vendor claims. Firms should run the free trial against their own hard questions — state-level nexus issues and industry-specific positions are the natural stress test, and the FAQ claims no industry limitations, which is exactly the claim to probe.

Second, the workflow integration is the real differentiator. Agentic tax preparation inside Drake, ProConnect, UltraTax or CCH Axcess means the output lands where the work already happens, and Agent Andrew reviewing 1040s through 1120-S before sign-off puts a second set of eyes in the loop without changing the firm's software stack. The security FAQ's "closed system built for confidential tax data, not a public AI" framing is aimed squarely at the objection that kills most AI purchases in accounting firms: client confidentiality. SOC 2 Type II plus PII redaction plus a no-training commitment is a credible package for that objection.

Third, the support layer surprised me: in-app University, biweekly webinars, monthly product updates, virtual/on-demand/1:1 training, and tiered onboarding (self-guided on Pro, white glove on Pro Plus, in-person on Enterprise). In my experience evaluating pro tooling, that onboarding structure is where vendor churn either happens or doesn't, and TaxGPT has clearly priced it deliberately across tiers.

Buying advice

For firms evaluating TaxGPT, my advice is to structure the trial as a bake-off rather than a tour. Pick ten research questions from the last season that consumed real staff hours — the state nexus question you escalated, the Section 174 position that needed three hours of reading, the foreign disclosure edge case — and run them through the research module with citations visible. Check the citations yourself against the source; the difference between a tool that cites the right regulation and one that cites a plausible one is the difference between time saved and malpractice exposure. Then run a completed prior-year return through Agent Andrew and count real findings versus noise. That second test is the one I would weigh most heavily, because return review is where an AI either earns a seat in the workflow or gets dismissed after a week.

On commercials, go in knowing that pricing is scoped per practice and entirely negotiable: team size, client count and the modules you need (research only versus research plus Cowork and tax prep) all move the number. Ask directly what happens per seat when seasonal staff join — tax firms' headcount swings harder than almost any other professional services vertical, and pricing that assumes twelve months of ten seats will hurt in April. The vendor's own comparison content confirms pricing is not published and demo-gated, so anything a third-party site quotes you is noise.

Finally, weigh the switching cost honestly. Client Intelligence only compounds once your client history lives inside it, and Cowork agents operate inside your existing tax software. Both create stickiness by design — helpful once you are committed, a reason to trial carefully before you commit.

What it does not do

It is not for individuals — no consumer filing, no 1040-for-you service, and the individual-filer framing in older writeups (including our own catalog entry) is wrong. It does not replace a licensed professional: the research is a drafting and verification accelerator, and partner review is assumed (Agent Andrew is explicitly a pre-sign-off reviewer, not the signer). Pricing opacity is a real friction — you cannot budget without a sales call, and I found no published seat or return-volume rates to sanity-check a quote. Published accuracy evidence is absent. Coverage is US and Canadian tax law; practitioners in other jurisdictions have nothing here. And the vendor's bold claims (10x productivity on G2, 90% research time reduction) are marketing until a firm measures its own before/after.

Verdict

TaxGPT is the most workflow-integrated AI tax product I reviewed in this batch: citation-backed research, persistent client memory, return review and agentic prep that runs inside mainstream tax software, with a security posture that answers the confidentiality question directly. The gap is evidence and pricing transparency — no published rates, no third-party accuracy studies, and ROI figures that are entirely vendor-derived. I rate it 4.0: core functionality is deep across the rubric's heaviest criterion — this is a real operating system, not a wrapper — and support and security signals are strong, but value-for-money is impossible to fully judge against unpublished pricing and the reliability score is capped by unverified performance claims. Firms serious about it should use the free trial as a bake-off against their hardest recurring research questions.

Sources

• https://www.taxgpt.com/pricing
• https://www.taxgpt.com/faqs
• https://www.taxgpt.com/what-is-taxgpt
• https://taxgpt.com/blog/taxgpt-vs-bizora
• https://www.g2.com/products/taxgpt/reviews
• https://agent-finder.co/reviews/taxgpt
• https://unclekam.com/tax-pro-tools/ai-tax-tools/taxgpt-review
• https://www.cpapilot.com/blog/cpa-pilot-vs-taxgpt-vs-chatgpt
• https://taxprepbench.com/compare/juno-vs-taxgpt/
• https://siteoptz.ai/tools/taxgpt