Tax Preparation

When a Source Page Can't Be Read: Poor Scans and the AI Pass

Bobby Huang8 min read
Jump to a section

Key takeaways

  • The Firm sets the test: a page passes only if a person can read every figure the file needs from it, digit by digit, without guessing.
  • The Firm can run the screen page by page, after folder setup and before the first instruction.
  • Under this approach, each marked page goes on the file index with the reason, the path chosen and who screened it; a blank reads as not screened.
  • Under the Firm's rule, marked pages stay out of the drafting pass, and the Preparer checks that every drafted row cites a page screened as readable.
  • The Firm decides whether a marked page waits for a clean copy or has its readable figures keyed by hand, each with its page reference, the mark and a reason.
  • Under the Firm's rule, nobody keys a figure they can't read with confidence; it waits for a clean copy or goes to the Reviewer with a note.

In this workflow, the Firm chooses which pages enter the AI drafting pass, and the Preparer checks each drafted row against its cited page. That check needs a page a person can read. If the edge is cut off, the print is blurred, or a total is written by hand in a cramped corner, the Preparer has nothing firm to hold the row against.

The Firm can choose the following workflow for handling unreadable source pages before the AI pass. The Firm can add a short screen: a person looks at every page, asks whether they can read the figures on it with confidence, and marks the pages where the answer is no. Those pages stay out of the drafting pass. Each marked page either waits for a clean copy or has only its readable figures keyed by hand, with a "keyed by hand" mark. Under this approach, the Firm supplies only screened, readable pages to the drafting pass, and the Preparer records a page reference for each drafted row before review.

What counts as a page you can't read

The Firm sets the test. One workable test: can a person read every figure the file needs from this page, digit by digit, without guessing? If any needed figure fails, the page gets marked.

Four things to look for:

Cut-off edges. A scan or photo that drops a margin, a column of amounts, or the bottom line where the total sits.

Blur. A figure where a 3 and an 8, or a 1 and a 7, can't be told apart.

Handwriting. A handwritten log, a note in the margin, or a figure written over a printed one.

Photos at an angle. A page shot on a desk, with lines that slope and figures that shrink toward one side.

The test is about the person, not the tool. The question is never how a tool would treat the page. It's whether the Preparer, and later the Reviewer, could check a row against it and be sure of what they're looking at.

Screen the pages before the pass

The screen fits between folder setup and the first instruction. Setting up the client folder addresses names, duplicates and versions; this separate screen addresses page legibility.

The Firm decides who screens. One approach: the Preparer who'll run the pass screens every page, and a second person looks again at any page the first wasn't sure about. Screen page by page, not file by file. A ten-page statement with one blurred page has nine readable pages and one marked one.

Look at each page at the size you'd use to check a row, not as a thumbnail. Read the figures the file needs: amounts, dates, totals, account numbers, form identifiers. If one needed figure can't be read with confidence, mark the page, even when the rest is sharp.

Mark each page on the file index

The file index already has one row per file. The Firm can add a page-quality note for each marked page, with four entries:

  • the page number in the file;
  • what makes it unreadable: cut off, blurred, handwritten, or at an angle;
  • the path chosen: clean copy requested, or keyed by hand;
  • who screened it, and the date.

A file with no marked pages gets "all pages screened, readable" and the screener's initials. A blank reads as not screened. The file index records who screened each page, so the Reviewer can tell a page that passed from a page nobody looked at.

Keep marked pages out of the drafting pass

Under the Firm's rule, a marked page doesn't go into the drafting pass. Set it aside from the pages the pass runs on, and note on the index that it was held back. The Preparer then checks that every drafted row cites a page screened as readable.

The reason is the check. Every drafted row gets checked against its page. A row drafted from a page nobody can read with confidence can't be checked that way, so the Firm keeps such pages out of the pass.

The marked pages stay in the client's file. Holding a page back from the pass doesn't remove it; it only means its figures reach the entry list another way, or not yet.

Two paths for a marked page

Each marked page takes one of two paths, and the Firm decides which one applies.

Request a clean copy. If the Firm chooses the clean-copy path, the item joins its one client request built from a checked list, while the page stays open on the index.

Key the figures by hand. Some marked pages still hold the needed figures in a part a person can read with care: a handwritten log written clearly, or a photo where the slope bends the labels but the amount column is sharp. The Firm can allow a person to key those figures by hand. Each keyed row carries its page reference. It also carries the mark "keyed by hand" and the initials of the person who keyed it. The Preparer writes the reason on each keyed row, such as "photo at an angle, amount column clear".

One limit sits under both paths. Under the Firm's rule, nobody keys a figure they can't read with confidence. A digit that might be a 3 or an 8 waits for the clean copy. An open page goes to the Reviewer with a note if no clean copy comes, and the entry list doesn't show that figure as settled.

When the clean copy arrives

Screen the clean copy like any other page. If it passes, update the index: the date it arrived, the page it replaces, and who screened it. Then run the pass on the new page, or key it, under the same rules as the rest of the file. The marked original stays in the file beside it.

A Doe example

The counts below are made up for illustration.

Jane Doe, a Preparer at Doe & Co. CPA, screens the folder for the client Doe Holdings before the first pass. The folder holds 5 files and 14 pages.

She marks 3 pages. Page 3 of a bank statement was photographed at an angle, and the right edge, where the page total sits, is cut off. Page 1 of a handwritten expense log has an unreadable date the file needs, but all 4 amounts can be read. Page 2 of the same log is a faint photocopy with 5 amounts; she can read 4, but one digit in the fifth is unclear.

She records all three pages on the file index with the reason and her initials. The drafting pass runs on the other 11 pages.

She takes the clean-copy path for the statement page and for the unclear amount on page 2 of the log. She keys the 4 amounts from page 1 of the log by hand, then the 4 she can read on page 2, each with its page reference, the mark and a reason. That's 8 keyed rows. The fifth amount on page 2 and the date on page 1 stay open, each with a note to Pat Doe, the Reviewer.

Both clean copies arrive. Jane screens them and marks them readable. She runs the pass on the statement page, so 12 pages have now been through it, and she keys the open amount from the clearer log page as the ninth keyed row.

Pat reads the 9 keyed rows against their pages, sees both clean copies logged on the index and the remaining unreadable date flagged as open, and decides whether to sign off.

Where TracePrep fits

We've built a TracePrep plugin that drafts these steps in your Firm's workflow, while your preparer decides and your Reviewer signs off. Join the waitlist and we'll tell you when plugin access opens.

Get on the list for plugin access

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.

Frequently asked questions

Who screens the pages?
The Firm decides. This piece assumes the Preparer who runs the pass screens every page, and a second person looks again at any page the first wasn't sure about. Under this approach, whoever screens puts their initials on the file index.
If a page passes the screen, are the drafted rows from it right?
Not on that basis. In this workflow, the screen only decides which pages go into the pass. Under this approach, the Preparer still checks every drafted row from a readable page against its page before the entry list moves on.
What if only one figure on a page is unclear?
Under this approach the page is marked, because one needed figure fails the test. The Firm can still let a person key the readable figures by hand, each with the mark, and leave the unclear one open until a clean copy arrives.
Can a person key a figure they're fairly sure of?
Under the Firm's rule, no. Under that rule, fairly sure isn't reading with confidence. The Firm can have the figure wait for a clean copy and, if none comes, send it to the Reviewer with a note.
What if a clean copy never comes?
The Firm decides what happens to that page. Under this approach, the page stays open on the file index, the Reviewer sees it with a note, and any unreadable figure remains unsettled on the entry list.

Share this article

Bobby Huang · founder of TracePrep Inc.

Bobby Huang is the founder of TracePrep Inc. TracePrep is a tax-prep platform for CPA, EA, and accounting firms, and it also produces the Reasonable Compensation Study.