CurateSuite
List7 min read

AI Tax Preparation: What AI Can and Cannot Do

What AI actually handles in US tax preparation: research, intake, document extraction, and drafting. Plus what still belongs to your team.

By CurateSuite
An accountant in a muted-blue shirt sits at a light oak desk in warm afternoon daylight, reviewing a printed tax return beside an open laptop. Document folders are stacked to one side. The atmosphere is calm and focused, with a large window and soft-focus bookshelves in the background.

For a US tax firm in 2026, the question about AI is not whether it helps. It does. The harder question is where it fits: which jobs it can handle, which it can only assist with, and which still need a licensed professional. Get that wrong and it costs you money or adds risk, sometimes both.

The sections below go through the tax workflow stage by stage and show where AI fits today. Capability details come from vendor-published specifications, assessed the same way across every tool. For specific product comparisons and pricing, the guide to AI tools for tax preparation in 2026 covers the full shortlist. For the wider picture of AI across all accounting work, see what AI can and cannot do in your accounting practice.

What AI handles in tax preparation

1. Tax research

Research has shifted faster than anything else in the tax workflow. A question that used to take 30 to 45 minutes in Checkpoint, Westlaw, or CCH can often be answered in five, and the better tools cite the underlying authority so you can verify the answer rather than take it on trust.

The distinction between tools matters. General-purpose AI assistants can produce confident-sounding answers with no primary source behind them. Purpose-built tax research platforms query the actual statutory and regulatory corpus and link every answer back to the underlying rule.

Blue J covers US federal tax, state and local tax, Canadian tax, and UK tax. It handles follow-up questions in context, so a multi-step research path does not restart from scratch. Vendor data puts average time savings at around three hours per user per week. Pricing for sole practitioners is $1,498 per year (billed annually as $125 per user per month), with a 7-day free trial. More than 4,000 firms use it, including Crowe, RSM, and Larson.

CoCounsel Tax works from inside Thomson Reuters Checkpoint, drawing on the Internal Revenue Code, Treasury regulations, IRS rulings, federal case law, and expert commentary. It is the natural choice for firms already paying for Checkpoint. Firms without a Checkpoint subscription face a larger commitment than a standalone tool, since they are buying the whole research library alongside the AI layer. Pricing is bundled and quoted on request.

TaxGPT covers AI research with citations to the underlying code and regulations, handles multi-state comparisons, and integrates with Drake, ProConnect, and UltraTax. A free tier covers up to 25 questions and one return review. Professional plans are custom-priced and require a demo before activation.

All three tools save research time. None of them replace the professional who interprets the answer, advises the client, and signs the position.

2. Client document collection and intake

Getting clients to deliver the right documents before the filing deadline is the task that eats the front of every season. AI tools in this category do two things: they generate personalized checklists from prior-year data so clients do not get asked for documents they already sent, and they chase missing items automatically rather than waiting for the preparer to follow up.

StanfordTax pulls prior-year data from Lacerte, UltraTax CS, Drake, ProConnect, ProSeries, CCH Axcess, and CCH ProSystem fx to personalize the current-year questionnaire and document checklist. It covers 1040, 1120, 1120S, and 1065 returns. The free tier covers all four return types with no client minimum, which makes it the lowest-risk way to test this category. The Premium plan is $18 per user per month.

Soraban goes further along the workflow. It replaces paper organizers with smart dynamic questionnaires, pushes extracted data into the tax engine with a vendor-claimed 97 percent accuracy rate, and handles branded return delivery with e-signature and payment routing. Pricing is per return: $25 for collect-only, $40 for collect-plus-deliver, with a 50-return annual minimum. That positions it for firms running 150 or more returns per year rather than sole practitioners.

3. Source document extraction

Once documents arrive, AI extracts structured data from varied source materials: W-2s, 1099s, K-1s, bank statements, and brokerage summaries. This replaces manual data entry, which carries transcription errors and takes real hours at volume.

StanfordTax and Soraban both include extraction as part of their intake workflows. Soraban's data push into the tax engine is the most integrated option for Lacerte, UltraTax, Drake, and ProSeries users. The tax and research category page lists the full range of tools in this space.

4. Return review and anomaly flagging

AI can catch obvious anomalies before the return leaves the desk: missing schedules, implausible deduction ratios, figures that have shifted significantly year over year. TaxGPT's return review feature works in this space, as do the review checks inside Soraban's delivery workflow.

This is a job where AI reduces the chance of a missed error rather than replacing the review itself. The partner sign-off stays in place.

5. Client communication drafts

Several tools generate first-draft emails, memos, and return delivery summaries from the data they already hold. Blue J drafts memos from the authority it retrieves. Soraban generates post-return summary emails as part of the delivery flow. Canopy includes email summarization as part of its AI feature set alongside checklist generation and form auto-fill.

A draft is a starting point, not a finished email. The point is that it already pulls in the right figures and references the right rules, which beats starting from a blank page.

6. Practice workflow and deadline tracking

TaxDome and Canopy sit at the practice management layer rather than the preparation layer, but for a tax firm they are part of the same stack. TaxDome gives solo and small firms a single platform for client portal, workflow, documents, e-signatures, billing, and built-in tax organizers. Pricing runs from $700 to $1,100 per user per year, billed annually, with no free trial. Canopy covers the same ground with AI checklist generation, email summarization, and form auto-fill layered in; Standard plans start at $74 per user per month (billed annually) with a 14-day free trial.

Both tools handle deadline tracking, document requests, and client approvals, which removes the coordination overhead that grows through the season as return counts increase.

At a glance

ToolWhat it doesStarting price
Blue JAI tax research with primary source citations$125/user/month (annual)
CoCounsel TaxResearch inside Thomson Reuters CheckpointCustom (Checkpoint bundle)
TaxGPTResearch, multi-state comparison, return reviewFree tier; custom for Professional
StanfordTaxIntake, personalized checklists, prep-ready binderFree tier; $18/user/month paid
SorabanIntake, data push to tax engine, return delivery$40/return (collect + deliver)
TaxDomePractice management, portal, tax organizers$700/user/year
CanopyPractice management with AI features$74/user/month (annual)

Prices come from vendor-published specifications and are re-checked before each article update. Custom means the vendor does not publish a list price.

What AI does not do in tax preparation

It does not replace tax preparation software

The term "AI tax preparation" implies a tool that prepares returns. No tool on this page does that. Drake, Lacerte, ProConnect, UltraTax, CCH Axcess, and the other US tax engines remain the spine of any tax practice. The AI tools in this article sit alongside them and remove hours from the surrounding work: research, intake, document extraction, and review.

It does not apply professional judgment to complex facts

Tax positions on novel transactions, estate planning structures, and complex pass-through scenarios involve characterizing the facts, weighing competing authorities, and advising a specific client on their specific situation. That is not what current AI tools do. Blue J and CoCounsel Tax retrieve and summarize authority well; they do not make the judgment call on how the rule applies to the client's facts, and the vendors say so directly.

It does not handle multi-state nexus analysis without specialist input

Multi-state nexus is complicated, shifts with state-by-state legislation, and varies by entity type and transaction type. AI research tools can answer jurisdiction-specific questions, but for clients with genuine multi-state sales tax exposure, specialist tools built for rate determination, registration tracking, and filing are a separate category from tax preparation AI.

It does not take responsibility for the return

Every tool in this category operates as an assistant to the preparer, not as a co-signer. The professional liability sits with the licensed preparer and the firm. AI tools can accelerate and check the work; they do not share the exposure.

How to build the stack

The order matters as much as the tools.

If document collection is the bottleneck, fix intake before anything else. StanfordTax's free tier costs nothing to test and gives you personalized questionnaires immediately. If you run 150 or more returns and want the data pushed into the tax engine, Soraban is the natural next step.

If research time on complex questions is the bottleneck, Blue J or TaxGPT are the tools to test. Blue J's 7-day trial makes this a low-risk evaluation.

If coordination across the whole season is the drain, TaxDome or Canopy handle the practice workflow layer and reduce the admin overhead across multiple return types and deadlines.

For guidance on evaluating any of these tools against your firm's actual workflow, see how to evaluate AI accounting software. For the full roundup of AI tools beyond tax preparation, see the main guide to AI tools for accountants. For the full tax and audit stack including audit tools, see the AI tools for tax and audit guide.

Common questions

What does AI actually do in tax preparation?

AI handles the surrounding work rather than the return itself. On the research side, Blue J and CoCounsel Tax answer tax questions from primary authority and cut routine question time from 30 minutes to around five. On the intake side, StanfordTax and Soraban personalize client questionnaires from prior-year data and automate document chasing. On the extraction side, those same tools pull structured data from source documents and push it into the tax engine. The return still gets built in Drake, Lacerte, ProConnect, or UltraTax. AI speeds up the work around it.

Does AI replace the accountant in tax preparation?

No. Professional judgment on complex tax positions, client advice, and the signature on the return all remain with the licensed preparer. AI compresses the research and intake phases; the interpretation, the advice, and the sign-off stay with your team.

What is the best AI tool for tax preparation?

The right tool depends on your bottleneck. For research, Blue J or CoCounsel Tax (if you are already on Thomson Reuters Checkpoint). For intake and document collection, StanfordTax's free tier is the starting point, and Soraban suits higher-volume firms. For practice workflow, TaxDome or Canopy. The guide to AI tools for tax preparation in 2026 compares specific options in depth.

Is there a free AI tool for tax preparation?

Yes. StanfordTax has a genuine free tier covering all four US return types (1040, 1120, 1120S, 1065) with no client minimum. TaxGPT's free tier covers up to 25 tax research questions and one return review. Blue J offers a 7-day free trial for sole practitioners, with no credit card required.

How does AI in tax preparation affect data security?

Client tax data is sensitive and most AI tools for tax professionals use cloud deployment. Before adopting any tool, confirm in writing that the vendor does not use your client data to train shared AI models, that data is encrypted at rest and in transit, and that you can export or delete on request. The guide to evaluating AI accounting software covers the full set of questions to ask.

Find the right AI tools for your firm

Seven questions, two minutes, a short list of the five tools that fit your firm best. Your results, instantly. No sign-up needed.

Take the matchmaker

Tools referenced in this article

More articles

Last updated 2026-06-23. Tool comparisons are based on vendor-published specs. See our methodology.