Prior-Year Comparison With AI: What Changed in the Client's File

Last period's final file shows what this client's file looked like when it was finished: which documents came in, from which payers and accounts, and which pages the workpapers cite. This period's folder is the same client, partway through. Compare the two copies to draft a starting list of differences.

Ask an AI tool to draft the list, citing the matching documents by file and page and marking the side with no match as "none." Decide what each difference means yourself. A missing statement might be a closed account, a client who hasn't sent it yet, or a file saved under an odd name. That call belongs to the Preparer, one row at a time.

Pulling figures and tying them to their pages has its own walkthrough; this one stays on the comparison.

Set up two copies

You need two folders. The first is last period's final workpaper folder: the version the Reviewer signed off on, not a working draft or a midway export. If the Firm keeps drafts beside the final, pull only the final set. The second is this period's client folder as it stands today.

Copy both. Here's why, in the vendors' words. Anthropic's help article for Claude Cowork says: "On desktop, Claude can read from and write to your local files without manual uploads or downloads." OpenAI's ChatGPT quickstart says: "ChatGPT can read and modify files in the folder you choose." Both statements describe a tool that can change files as well as read them. Neither quote mentions comparing two periods, so try that on your copies first.

Point the tool at the two copies and nothing else. The originals stay as the record: last period's file as finished, this period's as the client sent it. The tool and setup must be ones the Firm has approved for client data.

Before the comparison, have the tool name every file it opened in each copy, the period printed on it, and any page it couldn't read. Compare the reported list with each folder's listing, inspect omitted or unreadable files yourself, and leave affected rows unresolved until those files are checked. A skipped file can make a document look absent when it's sitting right there.

The instruction

Tell the tool to match documents across the two folders by what each document is and which account or payer it belongs to, not by file name.

Ask for one row per difference, using four row types:

• Only last period: a document in last period's final folder with no match in this period's folder.
• Only this period: the reverse.
• Changed payer or account: a document of the same kind that now names a different payer, employer, institution, or account.
• Same source, different figure: a matched document where a figure the workpaper cited last period reads differently now.

Require each row to cite the file name and page in both folders, or say "none" for the folder with no match. Require a period label on every quoted figure, and check each label against the source page.

Add three rules. Never net, total, or explain a change; quote the figures as they appear and stop there. Never guess a match: if the pages don't settle whether two documents are the same source, return both as "only" rows flagged "no match found" and say why. And list separately any file the tool couldn't classify.

A "same source, different figure" row doesn't say anything is wrong. Its job is to put both pages in front of you.

Check the draft by hand

Open every available cited page for each row; for a side marked "none," inspect that folder's file list and document contents before confirming the absence.

For each row, confirm the document is what the tool says it is, for the right person or entity, account, and period. Two cases deserve a slower look. The first is a prior-period document sitting in this period's folder. Check the printed period on each page against the period recorded in the file list, and flag any prior-period document found in this period's folder. The second is a corrected document that replaces an earlier one. Find out which version each folder holds and record both. The failure-modes piece shows how to test a tool for the wrong-version problem on synthetic documents first.

Then mark each row one of three ways:

• Confirmed: the available cited pages support the row, and any absence has passed your manual folder check.
• Wrong: the match, the type, or a figure doesn't hold up. Note what's actually on the page.
• Needs a client question: the pages check out, but you can't tell what the difference means without asking.

Read the "no match found" rows and the unclassified files too. If you can pair two of them by hand, do it and note how.

What to do with each row type

These are steps for the Preparer's file, not conclusions about tax treatment. Fit them to the Firm's procedures.

Only last period. Check whether the engagement notes or intake answers already explain the absence, and record what you find either way. If nothing explains it, put the item on the client request. The missing-items walkthrough turns those rows into one plain request.

Only this period. Note the document and its page in the file, with anything the engagement notes say about it. If they say nothing, mark it as a client question.

Changed payer or account. Note the old and new names with both pages. If the engagement notes don't mention the change, mark it as a client question.

Same source, different figure. Write a short note naming the figure, both pages, and what changed, with no conclusion about why.

The Firm's procedures set which rows need a note for the Reviewer. The list feeds the client request and the Reviewer's questions.

A worked example

Jane Doe is a made-up returning client who owns a made-up entity, Doe Holdings. Alex Doe is the Preparer. After the hand check, part of Alex's list looks like this:

Each note says what's on the page and what's still open, never what a change means.

Keep the checked list with the workpaper

File it with this period's workpaper: the two file lists, the tool's draft, your mark on each row, and your notes. The Reviewer can then see what changed and open the pages behind each row. For the Reviewer's side, see for reviewers.

Whether to use a tool's pass here, and which checks to require, is the Firm's call. The Reviewer concludes and signs off on the workpaper.

The prior-period checklist

1. Pull last period's final workpaper folder, not drafts, and this period's client folder.
2. Copy both and point an approved tool at the copies only.
3. Compare the tool's list of files read with each folder's file list.
4. Match documents by what they are and which account or payer they belong to.
5. Ask for four row types, with file-and-page citations for available documents, "none" for an unmatched side, and every figure labeled by period.
6. Forbid netting, totaling, explaining, and guessed matches.
7. Open every available cited page, check printed periods and corrected versions, and manually check the relevant folder for each side marked "none."
8. Mark each row confirmed, wrong, or needs a client question.
9. Send unexplained absences to the missing-items request and note the rest.
10. Keep the lists, draft, marks, and notes with the workpaper.

Where TracePrep fits

In TracePrep's Reasonable Compensation Study, every figure traces to its source, occupation, geography, and period, at field level. A Reviewer signs off on each Study, and the audit log records who prepared it, who changed what, and who approved it. Finalized evidence stays with the Firm.

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.

Join the waitlist