AI Tax Software for CPA Firms: 2027 Season Prices Compared

As of 1 November 2026, the best AI tax software for CPA firms depends on volume: Cranston charges a published $40 per prepared return, TaxScout AI starts at $49 a month for 150 returns, and Instead, Filed and Black Ore quote on request. Pick by return mix, the tax engine you already run, and written answers on client data.

TL;DR:

Cranston publishes $40 per prepared return, so 300 returns cost $12,000 a season; TaxScout AI's Firm plan covers 500 returns for $1,430 prepaid.

Instead's platform and e-file are free, with AI agent credits sold by volume and valid for 12 months.

Of the pages read, only Thomson Reuters states that client data is never used to train third-party AI models; most vendors say nothing.

The IRS Office of Professional Responsibility said in June 2026 that Circular 230 duties apply in full to AI output.

Pilot two tools on 20 last-year returns each before the 2027 season, and score time saved against review errors.

What is AI tax software for CPA firms, and what does it actually do?
Which AI tax tools should a CPA firm shortlist for 2027?
How much does AI tax software cost a CPA firm?
Is client tax data safe in AI tax software?
What do the IRS and AICPA expect from firms using AI?
How should a firm test AI tax software before the season?
Why does independent evidence matter when choosing AI tax software?
Realistic expectations: review time, not preparation time, is the cost
Find your next AI tax software tool through The AI Ledger
Frequently asked questions

What is AI tax software for CPA firms, and what does it actually do?

AI tax software reads client documents, drafts a return or research memo, and flags items for a person to review. It falls into three groups, and mixing them up is the most common buying mistake.

Return preparation: tools such as Cranston, Filed, Black Ore Tax Autopilot and TaxScout AI draft 1040s and sometimes business returns from source documents.
Platforms with their own engine: Instead bundles AI agents with its own e-file, which the vendor says covers the IRS, 48 states and five entity types: 1040, 1120, 1120-S, 1065 and 1041.
Research and drafting: Blue J and Thomson Reuters CoCounsel answer technical questions with citations and draft memos. They do not prepare returns.

Adoption is real but uneven. In the 2026 tax software survey by The Tax Adviser, 65% of 1,808 AICPA member respondents used AI for tax research, 32% for client communication, and 16% had no plans to use it, down from 24.3% a year earlier. The same survey shows where the engines sit: UltraTax CS 22.3%, Lacerte 16.7%, Drake Tax 14.3% and CCH Axcess Tax 13.8%.

That spread matters because most AI preparers work beside your engine, not instead of it. Cranston lists Drake Tax, CCH Axcess Tax, UltraTax CS, Lacerte and ProSeries. Filed lists Drake, CCH Axcess, Lacerte and ProConnect, and runs as a sidebar inside the tax software. If your engine is not on a vendor's list, the tool is a poor fit regardless of its demo.

Pro Tip: Ask each vendor for the exact list of forms and states it prepares without a human touching them. "Handles 1040s" can mean anything from a W-2 filer to a return with three K-1s and rental schedules.

Which AI tax tools should a CPA firm shortlist for 2027?

Shortlist by return mix first and brand second. The table uses editor scores and prices from The AI Ledger directory, each verified against the vendor's site on its last verified date, plus the vendor pages read on 1 November 2026.

Tool	Best for	Editor score	Price signal	Client data training
Instead	Firms wanting AI plus a free e-file platform	4.5	Free platform; credits from $30	Not stated
Black Ore Tax Autopilot	High-volume 1040 preparation	4.4	On request	Not stated
Filed	Firms keeping CCH, Drake, Lacerte or ProConnect	4.4	Flat plan scoped to taxpayers	Not stated
Cranston	Per-return pricing across 1040 and business forms	4.2	$40 per return	Not stated
Accrual	Larger firms with funded pilots	4.2	On request	Not stated
CCH Axcess Expert AI	Existing CCH Axcess Tax firms	4.2	On request	Not stated
TaxScout AI	Small firms wanting a cheap 1040 pilot	3.9	$49 to $825 a month	Not stated
Blue J	Tax research	4.5	$1,498 per user a year	Trained on primary sources and Tax Notes content
Thomson Reuters CoCounsel	Research and drafting on Checkpoint	4.4	Essentials $1,300 per user a year	Never used to train third-party models

Two recent moves affect the shortlist. Wolters Kluwer announced agentic return preparation in CCH Axcess Tax on 23 October 2026. CPA Practice Advisor's 28 October report describes it as drafting individual returns from client documents and prior-year data, including K-1 data, with a professional making the final decisions. The report gives no general availability date or price, so treat it as roadmap until your account manager confirms what is live in your licence.

Thomson Reuters also sells Ready to Review for 1040 preparation inside its ecosystem, provisioned through CoCounsel Tax. The directory lists it at an editor score of 4.2 out of 5 with custom pricing.

Pro Tip: If you run CCH Axcess Tax, ask Wolters Kluwer what is included in your current licence before paying a third party for the same job. The reverse also holds: a vendor with deeper preparation may beat a bundled feature.

How much does AI tax software cost a CPA firm?

The cheapest published option for 300 returns is TaxScout AI at $1,430 for the year, against $12,000 for Cranston at $40 a return, but the two are not like for like. Always compare cost per reviewed, accepted return, not the licence fee.

Cranston's pricing page states $40 per prepared return, state returns included, with no seat or onboarding fees. A client with a personal and a business return counts as two. Volume pricing starts past 2,000 returns a season. TaxScout AI's pricing page lists five plans on a 12-month term:

Plan	Monthly	12 payments	Prepaid year	Return credits	Overage
Solo	$49	$588	$470	150	$4
Firm	$149	$1,788	$1,430	500	$4
Practice	$415	$4,980	$3,990	1,500	$3
Volume	$825	$9,900	$7,900	3,500	$2

The worked example below uses published prices only. Filed, Instead credits, Black Ore, Accrual and CCH are quote-only, so they cannot be priced honestly here.

Returns a season	Cranston at $40	TaxScout AI, prepaid	Blue J, 3 users
100	$4,000	$470 (Solo)	$4,494
300	$12,000	$1,430 (Firm)	$4,494
1,000	$40,000	$3,990 (Practice)	$4,494

The sums: Cranston is returns times $40. TaxScout's 300-return case fits inside the Firm plan's 500 credits, so the cost is one prepaid year at $1,430, or 12 times $149, which is $1,788 billed monthly. Blue J is 3 users times $1,498, which is $4,494, and it is a research tool rather than a preparation tool, so it is a separate budget line. Blue J's Team plan is quoted, so the Individual rate is a ceiling for planning only.

Three costs sit outside the licence. First, your tax engine still bills you. Second, reviewer hours: a returns-per-hour figure from a vendor is self-reported, so measure your own. Third, unused credits. TaxScout credits belong to the term, not the tax year, while Instead credits expire 12 months after purchase.

Decision rule: choose Cranston if you want per-return billing that scales down in a slow year. Choose TaxScout if you have a spare weekend to pilot and can tolerate a young product, which the directory rates 3.9 out of 5. Choose Filed or CCH if you value working inside your engine and will negotiate a multi-year price. Choose Instead if you also want to move e-file off your current stack.

Is client tax data safe in AI tax software?

Treat it as unproven until the vendor puts it in writing. Of the vendor pages read for this article, only Thomson Reuters made a plain statement on training: customer data "remains under your firm's control and is never used to train third-party AI models."

The others varied. Instead's pricing page shows an AICPA SOC logo and links to a security page, but it makes no statement on retention or training. Filed says it holds ISO 27001 certification issued by Schellman, displays an AICPA SOC badge, and offers a custom data processing agreement on Enterprise. It does not say whether data trains models. Cranston's page links to security and privacy documents without naming a certification, and the directory notes the firm is still completing its SOC 2 Type II examination. TaxScout states 7-year retention and a 90-day export window after cancellation but is silent on training.

Where the vendor does not say, the table above reads "not stated". That is not an accusation. It is the gap you close with a data processing agreement before client files go in.

Ask for the SOC 2 Type II report, not the badge, and check the period it covers.
Ask where data is hosted and which model providers see documents.
Ask for training terms in the contract, including whether firm-specific "memory" counts.
Ask what happens at exit: export format, deletion date and certificate.

Pro Tip: Send the same five security questions to every vendor in one email and compare the replies side by side. Vendors that answer in writing within a week are usually the ones that will answer during an IRS or state inquiry.

What do the IRS and AICPA expect from firms using AI?

They expect the same professional standards as before, applied to a new tool. Journal of Accountancy reported on 26 June 2026 that the IRS Office of Professional Responsibility had outlined how Circular 230 applies to AI: due diligence on facts, citations and calculations, competence with the tool, human judgment on the final answer, and use of secure, enterprise-approved AI for client data.

The bulletin also raised fees. Charging for hours you did not spend because AI shortened the work could raise an unconscionable fee concern, and it says cost reductions should be credited to the client's account. On disclosure, it says practitioners should tell clients about AI use "as appropriate", without prescribing a form. The AICPA published introductory guidelines for responsible AI use in federal tax practice, and the AICPA site is the place to confirm current wording.

For a firm, that translates into four written controls:

A list of approved tools and who may use them.
A rule that no AI-drafted return is signed without a named reviewer.
A log of what the AI produced and what the reviewer changed.
A client engagement letter paragraph on AI use and billing.

The filing calendar is not a reason to skip this. The 2026 season opened on 26 January 2026, according to ThinkAdvisor's report of the IRS announcement. As of 1 November 2026 the IRS had not announced the 2027 opening date, so plan a pilot in November and December, before the January rush.

How should a firm test AI tax software before the season?

Run a controlled pilot on last year's returns, with a scoring sheet, before touching a live client. The steps below take about two weeks of part-time effort.

Pick 20 closed returns across your mix: simple W-2, Schedule C, rental, K-1 and one state with credits. Strip or mask identifiers if the vendor's terms are unsigned.
Run each shortlisted tool on the same set, with a preparer who has not seen the final answers.
Score every line against the filed return: correct, wrong, or missing. Record minutes of review per return.
Check the audit trail: can a reviewer trace each figure to a source document?
Price the result at 300 and 1,000 returns using the tables above, adding reviewer time at your internal hourly rate.
Decide with a threshold, for example "at least 90% of lines correct on simple returns and review time under 15 minutes", agreed before you start.

Where each tool tends to fall short is in the edges. Per-return tools cost the same for a one-page 1040 as for a messy one. Flat plans reward volume but lock you in for a year. Research tools do not prepare anything. Platforms with their own e-file ask you to change process, which is costly in January.

Pro Tip: Include two returns you know contain a deliberate trap, such as a missing 1099-B basis or a K-1 with a late amendment. How a tool flags what it cannot resolve tells you more than a clean pass.

Why does independent evidence matter when choosing AI tax software?

Vendor pages describe the best case, and most of the figures on them are self-reported. Thomson Reuters notes on its own CoCounsel page that its usage metrics are self-reported and not independently verified. An editor score and a last-verified date give you a second, dated opinion and show you when a listing was last checked against the vendor's site.

The AI Ledger scores tools on stated features, published pricing, data handling and fit, and records the date each listing was verified against the vendor's site. It does not claim hands-on testing of every product, and scores are not for sale. Use the full directory to compare scores, then confirm prices and security terms with each vendor yourself.

Realistic expectations: review time, not preparation time, is the cost

AI will draft more of a simple return than it did last year, but you still own every line you sign, so budget reviewer time as carefully as licence fees. Treat any vendor's saving claim as a hypothesis to test on your own returns.

Find your next AI tax software tool through The AI Ledger

Start with the tax prep category page to see every listed tool with its editor score, price and last verified date. Then use the finder to narrow by firm size and software, and the compare page to put two or three shortlisted tools side by side.

Frequently asked questions

What is the best AI tax software for CPA firms?

There is no single best. Instead and Blue J hold the highest editor scores at 4.5 out of 5, but they do different jobs: one is a preparation platform, the other is research. Match the tool to your return mix and tax engine, and pilot two tools on 20 closed returns before committing.

How much does AI tax software cost per return?

Where prices are published, they range widely. Cranston charges $40 per prepared return. TaxScout AI's plans work out between $2.26 and $3.13 per return at full use, from $49 to $825 a month. Filed, Black Ore, Accrual and CCH quote on request, so ask for a written price per return.

Does AI tax software work with Drake, Lacerte and CCH Axcess?

Several do. Cranston lists Drake Tax, CCH Axcess Tax, UltraTax CS, Lacerte and ProSeries. Filed lists Drake, CCH Axcess, Lacerte and ProConnect. Instead uses its own platform and e-file. Confirm the specific forms and states supported for your version, because integration lists are not the same as full coverage.

Is it safe to put client tax returns into AI tools?

It can be, with controls. Thomson Reuters says customer data is never used to train third-party AI models. Most other vendors reviewed do not state their training terms on the pages read. Get the SOC 2 Type II report, hosting location and training terms in writing, and follow the IRS guidance to use secure, enterprise-approved tools.

Do I have to tell clients I use AI to prepare their returns?

The IRS Office of Professional Responsibility bulletin of June 2026 says practitioners should disclose AI use "as appropriate" and credit cost savings to clients, without prescribing a form. Many firms add a short paragraph to the engagement letter. Check the latest AICPA guidance and your state board's rules.