# OCR Tax Returns and IRS Notices for a Tax Controversy

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2026-08-04T16:44:22Z
- Updated: 2026-10-11T19:22:20Z
- Category: Practice areas
- Canonical: https://www.schemalaw.com/blog/ocr-tax-returns

> How to OCR tax returns, IRS notices and receipts for an audit, appeal or collection matter, test transcripts for search, and index the file by year and issue.

To OCR tax returns and IRS correspondence for an audit, appeal or collection matter, scan each return and each notice as its own file, convert it to a searchable PDF, and log the tax year, form or notice number and date from every one. Convert schedules and receipts to tables only where you will add up the figures.

The examples use invented taxpayers. Gideon and Marta Albrecht file jointly. Gideon runs a one-truck tree service, and the IRS is examining the business expenses on their 2023 and 2024 returns. They arrive with a banker's box and a shoebox.

Line labels on a return are reliable search keys. For the audit, find adjusted gross income and taxable income on each converted year, then find every deduction, withholding amount, and capital gain on the schedules. The IRS notice names the income tax year. Confirm each figure on the image before it goes into a workpaper for the examination or the appeal.

## Which records make up a tax controversy file?

A tax controversy file is made of what the taxpayer filed, what the IRS sent, what the IRS's own systems show, what backs up the numbers, and what the representative adds. Returns, notices, transcripts and substantiation each arrive in a different form, and each needs different handling once scanned.

| Kind | Examples | Arrives as | First step |
| --- | --- | --- | --- |
| What was filed | Returns with schedules, amended returns, extension requests, information returns the taxpayer received | Preparer's printout, client's photocopy, PDF from tax software | Test for a text layer; OCR the scans |
| What the IRS sent | Notices, letters, document requests, examination reports | Paper in opened envelopes; phone photos | Scan every page, both sides; OCR; log number and date |
| What IRS systems show | Account transcripts, return transcripts, wage and income transcripts | Downloaded files or prints of them | Test for a text layer before doing anything |
| What backs it up | Receipts, invoices, bank and card statements, mileage logs, calendars | Shoebox, envelopes, spiral notebooks | Sort by year and category before scanning |
| What the representative adds | Authorization forms, your letters, proof of mailing, notes of calls | Your own files | Save as created; no OCR needed for native files |

The IRS's page on [burden of proof](https://www.irs.gov/businesses/small-businesses-self-employed/burden-of-proof) explains why the fourth row is the heavy one: "The responsibility to prove entries, deductions, and statements made on your tax returns is known as the burden of proof." In an examination, the agency's [page on IRS audits](https://www.irs.gov/businesses/small-businesses-self-employed/irs-audits) says, "The IRS will provide you with a written request for the specific documents we want to see." That request is the list your scanning plan should be built around.

Returns turn up in other practice areas with a different job to do. As income evidence in a divorce they are covered in the family law post, under [what scanned tax returns and pay stubs give you once converted](https://www.schemalaw.com/blog/family-law-document-ocr#what-do-scanned-tax-returns-and-pay-stubs-give-you-once-converted). The return a debtor gives a trustee is covered in the post on [OCR for bankruptcy documents](https://www.schemalaw.com/blog/ocr-bankruptcy-documents).

## IRS notices and letters: finding the notice number and date on a scan

Find the notice or letter number in a corner of the page. The IRS says, in its guide to [understanding your IRS notice or letter](https://www.irs.gov/individuals/understanding-your-irs-notice-or-letter), "You can find the CP or LTR number on the right corner of the letter." Read the number and the notice date from the image, not from converted text.

Corners are a weak spot on a scan. A notice that was folded in three, stapled, hole-punched or clipped to a return envelope can lose exactly the part of the page where the number sits. Before you rely on a search to find notice numbers across the file, look at the corners of each first page and each last page on screen. If the number is cut off or covered, rescan that sheet flat on the glass.

### What to capture from each notice

Layouts differ from one notice to another, so read the whole first page and the whole last page for these.

| Field | Why it goes in the log |
| --- | --- |
| Notice or letter number | Tells you what kind of notice it is |
| Notice date | The date the lawyer will count from |
| Tax year or period | One box of mail can hold notices for several years |
| Form the notice concerns | Different taxes and penalties are tracked separately |
| Response due date, if one is printed | Goes straight into the calendar |
| Amount stated | Read from the image, digit by digit |
| Contact name, unit, and address or fax for replies | Where your response goes |
| Page count, if the pages are numbered ("Page 1 of 6") | Proves the scan is complete |
| Enclosures listed | Confirms nothing was left in the envelope |

Type those into a log, one row per notice. Do not paste them from converted text without reading the image, because the fields that matter most here are numbers and dates.

The same IRS guide tells taxpayers to "Review it carefully and keep it for your records" and, where a reply is requested, to act by the due date. Which deadline a particular notice starts, and what happens if it passes, is a legal question. The log's job is to make sure no dated notice sits unread in a scanned file.

### Envelopes, duplicates and missing pages

Scan the envelope when the client kept it, since a postmark can matter to the lawyer. Watch for duplicates: the box may hold two copies of the same notice. Keep both scans and mark one as a duplicate in the log instead of deleting it. And compare any printed page count with the pages you have. In the Albrecht box, a letter numbered "Page 1 of 8" has six sheets, and the two missing pages turn out to be the response form the clients had already mailed back.

## Are IRS account transcripts searchable when you download them?

The IRS documentation does not say. Its transcript pages describe what each transcript contains and how to get one, and they do not state a file format or whether the text can be searched. Test each transcript you receive: search for a word you can see, and try to select a single line.

What the IRS does say is useful for knowing what you hold. Its page on [transcript types for individuals and ways to order them](https://www.irs.gov/individuals/transcript-types-for-individuals-and-ways-to-order-them) describes five products. A tax return transcript "shows most line items from your original Form 1040-series tax return as filed, along with any forms and schedules." A tax account transcript "shows basic data such as filing status, taxable income, and payment types. It also shows changes made after you filed your original return." A record of account transcript combines those two, a wage and income transcript "shows data from information returns we receive such as Forms W-2, 1098, 1099, and 5498", and a verification of non-filing letter is the fifth.

The same page says an individual can register for an online account "to view, print, or download all transcript types", can use a mail service, or can submit Form 4506-T. For representatives, the page on the [Transcript Delivery System](https://www.irs.gov/tax-professionals/transcript-delivery-system-tds) says: "Use TDS to view your client's return and account information quickly, in a secure, online session."

So a transcript can reach your file by several routes, and the route decides what you have:

| How it reached the file | What to test |
| --- | --- |
| Downloaded and saved by you or the client | Search for a transaction description you can see |
| Viewed on screen and printed to PDF | Same test; the result depends on how the print was made |
| Mailed by the IRS and scanned | An image; run OCR on a copy |
| Printed by the client, written on, then scanned or photographed | An image with handwriting over it; run OCR and read the notes by eye |
| Faxed from a prior representative | An image, and a faint one; run OCR and check every figure |

### If the transcript needs OCR

An account transcript is a list of dated transactions and amounts. Three checks apply after conversion:

- **Codes.** Where a transaction line carries a numeric code, a misread code changes what the line means. Read every code against the image.
- **Dates.** Check the format the transcript prints and keep it consistent in any spreadsheet you build.
- **Signs.** Credits may print with a minus sign or in a separate column. Confirm that a faint minus was not dropped.

A transcript is also not a copy of the return. If you need the return itself, the IRS's page [about Form 4506](https://www.irs.gov/forms-pubs/about-form-4506) describes that form as the way to "Request a copy of your tax return", and points to the transcript service for cases "when you don't need an actual copy of your original tax return."

## Converting prior-year tax returns and their schedules

Convert a prior-year return by scanning the complete copy in form order, running OCR to a searchable PDF, and typing the handful of lines you need into a comparison sheet from the image. Export schedules to a spreadsheet only when you need to total or compare their rows. Form pages are hard on table detection.

### Before scanning

1. **Find out which copy you have.** A preparer's packet can include a client copy, a filing copy, worksheets and state returns. Look for words such as "client copy" or "do not file" printed across the page, and note them.
2. **Put the pages in order.** Main form, then numbered schedules, then other forms, then attached statements. Count the pages.
3. **Remove staples and flatten folds.** A return kept in a drawer for years may have both.
4. **Scan both sides.** Some forms print on the back.
5. **Name the file by taxpayer, year and form.** For example `Albrecht 2023 Form 1040 as filed.pdf`.

### What OCR does with a form page

A return page is a grid of line numbers, descriptions and amount boxes. Recognition reads the printed labels well when the copy is clean. The trouble is in pairing each amount with its line. A converter may output the left column of descriptions as one run and the right column of amounts as another, or attach an amount to the line beside it. That is why the figures you rely on get typed from the image, with the line number, and not copied out of converted text. Checkboxes, such as filing status, come through as a printed box and a label with no record of which was ticked. The post on [OCR for scanned forms](https://www.schemalaw.com/blog/ocr-scanned-forms) covers fields and checkboxes.

Other things to look for on a scanned return:

- **Amounts split across boxes.** Some forms print dollars and cents in separate boxes, which can convert as two numbers.
- **Losses.** A negative figure may be shown in parentheses or with a minus sign. Check that the sign survived.
- **"See statement".** A line that refers to an attached statement holds no number of its own. The statement is at the back of the return and has to be in your scan.
- **Handwritten returns and handwritten changes.** Recognition is weaker on handwriting than on print. Read these by eye.
- **Identifying numbers.** Return pages carry Social Security or employer identification numbers. The post on [OCR for documents with Social Security numbers](https://www.schemalaw.com/blog/ocr-documents-with-social-security-numbers) covers what to do with converted copies.

### A year-over-year sheet

For the lines in dispute, build a small table with one column per year and one row per line. For the Albrecht examination, which concerns the business schedule:

| Line as labeled on the form | 2022 | 2023 | 2024 | Source pages |
| --- | --- | --- | --- | --- |
| Gross receipts | 148,210 | 151,940 | 139,675 | p. 7 of each return |
| Car and truck expenses | 9,120 | 21,480 | 22,035 | p. 7 |
| Supplies | 11,300 | 12,050 | 10,870 | p. 7 |
| Repairs and maintenance | 2,410 | 2,980 | 9,640 | p. 7 |

Laid out this way, two things stand out: car and truck expenses more than doubled from 2022 to 2023, and repairs more than tripled in 2024. Those are the categories where the substantiation needs to be most complete. The table says nothing about whether the deductions are correct. It tells you where to look.

Schema Law reads PDF, PNG or JPEG uploads of up to 20 MB into text and tables and shows the parsed pages on screen, which is where to compare a form page with its conversion before exporting. Searchable PDF is included on every plan. The [pricing page](https://www.schemalaw.com/pricing) lists the page limit per document for each.

## Receipts, mileage logs and other substantiation records

Substantiation records are the taxpayer's proof, and they are the worst paper in the file: small, faded, handwritten or crumpled. Scan them in groups by year and expense category, convert them so they can be searched, and build the schedule of amounts by typing from the images, not by trusting recognized totals.

The IRS states the standard on its burden of proof page: "You generally must have documentary evidence, such as receipts, canceled checks, or bills, to support your expenses", and "Additional evidence is required for travel, entertainment, gifts, and auto expenses." Its topic page on [business use of a car](https://www.irs.gov/taxtopics/tc510) adds: "The law requires that you substantiate your expenses by adequate records or by sufficient evidence to support your own statement." What counts as adequate for a given deduction is a legal question. On the document side, the audits page says the IRS accepts "some electronic records", so ask the examiner what form they want before you prepare a submission.

### Receipts

- **Thermal paper fades.** Scan faded receipts in grayscale or color, not black and white, so that pale print is not thrown away. The post on [OCR for faded text](https://www.schemalaw.com/blog/ocr-faded-text) explains why.
- **One receipt per image where you can.** Six receipts taped to a sheet can convert as one jumble of lines. If the client already taped them, leave them and type the amounts.
- **Capture four things.** Date, payee, amount and what was bought. Add the page reference.
- **Match to the bank or card statement.** A receipt shows what was bought and a statement shows that it was paid. Put the statement page beside the receipt page in your schedule.

### Mileage logs

Mr. Albrecht's mileage log is a spiral notebook kept in the truck: a date, a customer name and odometer readings in pencil, one line per job. That is handwriting in columns, and the post on [handwritten ledgers to Excel](https://www.schemalaw.com/blog/handwritten-ledger-to-excel) covers the method. Three points are specific to a log:

1. Scan the whole notebook, including the cover and blank pages, so nobody can say pages were left out.
2. Type every line. For each, compute miles as the ending reading minus the starting reading and compare with any miles figure written on the line.
3. Check that each day's starting odometer is no lower than the prior entry's ending reading. A reading that goes backwards is a misread digit or a problem with the log, and you need to know which before the examiner asks.

### Calendars, invoices and everything else

Appointment calendars, customer invoices and supplier statements support the log and the receipts. Convert them to searchable PDF so that a customer name or a date written in the log can be found across all of them.

## How do you index a tax controversy file by year and issue?

Index the file with one row per document and two sorting keys: the tax year and the issue the document speaks to. Add the document type, its date, the file name and page count. Then every question in the case, such as "what supports 2023 truck expenses?", is a filter and not a search through folders.

A document can concern several years and several issues. Give it one row per combination, or use a column per issue with a mark in each that applies. Either works. What fails is an index sorted by date received, which scatters each year's proof across the whole list.

| Year | Issue | Type | Document date | Description | File | Pages |
| --- | --- | --- | --- | --- | --- | --- |
| 2023 | All | Return | 9 Apr 2024 | Joint return as filed, preparer's client copy | Albrecht 2023 Form 1040 as filed.pdf | 31 |
| 2023 | Car and truck | Notice | 12 May 2026 | Document request, items 1 to 4 | IRS 2026-05-12 document request.pdf | 5 |
| 2023 | Car and truck | Substantiation | Jan to Dec 2023 | Mileage notebook, scanned whole | 2023 mileage log.pdf | 84 |
| 2023 | Car and truck | Substantiation | Various | Fuel receipts with card statement pages | 2023 fuel.pdf | 112 |
| 2024 | Repairs | Substantiation | 3 Aug 2024 | Invoice, transmission rebuild | 2024 repairs.pdf, p. 14 | 2 |
| 2023, 2024 | All | Transcript | 20 May 2026 | Account transcripts, downloaded | Transcripts 2023-2024.pdf | 6 |

### Building it

1. Start from the IRS's written document request. Each numbered item becomes an issue label.
2. Add a row for every notice in your notice log.
3. Add the returns and transcripts for the years in question and for the year before, which gives the comparison.
4. Add substantiation files as you scan them, already grouped by year and category.
5. For each requested item, filter the index and read what you have. An empty filter is a gap to raise with the client now.
6. Record what was sent to the IRS, on what date and by what method, in two more columns.

Because every file in the index is searchable, the index does not have to describe each page. A search across the folder for a customer's name, a payee or an amount will find the page, and the index tells you which year and issue it belongs to.

One more column earns its place: "checked against image by", with initials and a date. A figure that goes into a letter to the IRS should come from a page someone has read, and the column shows who.

## Try it on a real document

Upload a scanned PDF and export it as Word, a searchable PDF or plain text on the free plan. Your documents are never used to train an AI model. [Start free](https://www.schemalaw.com/api/auth/google)