AI Tax Prep in 2026: Not Quite There Yet

WCG vigorously tested and demo’d the following providers. There might be more in this space, but these are the ones we chose to chase down.

TaxGPT

Not ready for production at a firm of our scale and complexity. They used pixel recognition within a browser which is basically AI navigating through a website like Proconnect and clicking and typing. It chewed up your laptop as well, so you couldn’t do other work. No integration to Office 365 unless you used the website version of email which no one on the planet does when we have Outlook. We will leave it there for now and revisit. They are the least prepared of these five.

Thomson Reuters Ready to Review

Works inside Thomson Reuters’ own environment, which creates problems we will get to shortly. Beyond the structural issue, it has a more immediate limitation: it does not handle K-1s or consolidated 1099 brokerage statements. On the roadmap for November 2026. Seriously? K-1s and 1099 consolidateds are on the roadmap? They are the roadmap for any serious tax return preparation.

For WCG, that is disqualifying. The bulk of our complexity lives in exactly those document types. Thomson Reuters does have a PII/SOC2 compliant research environment through their Co-Counsel product, which is useful for other purposes, but Ready to Review is not a fit for our practice.

Instead

Prepares tax returns in its own proprietary software. The concept is appealing until you think through what happens when a tax return hits a wall. No AI system today completes 100% of tax returns without some human involvement. Think Form 3115, an unusual depreciation method, or a client who surfaces a complex K-1 mid-engagement. When a tax return is 90% done inside a proprietary system and the last 10% requires a human decision, you are stuck. You cannot finish it there, and you cannot transfer a half-completed tax return into Lacerte or UltraTax or Proconnect or any other recognized software to pick up where AI left off. You start over. That is not a workflow inconvenience; that is a built-in reset cost every time AI hits its ceiling.

Oh, and you wasted money and time. Not just time.

There is also a longer-term problem. Client complexity tends to grow. A straightforward 1040 one year becomes a K-1-heavy filing the next. If your AI tool prepared prior-year tax returns in proprietary software, your history is trapped in that system. That creates problems for comparison, continuity, and review.

Black Ore

Occupies a third category that is worth understanding on its own terms. It is not proprietary software like Instead or Thomson Reuters Ready to Review, and it is not a straight data push like Juno. Black Ore ingests source documents, applies its own federal and state tax logic, runs internal calculations, builds workpapers, and performs anomaly detection before exporting mapped inputs into Lacerte. That is meaningfully more sophisticated than OCR and data entry.

The problem is where that sophistication lives. All of that work happens inside Black Ore’s own platform. Lacerte receives the resulting inputs, and then does what Lacerte always does: runs its own calculations, applies its own diagnostics, and performs its own consistency checks from scratch. Whatever Black Ore caught internally gets rechecked anyway. Whatever Black Ore missed shows up in Lacerte’s diagnostic queue, waiting for your team. The pre-prep intelligence layer does not shorten the back end of the workflow in any meaningful way, because the back end belongs to Lacerte regardless.

Your team still handles the actual Lacerte diagnostic cycle, e-file setup, 8879 generation, printing, and delivery. Black Ore replaces the first-level preparer, not the last mile. For a firm that already has SurePrep pushing data and experienced reviewers handling Lacerte, the question becomes: how much are we paying to automate the middle of a process whose beginning and end remain unchanged?

Black Ore’s pricing of $100 to $150 per tax return on average answers that question. At $100,000 to $150,000 annually, a firm is taking a significant financial position on a product in its early seasons. That is a lot of money to shorten the middle while the bookends stay the same.

Oh, and the $100 to $150 is their average… and it comes from $1 per scanned page and $3 per K-1 scanned page. We believe our cost would be more like $200 to $250 given the per page fee structure. So, you take SurePrep at $25 per tax return and add an hour of labor at $80 per hour (super high side) and you are staring at $105 compared to $200 to $250 plus the same $80 labor add-on. Not good.

Juno

Also integrates with Lacerte via API, and the architecture is sound for the same reasons as Black Ore. After working through demos and follow-up questions, though, it became clear that Juno is closer to a data entry tool than a full preparation system. Their own sales rep confirmed, candidly, that end-to-end tax return completion is not where they are headed. What they deliver is closer to SurePrep with a better interface. We already have SurePrep.

Sidebar: We giggled at the “end to end tax prep is not where we are heading.” Oh really? Your customers might have a different perspective on that. Adapt or die. We’ll see.

What WCG actually wanted from Juno was their PII/SOC2 compliant sandbox for uploading and analyzing existing tax returns. Juno declined to sell sandbox access on its own. So we are purchasing a block of 500 tax returns at $44 each, tax returns we have no plans to use, solely to get access to the compliance and secure environment with ZDR (which is fancy AI speak for zero data retention). If that arrangement sounds odd, it should. It is also a fair summary of where the AI tax prep industry sits right now.

Evolution Speed

We just highlighted a bunch of deficiencies in this space. Perhaps this is unfair. They are new. They are trying. We wish they understood their own limitations better, and applied that actual tax return preparation, but that will come. Tech-forward CPA firms like WCG will continue to provide pushback and feedback and resistance and everything else to help shape this space. We must in a very self-serving way. We must move off of compliance work and more towards advisory work. And to do this with fewer CPAs and tax professionals coming out of school. In other words, the pipeline is thinning yet the workload is not.

At the speed all this stuff is evolving, we suspect standards will emerge, and then we can have a gold standard be selected. Today? No standard let along a gold one.