What is Bates numbering? The complete guide for litigation teams

A Bates number is an address. It names 1 page in 1 production, and it keeps naming that page for the life of the case. Page numbers restart in every document and shift when someone re-sorts a PDF. A Bates number does not move. That is the whole point of the system, and every rule in this guide exists to protect it.

The name comes from Edwin G. Bates, whose hand stamp printed a number on a sheet of paper and advanced to the next number by itself. The Henry Ford museum, which holds a Bates numbering machine made between 1919 and 1925, describes its mechanism as an improvement patented by Bates in 1891. Law offices ran those stamps over paper productions for most of the 20th century. Software does the stamping now, but the idea has not changed: 1 page, 1 number, forever.

People use 4 terms loosely. They mean different things, and the difference shows up when a deficiency letter lands.

Bates number

The identifier itself, such as PLTF-000412. It lives in the image, in the file name and in the load file.

Bates stamp

The visible mark burned into the page image. Older usage meant the physical hand stamp. A page can carry a Bates number in its file name without a visible stamp, which is how many native files are produced.

Bates range

First page to last page of a document or a production: PLTF-000100 to PLTF-000245. A range should always equal the page count it claims.

Control number

An internal ID a review platform assigns before production. It is not a Bates number, and it should never appear in a citation that leaves the building.

Endorsement

The term many agency specifications use for the stamp itself. The SEC standards say images "must be endorsed with sequential Bates numbers in the lower right corner of each image."

Slip sheet

A placeholder page that holds a Bates number for something not imaged: a withheld privileged page, a spreadsheet, an audio file, a DICOM study.

Who uses the number? Everyone downstream. The deposition taker says "I'm showing you PLTF-000412." The expert lists "materials reviewed: DEF-000001 to DEF-002318." The medical chronology cites the page for every entry. The judge reading a motion to compel reads "the gap between DEF-000430 and DEF-000442 is not explained." Each of those sentences fails if the numbers drift.

Medical records are crowded at the edges. Fax headers run across the top, flowsheet columns run to the right margin, and signature blocks sit at the bottom right, exactly where the stamp wants to go. Here is a produced page with every element that matters, pinned.

Placement conventions are strong but not universal. The SEC wants the number in the lower right corner of each image. Many ESI protocols say "lower right, not obscuring any content." If yours says nothing, use the lower right and write the choice down so the next production matches.

Our rule for medical records: before stamping a batch, scale every page to about 95% and add a white footer band. It costs nothing, and it ends the argument about whether a stamp covered the 21:52 on a signature line. A stamp over a time entry turns a routine page into an authenticity dispute.

There is no national standard for the format. There are strong conventions, and the federal agencies that receive the most productions have written theirs down. The DOJ Antitrust Division's specification is the clearest: numbers must be unique, sequential across the entire production, "of a consistent length, include leading zeros," and carry no more than 3 segments, for example a company identifier, a custodian segment and a page counter.

Ranges are written first page to last page: PLTF-000100 to PLTF-000245. Some teams shorten that to PLTF-000100-245. Pick 1 style and keep it in every letter, log and index. A range must equal its page count: 000100 to 000245 is 146 pages, and if the PDF has 145, something is missing or doubled.

Here is something that surprises new paralegals: the text of Federal Rule of Civil Procedure 34 never uses the word "Bates." No federal rule makes you stamp pages. Bates numbering is required by ESI protocols, stipulated orders, local rules, agency specifications and the plain need to cite a page. The rules below are what sits underneath it.

2 advisory committee sentences from the 2006 amendments to Rule 34 explain why a stamped TIFF alone is not always enough.

"If the responding party ordinarily maintains the information it is producing in a way that makes it searchable by electronic means, the information should not be produced in a form that removes or significantly degrades this feature."

That is why stamped images travel with extracted text and a load file. A flat, unsearchable PDF of an EHR export, stamped and nothing else, invites a motion. The same note adds that a "reasonably usable form" does not let a party convert ESI into a form "that makes it more difficult or burdensome for the requesting party to use the information efficiently."

The Sedona Conference says the same thing in its own words. Principle 12 of The Sedona Principles, Third Edition (October 2017): production "should be made in the form or forms in which it is ordinarily maintained or that is reasonably usable given the nature of the electronically stored information and the proportional needs of the case."

For medical records specifically, HIPAA adds a layer. Under 45 CFR 164.512(e), a covered entity can disclose records in response to a subpoena or discovery request when, among other routes, the party seeking them has sought a qualified protective order. The order usually dictates the legend tiers and who may see which ranges. State rules differ on privilege logs, confidentiality designations and whether a party must produce records it obtained by authorization, so check your jurisdiction.

Our view: put the Bates format, the legend wording, the load-file fields and a 502(d) order in the ESI protocol before anyone produces anything. Every one of them is cheap to agree on in month 1 and expensive to fight about in month 9.

Stamping is step 4 of 6. Most Bates disasters come from doing it at step 2, before the set is final. Here is the order we recommend, then the same order on a dated hypothetical.

1. Collect and freeze. Keep an untouched copy of every set exactly as it arrived. If alteration or authenticity comes up later, the unstamped original is the only clean witness. Record a hash (MD5 or SHA-1 are the fields the DOJ specification lists) for each file.
2. Organize. Sort by provider and date of service. Decide the production order now, because the Bates sequence will freeze it.
3. Cull duplicates carefully. Exact duplicates share a hash and can go, if the protocol allows it. Near-duplicates stay. 2 copies of a nursing note that differ by 1 late addendum are evidence, not clutter. See altered medical records.
4. Privilege and redaction review. Mark each withheld page for a slip sheet. Mark redactions and keep the unredacted original in the review set.
5. Stamp. Numbers and legends in 1 pass. 2 people stamping parts of 1 production with overlapping start numbers is the classic way to create duplicate numbers.
6. QC, then produce. Check first and last numbers, range length against page count, legends by range and a sample of pages for stamp overlap. Then send, log and keep an exact copy of what went out.

Pre-production QC checklist

• Untouched originals savedEvery set as received, hashed, before any stamp
• Prefix and width match the protocolSame separators and digit count as earlier productions
• Start number continues the sequenceLast produced number plus 1, checked against the log
• Range length equals page countLast minus first plus 1, for every document and the whole set
• No stamp on contentSample flowsheets, signature pages and fax-header pages
• Legends correct by rangeExact wording from the protective order
• Slip sheets for every withheld pageEach one on the privilege log by Bates number
• Natives named by Bates numberSlip sheet in the images, hash in the load file
• Families intactAttachments follow their parent, BEGATTACH and ENDATTACH filled
• Load file opens cleanlyHeader row, delimiters, 1 OPT line per image
• Redactions burned inText layer and extracted text redacted too, as well as the image
• Exact copy of the production keptWhat went out, not what you meant to send

0 of 12 checked

A modern production is 4 things on 1 volume: page images, extracted text, native files, and load files that tie them together. The load files are how a review platform knows that PLTF-000213 to PLTF-000217 is 1 document and that PLTF-000218 is the first page of the next one. If you receive productions and have never opened a DAT file in a text editor, do it once. It demystifies the whole thing.

The SEC Data Delivery Standards (revised August 2026) are a good public reference. They call for black and white images as "300 DPI Group IV single-page TIFF files," color as JPEG, a unique Bates-number file name for every image, and no more than 2,000 image files per folder. They also say PDFs are not an acceptable format for images in that structured production. Your ESI protocol may say otherwise; many state-court medical record productions go out as stamped searchable PDFs.

The OPT image cross-reference file

The OPT (or LOG) file has 1 line per page image and 7 comma-separated fields: ImageID, VolumeLabel, ImageFilePath, DocumentBreak, FolderBreak, BoxBreak, PageCount. A "Y" in the fourth field marks the first page of a new document. Here is what the start of the hypothetical orthopedic set looks like.

The DAT metadata file

The DAT file holds 1 row per document with a header row of field names. In the Concordance default format the SEC describes, the field separator is ASCII character 020 and every value is wrapped in the thorn character, ASCII 254. Date, time and time zone go in separate fields. The Bates fields carry different names in different specifications, so match the one in your protocol.

The DOJ Antitrust Division specification also asks for a sample production of 500 to 1,000 records before the real one, so load problems surface early. That is a good habit on any large production, including a hospital's. The FTC Bureau of Competition Production Guide lists BEGBATES and ENDBATES as the field names and 300 DPI Group IV TIFF for black and white images.

Some records do not survive being turned into pages. A billing spreadsheet loses its formulas. An audio file of a 911 call has no pages. A DICOM imaging study is hundreds of slices plus metadata. An EHR audit trail export is a table meant to be sorted and filtered, not read on paper (see the EHR audit trail guide). For these, the usual answer is a native file with 1 Bates number, a slip sheet in the image set and a hash in the load file.

The agencies differ in the details, which is a reminder to read your own protocol. The SEC wants Excel spreadsheets to have "a placeholder image named by the Bates number of the file." The DOJ specification says the Bates range for a spreadsheet "should be a single number" and that the placeholder must show at least the BEGDOC#, file name and file path. And the SEC says native productions made without load files, with prior approval, "should not be Bates numbered" at all.

A tip for exhibits from natives: cite the native's Bates number plus a locator, such as "PLTF-000473, series 4, image 112" for a DICOM study or "DEF-001920, row 3,418" for an audit log. Agree the locator style with the other side before depositions. If you want to know what AI can and cannot do with a native imaging study, the DICOM guide covers it.

Medical records break the tidy ESI model in 3 ways. They arrive from many custodians, each with its own page numbering, fax headers and certification. They arrive late and in pieces, so supplemental productions are the norm. And they are full of near-duplicates, the same note printed on different days, which can be the most important pages in the case.

That produces 2 practical rules. First, when you stamp incoming records for internal use, leave the producing party's numbers visible and track both. Your internal prefix should never replace theirs. Second, keep a Bates-range log per provider from day 1. It becomes the index for the cover letter, the materials-reviewed list for your expert and the check against every citation in the chronology. If you are still gathering the records, how to get medical records for a lawsuit covers requests, authorizations and subpoenas.

The ledger is boring, which is the point. If opposing counsel writes "your production is missing pages," you answer with arithmetic, not a search.

BATES-RANGE LOG
Case: [CAPTION]                    Producing party: [PARTY]
Prefix: [PREFIX]   Separator: [-]   Width: [6] digits   Legend: [WORDING]

Prod # | Date sent  | Provider / custodian | Dates of service      | Bates start   | Bates end     | Pages | Natives | Withheld (slip sheets) | Notes
-------|------------|----------------------|-----------------------|---------------|---------------|-------|---------|------------------------|------
[1]    | [MM/DD/YY] | [PROVIDER]           | [MM/DD/YY]-[MM/DD/YY] | [PREFIX-000001] | [PREFIX-000212] | [212] | [0]     | [none]                 | [Custodian cert received MM/DD]
[1]    | [MM/DD/YY] | [PROVIDER]           | [MM/DD/YY]-[MM/DD/YY] | [ ]           | [ ]           | [ ]   | [ ]     | [Bates nos. + reason]  | [Exact dupes removed: N, by MD5]
[2]    | [MM/DD/YY] | [PROVIDER, supplement] | [ ]                 | [continues]   | [ ]           | [ ]   | [ ]     | [ ]                    | [Why late; what it adds]

CHECKS FOR EACH ROW
[ ] End minus start plus 1 = pages + natives
[ ] Start = previous end + 1
[ ] Provider's own page marks and fax headers visible
[ ] Unstamped original saved at: [PATH]
[ ] Other side's Bates numbers (if any) recorded in Notes

[FIRM LETTERHEAD]
[DATE]

VIA [SECURE FILE TRANSFER / OVERNIGHT]

[OPPOSING COUNSEL NAME]
[FIRM]
[ADDRESS]

Re: [CAPTION], [CASE NO.]
    [PARTY]'s Production [NUMBER] in response to [REQUEST SET AND NUMBERS]

Counsel:

Enclosed is [PARTY]'s Production [NUMBER], Bates numbered [PREFIX-START] through [PREFIX-END] ([TOTAL] numbers). The production is designated [LEGEND] under the [Stipulated / Qualified] Protective Order entered [DATE].

Form: [single-page Group IV TIFF with document-level text / searchable PDF], with [DAT and OPT] load files. Native files are named by Bates number and represented by slip sheets in the image set.

Index:
  [PREFIX-000001] to [PREFIX-000212]  [PROVIDER]  records, [DATES OF SERVICE]
  [PREFIX-000213] to [PREFIX-000354]  [PROVIDER]  records, [DATES OF SERVICE]
  [PREFIX-000473]                     [PROVIDER]  [DICOM study / spreadsheet], native, MD5 [HASH]

Withheld: [N] pages withheld as [privileged / work product], represented by slip sheets at [BATES NUMBERS] and listed on the enclosed privilege log.

Duplicates: Exact duplicates, identified by [MD5] hash, were not produced. No near-duplicate or differing versions were removed.

This production [does / does not] complete [PARTY]'s response to Request Nos. [ ]. [PARTY] will supplement as additional records are received, continuing the Bates sequence.

[SIGNATURE BLOCK]
Enclosures: [MEDIA / LINK], privilege log, Bates-range log

[PREFIX-000401]

[DOCUMENT WITHHELD]
Reason: [Attorney-client privilege / Work product / Produced in native format]
Privilege log entry: [NUMBER]

-- or, for a native file --

[PREFIX-000473]
PRODUCED IN NATIVE FORMAT
File name: [PREFIX-000473].[EXT]
Original file name: [NAME]
Custodian: [PROVIDER]
MD5: [HASH]

[CONFIDENTIALITY LEGEND]

A deposition transcript that reads "this page here" is hard to check on appeal. One that reads "PLTF-000412" can be checked by anyone with the production. Cite the page, every time, in every document that leaves the office.

When the same page appears in 2 productions

In medical cases it always does. The plaintiff produces the orthopedic chart obtained by authorization; the defense produces the same chart obtained by subpoena, with different numbers and sometimes different contents. Pick 1 set as canonical for citations, usually the most complete one, and keep a table that maps the other set's numbers to it. That table is built by a person, page by page, and it is where the interesting findings live.

The step-by-step method for this comparison, including how to handle addenda and differing versions, is in reconciling a defense record production. Treat a partial match as a question, not a conclusion. An addendum can be a proper late entry. It can also be something else, and that is for the reviewer and the experts to work out.

Ranked by how much damage each does once a production is out the door, based on what the rules above make hard to undo.

AI ediscovery is older than most people think. Technology assisted review (TAR), also called predictive coding, trains a classifier on attorney decisions and ranks documents by likely relevance. In February 2012, Magistrate Judge Andrew Peck approved its use in Da Silva Moore v. Publicis Groupe, 287 F.R.D. 182 (S.D.N.Y. 2012).

"This judicial opinion now recognizes that computer-assisted review is an acceptable way to search for relevant ESI in appropriate cases."

3 years later, in Rio Tinto v. Vale, 306 F.R.D. 125 (S.D.N.Y. 2015), the same judge wrote that it is now black letter law that courts will permit a producing party that wants to use TAR for document review to do so. Sedona Principle 11 points the same way: a responding party may meet its obligations "by using technology and processes, such as sampling, searching, or the use of selection criteria."

TAR sorts documents. It does not number pages, and it does not care much whether a page is stamped. The newer wave of legal AI tools does care. Generative AI built on a large language model (LLM) now drafts chronologies, deposition summaries and record summaries from the produced pages. An AI medical chronology is only usable in litigation if every line carries a page-level citation a person can click and check. And that citation is only stable if the page numbers are.

Where AI helps and where it fails on Bates-stamped records

The last failure is the one that ends careers. In Mata v. Avianca, 678 F. Supp. 3d 443 (S.D.N.Y. 2023), lawyers filed a brief citing court decisions that did not exist, generated by ChatGPT, and were sanctioned $5,000 under Rule 11. A fabricated case citation and a fabricated Bates citation are the same failure: a reference nobody checked. The fix is also the same. Check every cited Bates number against the range log before anything is filed or sent. A citation to PLTF-000999 in a production that ends at PLTF-000487 should never survive QC.

Why AI makes stamp-before-review a hard rule

If the model reads the set before it is stamped, it cites PDF page numbers. Then someone dedupes, re-sorts and stamps, and every citation drifts. 500 chronology entries now point at the wrong page, and nobody notices until a witness is handed the wrong exhibit. Our rule: freeze the set, stamp it, then run the AI. When a supplement arrives, stamp it in sequence and add it, rather than re-running the whole set in a new order.

Checklist for legal AI tools that touch a production

• Signs a business associate agreementHIPAA compliant AI starts with a BAA before any PHI moves
• SOC 2 report availableAsk for it, read the exceptions
• No training on your dataIn the contract, not the FAQ
• Citation on every linePage-level, clickable, to the source page
• Keeps Bates numbers stableThrough dedupe, sort and export
• Flags low-confidence OCR pagesFaxes and handwriting get a human look
• Keeps the stamp out of the textBates and fax headers not read as clinical content
• Human-in-the-loop by designA person reviews, edits and signs the output
• Keeps an audit trail of AI useWho ran what, on which pages, when
• Exports with citations intactWord, PDF and hyperlinked exports keep the page links

0 of 10 checked

[__]. Bates Numbering.
(a) Each page of each document produced in image form shall bear a unique Bates number, burned into the image in the lower right corner without obscuring any content, in the format [PREFIX]-[NNNNNN] (fixed width, leading zeros, no spaces).
(b) Each party shall use a unique prefix. Numbering shall be sequential within and across productions and shall not restart. Once produced, a Bates number shall not be reassigned.
(c) Documents withheld in full on a claim of privilege shall be represented by a slip sheet bearing the Bates number and the words "Withheld: [Privilege]," and shall be listed on the privilege log by Bates number.
(d) Files produced in native form shall be named by their Bates number, represented by a slip sheet in the image set, and accompanied in the load file by their [MD5/SHA-1] hash.
(e) Load files shall include at least: BEGBATES, ENDBATES, BEGATTACH, ENDATTACH, CUSTODIAN, NUMPAGES, [HASH], TEXTPATH and NATIVELINK. Image cross-reference files shall be in [OPT] format.
(f) Confidentiality designations under the Protective Order shall be stamped on each page [lower left / bottom center].
(g) Any chronology, summary or index a party serves or files that cites produced documents, whether prepared by a person or with the help of software, shall cite by Bates number.
(h) The parties request entry of an order under Federal Rule of Evidence 502(d).

More on accuracy thresholds and what courts expect is in is AI accurate enough for court, and on the privacy side, HIPAA compliant AI medical record review. The short version for AI medical record review: the model drafts, a person checks every cite against the page, and the signature on the work product belongs to the person.

Everything above stands without this chapter. This part is ours, and you should read it as a vendor describing its own product.

Medrecords AI is medical record review software. Bates numbering and stamping applies sequential numbers and confidentiality legends across the uploaded file in 1 pass. You set the prefix, start number and padding. The numbers stay stable through deduplication, re-sorting and export, so a page keeps its number for the life of the file.

The chronology cites the source page of the uploaded file on every line, and each cite links to that page. When the file is stamped, the citations speak Bates. OCR routes each page to the right engine for typed text, handwriting or tables, and flags low-confidence pages for a person to read. Hyperlinked exports keep the links to the source pages.

What it does not do

Pricing, if it helps the decision: Self-Service bills 10 cents a deduplicated page, down to 5 cents at volume, duplicates free; Enterprise On-Prem is an annual license. Details are on the pricing page, and the security page covers the controls.

Why is it called Bates numbering?

It is named after Edwin G. Bates, whose hand numbering machine stamped a number on a page and advanced to the next number by itself. The Henry Ford museum dates his patented improvement to 1891. Law offices used the stamp on paper productions for decades, and the name stayed when software took over.

What is the difference between a Bates stamp and a Bates number?

The Bates number is the identifier, such as PLTF-000412. The Bates stamp is the visible mark of that number on the page image. A native file can carry a Bates number in its file name and slip sheet without any stamp on the file itself.

Where does the Bates number go on a page?

Usually the lower right margin, clear of content. The SEC Data Delivery Standards require the lower right corner. The confidentiality legend goes in another corner or at the bottom center. On medical records, check that the stamp does not cover a signature, time, dose or flowsheet value.

Should Bates numbers restart for each production?

No. A later production from the same party continues the sequence, so every page in the case has 1 unique number. Restarting creates 2 pages with the same number and breaks earlier citations. The DOJ Antitrust Division specification also says the digit count should not change in later productions.

Do the Federal Rules require Bates numbering?

Not by name. Rule 34 governs the form of production and Rule 26 governs privilege claims and the discovery plan, but neither text says "Bates." The requirement usually comes from the ESI protocol, a stipulated order, local rules or agency specifications. Check your court's rules and any standing order.

How are privileged pages handled in a Bates range?

With a slip sheet: a placeholder page carrying the Bates number and a short reason, such as "Withheld: attorney-client privilege." The withheld item then appears on the privilege log by that number, which is how FRCP 26(b)(5)(A) claims are usually made. State rules on privilege logs vary.

Do native files get Bates numbers?

Usually 1 number for the whole file, used as the file name and printed on a slip sheet in the image set, with the file hash in the load file. Some specifications differ: the SEC says natives produced without load files, with prior approval, should not be Bates numbered. Follow your protocol.

Can AI apply Bates numbers to medical records?

Software has applied Bates numbers for decades, and many AI record review tools include stamping. The questions to ask are whether numbers stay stable through deduplication and re-sorting, whether the stamp stays out of the extracted text, and whether every AI-drafted line cites a Bates page you can open.

Can ChatGPT summarize a Bates-stamped production for a lawsuit?

A general chatbot can summarize text, but consumer versions usually do not sign a business associate agreement, do not cite pages reliably and may invent citations. Mata v. Avianca shows what happens when unverified AI citations get filed. Use a tool that signs a BAA and cites every line, and check each cite.

Is it HIPAA compliant to upload Bates-stamped medical records to AI?

It can be, if the vendor handles PHI as a business associate under a signed BAA, and your use fits the protective order in the case. Many qualified protective orders limit use of PHI to the litigation, so read the order before you upload anything to any vendor.

Does AI technology assisted review replace Bates numbering?

No. TAR decides which documents are likely relevant; Bates numbers identify the pages that get produced. Courts have accepted TAR since Da Silva Moore in 2012, but the produced set still needs a stable number on every page for citations, exhibits and privilege logs.

This guide was written by Medrecords AI from primary sources, read in September 2026. Rule text is quoted from the versions published by the Legal Information Institute. Agency specifications are quoted from the documents the agencies publish. Hypotheticals, including every Bates number, date and page count in them, are invented and labeled. Nothing here is legal advice, and local rules, standing orders and ESI protocols control.

Product facts in chapter 12 are what Medrecords AI publishes on its own product pages. Related guides: medical record organization, deduplicating medical records and a verified chronology from scanned records.