# How to Redact Documents Consistently Across Multiple Files > Use shared detection settings across a batch, build a cross-document entity registry, and run a QA sample review before finalizing any large production set. - **Author:** Neetusha - **Published:** 2026-06-22 - **URL:** https://www.redactifyai.com/answers/how-to-redact-multiple-files-consistently/ --- Inconsistent redaction across a document set is one of the most common sources of discovery sanctions. If a name appears redacted in document A but unredacted in document B from the same production, opposing counsel will notice, and courts have held that inconsistent redaction can waive privilege or indicate inadequate review. Four practices eliminate most consistency failures. ## 1. Lock down your detection settings before processing begins Before running any files, define a shared entity type checklist for the batch: which categories must be redacted (names, SSNs, dates of birth, account numbers, medical record numbers, etc.) and which should not be (company names in headers, dates that are not birth dates, etc.). Write this down. Apply the same settings to every file in the batch. Changing settings mid-batch, for example adding phone numbers after processing half the files, creates a two-tier production where early files are inconsistently redacted. ## 2. Batch-process all files together Processing files one at a time with manual settings each time introduces human error. Batch processing applies the same rule set uniformly across every document in the queue. This eliminates the risk of forgetting a category on file 37 of 200. It also speeds up the job significantly, because the tool loads the detection model once and runs it across all files. ## 3. Build a cross-document entity registry If a person's name, account number, or medical record number appears in file 1, it must be redacted every time it appears in files 2 through 200. A cross-document entity registry is a list of confirmed sensitive values that the tool applies globally across the batch. Some tools call this "find and redact across files." Without it, a name that the NER model catches in one document may slip through in another where it appears in an unusual format, like a table header or a footer. ## 4. QA sample before finalizing No automated tool is perfect. Before certifying a production, pull a random sample, typically 5-10% of the batch or at least 20 documents, and manually review each one for missed identifiers. Pay special attention to headers, footers, tables, captions, signature blocks, and exhibits, because these areas have different formatting from body text and are the most common failure points. Document your QA process; this becomes your evidence of reasonable care if the production is later challenged. [FRCP 26](https://www.law.cornell.edu/rules/frcp/rule_26) governs the discovery process and requires parties to produce documents in a reasonably usable form with privilege protections intact. The [Sedona Conference](https://thesedonaconference.org/), whose commentary courts frequently cite in discovery disputes, has consistently emphasized that parties must take proportionate and consistent steps to protect privileged and sensitive information throughout a production. RedactifyAI supports batch processing across PDF, Word, and Excel files, applies a shared entity detection configuration to every document in the queue, and generates an audit trail that records which entity types were redacted in each file. This creates a uniform, documented record across the entire production set.