OCR
OCR, optical character recognition, is the generation of machine-readable text from a page image by recognizing the shapes of the characters on it. In eDiscovery it supplies text for documents that have none of their own: scanned paper, image-only PDFs, photographs of documents, faxes, and pages whose original text was destroyed by redaction burn-in.
Also called: optical character recognition · OCR text · searchable PDF · image-to-text · text recognition
Updated September 6, 2026
How does OCR work?
Processing tries to extract text from every file. Native electronic documents have text inside them and give it up exactly. Images do not: a scan is a picture of a page, and a search index built over it would find nothing. OCR fills that gap: the image is analyzed, character shapes are recognized, and a text file is written alongside.
Three places in the workflow need it. Scanned paper and image-only PDFs at ingestion, without which those documents are invisible to every search term, every cull and every TAR classifier. Redacted documents at production, because burn-in destroys the underlying text. And documents produced as images by the other side, which arrive with whatever text the producing party generated, or none.
OCR output is never treated as the document. It is a search aid: good enough to find a page, not authoritative as to what the page says. Where accuracy matters, the image governs.
The practical consequence is that a collection’s OCR quality sets a ceiling on how well every downstream step works. A search term that would have found a document does not find it if the OCR read “Apex” as “Ap ex”, or “invoice” as “lnvoice”.
What does OCR look like in practice?
Volume, not accuracy, is what makes OCR a cost question. A gigabyte of scanned images is roughly 15,500 pages; a gigabyte of plain text is closer to 678,000. Both figures come from the pages-per-gigabyte tables the volume translator uses, drawn from published analysis of about 2,000 hosted matters.
A per-page OCR charge on a 40 GB scanned collection is therefore priced against roughly 620,000 pages, while the same 40 GB of email is a much smaller page count. Per-page pricing and per-gigabyte volume measure different things, which is why the two never reconcile on an invoice. (Pages per gigabyte are rules of thumb; image resolution and page density move them substantially.)
How does OCR show up on an eDiscovery invoice?
OCR and imaging is its own category in the Fee Field Guide, priced per page or per gigabyte, and it commonly appears alongside, not inside, per-GB processing. Where a collection is largely scanned material, it can be the largest processing-side line.
It also recurs at production. Every redacted page is re-OCR’d after burn-in, which some providers bill as new OCR against the production volume, on top of format conversion and endorsement charges for the same page.
Ask: is OCR included in processing; is it billed per page or per GB; is it run on everything or only on files with no extractable text; and is re-OCR after redaction charged separately. Where the protocol allows native production, a large part of the imaging and OCR charge simply does not arise.
Sources
- EDRM, Processing Standards on text extraction and OCR.
- The Sedona Conference, Glossary: eDiscovery & Digital Information Management, “optical character recognition.”
- Pages-per-gigabyte figures:
src/data/volume.json, from published analysis of ~2,000 hosted matters and industry page-per-GB tables. - Fee structure: the Fee Field Guide dataset.
Related terms
TIFF production · Redaction · Processing · Document production · Metadata · Culling
eDiscovery at one flat monthly rate
$3,250 a month on a 12-month agreement. No per-user fees and no per-GB hosting charges.
Common questions
Is OCR text as accurate as extracted text?
No, and the gap is the whole reason the distinction is drawn. Extracted text is copied out of the electronic file and is exact. OCR text is a machine's reading of an image and contains errors, which vary with scan quality, skew, handwriting, tables, stamps and font. Clean 300 DPI office print recognizes well; a faxed, stapled, handwritten-annotated page does not.
Do redacted documents need OCR?
Yes. Burning a redaction into a page image destroys the text underneath, so the original extracted text can no longer be produced with that document; it would disclose exactly what was redacted. The redacted image is re-OCR'd and the OCR text produced in place of the extracted text. Skipping this step is one of the most common ways redacted material leaks.
Is OCR included in processing fees?
Sometimes. It is frequently broken out as a separate per-page or per-GB line, which the Fee Field Guide notes as a holdover from per-page photocopying pricing. Ask directly whether OCR is included in processing, whether it is charged per page or per gigabyte, and whether the charge applies to the whole collection or only to files with no extractable text.