Teaching a New Preparer to Check AI Drafts: A Two-Week Practice Plan

A new Preparer at a firm that uses an AI tool for drafting needs one skill before any other: checking a draft that already looks finished. In these synthetic drills, the senior gives each drafted row an amount, a label and a page reference. A wrong one looks like a right one. Reading rows isn't the same as checking them.

So the Firm can choose to train a new Preparer to check AI drafts before they touch a live one. This plan runs two weeks on synthetic Doe files with errors planted on purpose. Each drill gets logged, and a senior signs off on each one. The tool only drafts. A person checks, decides what each row needs, and the Reviewer signs off on the real work later.

Build the drill files first

A senior builds the drill set before the new Preparer starts. Each drill file holds a few made-up documents for a fictional client, a drafted entry list from those documents, and an answer key. The key lists every planted error by page and row for the senior to compare with the trainee's log.

The senior seeds errors by category. The Firm picks its own list. This plan uses six:

• a wrong amount
• a missing row
• an extra row
• a wrong page reference
• a wrong label
• an out-of-date document in the file

Each drill file gets a version label, such as "drill 3, v1". If the senior fixes a file or changes its key, the label changes, so the log always points at the file the trainee saw.

The Firm can use synthetic Doe documents from its known-answer folder to seed the drill files. That folder tests the tool; this plan tests the person.

The two-week schedule

The Firm sets the pace. Here's one schedule, ten working days, with the seeded error count for each drill.

Day 1: watch. The trainee sits with the senior, who checks one drafted list out loud. The senior works from the Firm's readback checklist and says why each row passes or comes back. No drill, no log row.

Day 2: one category, 2 errors. A short file. The trainee knows the category in advance. When they're done, they open the key and compare.

Day 3: two categories, 3 errors. Still a short file, and the trainee knows both categories.

Day 4: mixed, 4 errors. The trainee isn't told which categories are in the file.

Day 5: review and repeat. The senior goes through the week's log with the trainee. Any category the trainee missed gets a repeat drill that day, on a fresh file.

Days 6 and 7: longer files, 5 errors each. No category hints. The trainee also writes a one-line reason for every row they flag.

Day 8: a clean file, 0 errors. The trainee isn't told it's clean. This drill shows whether they flag rows that are fine. A false flag costs a Reviewer time on a live file, so it goes in the log too.

Days 9 and 10: full files, 6 errors each. These look like a live job. The trainee doesn't know how many errors were planted.

Keep a drill log

The Firm can keep one log row per drill. A row holds the date, the drill file and its version label, the number of errors seeded, the number caught, the number missed, and any false flags. Each row also says who ran the drill and who signed off.

The trainee fills in their catches first: the page, the row, the category, and a one-line reason for each. Then the senior opens the key, adds anything missed, and writes the category of each miss beside it. The trainee doesn't see the key until their own entries are in.

The log is the record of how the trainee checks, not a score. A drill with a miss isn't a failure. It's a category the trainee needs to see again.

The senior's sign-off on each drill

The senior signs a drill only after comparing the trainee's entries with the key, row by row. The Firm sets the bar. In this plan it's every seeded error caught and every flag explained. A false flag with a clear reason can still pass if the senior agrees with the reasoning.

A drill that misses the bar doesn't get signed. Under this plan, the trainee gets a repeat drill in the missed category, on a different file, before the senior clears that category. The senior signs the repeat when it clears.

Under this plan, the senior signs on the same day as the drill, while the trainee still remembers why they flagged each row. The Firm can choose same-day discussion so the trainee and senior can review a miss with the drill file open.

The sign-off covers that drill and nothing else. It doesn't certify the trainee for live work, and it isn't a sign-off on any return. It's the senior's decision, made against the key, and the tool has no part in it.

A Doe example

The counts below are made up for illustration.

Morgan Doe joins Doe & Co. CPA as a Preparer. Pat Doe, a senior Preparer, builds ten drill files for a fictional client, Doe Holdings, and runs the plan.

In week one, the Day 2, 3 and 4 drills seed 2, 3 and 4 errors. Morgan catches 2, 2 and 4. The Day 3 miss is a missing row, so Pat doesn't sign Day 3. On Day 5, Morgan runs a repeat drill seeded with 2 missing rows and catches both. Pat signs the repeat. Week one: 11 errors seeded, 10 caught.

In week two, the drills seed 5, 5, 0, 6 and 6 errors, a total of 22. Morgan catches all 22. On Day 8, the clean file, Morgan flags one row as a wrong label. The key says the label is right. Morgan's reason shows she read a heading on the page as the row label, so Pat logs it as a false flag, talks it through, and signs.

Across two weeks, the log shows 33 errors seeded and 32 caught, with the one miss cleared by a repeat drill, and one false flag explained. Pat signs Day 10.

After the two weeks

The Firm decides what comes next. One choice is that Morgan starts checking live drafts with a Reviewer reading back every file she checks for a set period. A Preparer or Reviewer still signs off on every live file. The drill files and the signed log stay in the Firm's training file, so the next new hire starts from the same set. The Firm can also rerun a drill or two when it adds a new kind of document to the files the tool drafts from, so the check keeps pace with the work.

Where TracePrep fits

TracePrep's one shipped workflow today is the Reasonable Compensation Study for an S-corporation shareholder-employee. It's a different job from this training plan, and it keeps the same kinds of records:

Join the waitlist

This is general information, not tax or legal advice. Your licensed professional and your Firm's Reviewer are responsible for every return and every conclusion. Keep client data inside the systems your Firm has approved. As general guidance only, not advice: if you use an AI tool with client data, firms generally look for a business or commercial plan whose terms cover how that data is handled; read the vendor's current terms before you do.