Optical character recognition is what turns a scanned exhibit into a filing a court can actually search. Most federal courts, and a growing number of state courts, require electronically filed PDFs to be text-searchable, which means any document that started life on paper needs OCR processing before it hits the docket. Using OCR for court filings is a short technical workflow with a few places it can go wrong, and the consequences of getting it wrong range from a bounced filing to an accidental disclosure of information you thought you had redacted.
Why Courts Require Searchable PDFs
Federal courts require represented parties to file electronically through the Case Management/Electronic Case Files (CM/ECF) system.1United States Courts. Electronic Filing (CM/ECF) The electronic filing mandate itself comes from Federal Rule of Appellate Procedure 25(a)(2)(B), but the text-searchability requirement lives in local rules and administrative orders issued by individual district and appellate courts.2United States Courts. Federal Rules of Appellate Procedure – Rule 25 Many of those local rules require every PDF to be text-searchable, attachments and exhibits included where feasible.
The reason is practical. Clerks and judges rely on keyword searches to find information in the record, verify cited facts, and build the docket index. A 500-page exhibit without OCR is a stack of pictures to the court’s systems; nothing in it is findable.
When a filing fails the searchability standard, the clerk’s office usually rejects it and sends it back for a compliant resubmission. That cycle is dangerous close to a deadline. A late refiling after rejection can be deemed untimely, and in the worst case that means a missed statute of limitations or a dismissed appeal. Build in time for at least one round of rejection and refiling before any filing deadline.
When You Need OCR and When You Don’t
A native PDF, one created directly from a word processor or other software, already contains real text data. You can highlight it, copy it, and search it. No OCR is needed because the text was digital from the start.
A scanned PDF is a different animal. The output of a scanner is an image file: each page is a photograph of text, not actual text. On screen it may look identical to a native PDF, but the court’s search function will find nothing in it. These scanned images need OCR to generate a hidden text layer behind the visible image, which preserves the original look while making the content searchable.
A useful rule of thumb: if you created the document on a computer, save it directly as a PDF rather than printing and rescanning it. Printing to paper and scanning back in costs you quality and manufactures an OCR dependency that never had to exist. Reserve OCR for documents that only exist on paper, such as signed contracts, historical records, and opposing party productions that arrived as image files.
Scan Settings That Produce Clean OCR
Specific requirements vary by court, but a few technical standards hold up across most jurisdictions:
- 300 DPI resolution. This is the widely accepted minimum for producing characters clean enough for reliable recognition. Lower resolutions leave the software guessing, and in legal documents a misread “8” as “B” or “1” as “l” is not a harmless error.
- Black and white. Courts generally expect black and white scans unless color carries evidentiary weight, as with photographs or highlighted annotations.
- PDF/A format. Many courts require or prefer PDF/A, an ISO-standardized archival format (ISO 19005) designed to keep documents readable years into the future regardless of the software used to open them. PDF/A restricts features like encryption and external font dependencies that could otherwise render a file unreadable later.3Library of Congress. PDF/A Family, PDF for Long-term Preservation
Check your specific court’s local rules before filing. Some add requirements around file naming, page orientation, or color depth, and a technical rejection costs you the same time as a substantive one.
Running OCR and Verifying the Output
The process itself is not complicated. Professional software such as Adobe Acrobat Pro or ABBYY FineReader handles it: import the scanned PDF, select the text recognition function, and let the software generate the hidden text layer.
Set the recognition language to English for domestic filings. The software uses that setting to decide which character patterns to expect, and a mismatch tanks accuracy. If an exhibit contains passages in another language, some tools let you specify multiple languages or process sections separately.
After processing, verify the result. Open the file, search for a few words you can see on the page, and confirm they come up. Copy a paragraph into a text editor and look for garbled characters. On clean typed pages, OCR routinely clears 99 percent accuracy, but that number falls hard on degraded originals, faded ink, unusual fonts, creased pages, or heavy background noise. For handwritten text, current OCR is unreliable enough that you should not assume it will be searchable at all. Transcribe handwritten portions separately and attach the transcription alongside the original.
Watch file size. Adding a text layer increases the size, sometimes a lot on large exhibits, and most courts cap uploads through their filing systems. Scan at 300 DPI rather than a higher default, and use your software’s PDF optimization tools to compress the output without dropping below that clarity threshold.
Redact Before You Run OCR
This is the step that catches filers off guard. The hidden text layer OCR produces contains a machine-readable copy of everything on the page, including information that looks redacted in the visible image. Black out a Social Security number on the image, then run OCR on that image, and the software may still read the underlying characters and embed them in the text layer. Anyone who copies text from the blacked-out area recovers the information.
Federal Rule of Civil Procedure 5.2 requires filers to redact specific personal identifiers from court filings: all but the last four digits of Social Security and financial account numbers, only the birth year rather than the full date of birth, and only initials for minors’ names.4Legal Information Institute. Federal Rules of Civil Procedure Rule 5.2 – Privacy Protection For Filings Made with the Court Redaction is the filer’s responsibility, not the clerk’s. And because most paper filings get scanned into the electronic case file accessible through PACER, even a paper filing eventually becomes a digital document where a hidden text layer can expose supposedly redacted content.5Public Access to Court Electronic Records. PACER Federal Court Records
The correct order is redact first, OCR second. Apply redactions using a tool that permanently removes the underlying content rather than just covering it with a black box. Then sanitize the document to strip metadata, hidden layers, comments, and embedded data. Only after that should you run OCR on the redacted version.
If you have to OCR first because you need the text layer to locate the sensitive content, re-run the redaction tool on the processed file and confirm both the visible image and the hidden text layer have been scrubbed. Flatten or rasterize the final output as a belt-and-suspenders step: it merges all layers into a single image before a fresh text layer is generated from the sanitized content.
A filer who files unredacted information waives Rule 5.2’s protections for that information, and there is no reliable take-back once a document has been accessible on PACER, even briefly.4Legal Information Institute. Federal Rules of Civil Procedure Rule 5.2 – Privacy Protection For Filings Made with the Court
Bates Numbering in the Right Order
Bates stamping is standard for litigation exhibits, and it has a specific conflict with OCR. General-purpose PDF tools often apply Bates numbers as a text overlay on top of an image file. That overlay can confuse OCR engines, which may misread the stamp as part of the document text or skip OCR entirely on the affected pages.
The safer sequence is to apply Bates numbers first, then run OCR on the full document so the engine treats the stamp as part of the page image. Dedicated e-discovery platforms handle Bates stamping as part of their export workflow and preserve native file integrity while producing properly numbered copies. If you are working with a large volume of exhibits, that tooling tends to pay for itself in avoided rework.
Bookmarks and Hyperlinks for Navigable Filings
Searchability is the floor. Courts increasingly expect navigation features that make large filings usable. Bookmarks let a reader jump to specific sections, and internal hyperlinks let a brief point directly to the cited page of an attached exhibit.
If you create the filing in a word processor, generate a hyperlinked table of contents before converting to PDF. Those links carry over as PDF bookmarks. Use “Save As PDF” or “Create PDF” rather than “Print to PDF”; the print function flattens the document to an image and strips every hyperlink and bookmark in the process.
For multi-exhibit filings, add manual bookmarks for each exhibit tab in your PDF software. Name them descriptively (“Exhibit A – Purchase Agreement”) rather than generically. If the local rules call for cover sheets or separator pages, integrate those into the PDF and bookmark them as well.
Common OCR Failures to Catch Before Filing
Most OCR problems fall into a few predictable categories. Knowing them in advance saves you from a rejection or, worse, an exhibit whose text layer misrepresents what the visible page says.
- Low-resolution scans. Scanning below 300 DPI is the single most common cause of poor results. Characters become ambiguous and the software guesses wrong.
- Background noise and artifacts. Coffee stains, paper texture, fax artifacts, and creased pages can read as characters. Borders and text boxes are particularly bad: vertical lines routinely become “l” or “1.”
- Mixed content pages. Pages combining typed text with handwritten annotations, stamps, or signatures confuse OCR engines, which try to read everything and produce garbage where the handwriting is. Where possible, separate the content or flag those pages as partially non-searchable.
- Skewed pages. Documents fed through the scanner at a slight angle recognize poorly. Most OCR software includes a deskew function that straightens the image before processing; use it.
- Overlapping text layers. Running OCR on a file that already has a text layer, such as a native PDF that was printed and rescanned, creates duplicate or conflicting data. Searches return garbled results or miss content. Before processing, check whether the file already contains searchable text.
Whatever your source material, test the output. Search for specific terms, copy passages, and spot-check pages with complex formatting. A few minutes of verification prevents a filing rejection and the embarrassment of opposing counsel pointing out that your exhibit’s hidden text says something different from what the page shows.
Filing Through CM/ECF
Federal practitioners file through CM/ECF, which requires a linked PACER account. After a successful upload, the system generates a Notice of Electronic Filing (NEF) that serves as the official confirmation of your submission. The NEF logs the date and time, links to the filed document, and constitutes service on all registered CM/ECF users in the case. Once you have your NEF, filing and service are complete as to electronic participants.
Court clerks review submissions for compliance before accepting them into the docket. A filing that clears review becomes part of the permanent electronic record, available to authorized parties and, for most documents, to the public through PACER. A filing that fails review for searchability or formatting bounces back for correction. That round trip is the main reason the OCR step is worth doing carefully the first time.