Objective Document Coding Extractor
When you need consistent bibliographic fields for a set of documents, for example to build an exhibit index, feed a chronology or prepare a production, without mixing in subjective relevance or privilege calls.
Objective coding is the unglamorous foundation of document-heavy litigation: date, author, recipients, document type, Bates range and the people and entities named. It is mechanical, which makes it tempting to rush, but inconsistent coding breaks everything built on it, from chronologies and exhibit lists to privilege logs and witness kits. The value is consistency and accuracy, not judgment.
This prompt keeps Claude strictly on the objective side. You paste each document with its id and Bates range, set the conventions for dates, names and document types, and Claude fills a single table with the same columns for every document. Anything not on the face of the document is marked BLANK rather than inferred, ambiguous values are flagged with the text relied on, and email chains are handled consistently. Claude finishes with a list of questions for the review lead.
Claude is explicitly told not to make relevance, privilege or responsiveness calls; those subjective decisions belong to attorneys, and the document-review-classifier prompt handles a first pass at them. Even objective fields can be misread, especially on scanned or poorly formatted text, so build in sampling and quality control before the table is relied on for anything that leaves the team.
The Prompt
I am coding documents for [MATTER NAME]. I need objective fields only, entered consistently, for each document below. Do not make relevance, privilege or responsiveness calls; those are handled separately. <document id="1" bates="[BATES RANGE OR N/A]"> [PASTE DOCUMENT TEXT, INCLUDING HEADERS AND SIGNATURE BLOCKS] </document> <document id="2" bates="[BATES RANGE OR N/A]"> [PASTE DOCUMENT TEXT] </document> [REPEAT AS NEEDED] Coding conventions: - Date format: [YYYY-MM-DD]. Use the document's own date (sent date for email, signature or "dated as of" date for agreements). If there are several dates, use [RULE, e.g., the earliest] and list the others in Notes. - Names: [Last, First]. Record names as they appear; do not resolve nicknames or initials unless the document does. - Document types: choose from [LIST, e.g., Email, Letter, Memo, Agreement, Invoice, Spreadsheet, Presentation, Text Message, Other]. Output one table with columns: Doc ID | Bates Begin | Bates End | Date | Doc Type | Author/From | Recipients/To | CC | BCC | Title or Subject | People Mentioned | Entities Mentioned | Notes. Rules: - If a field is not on the face of the document, write BLANK. Never guess or infer a value. - If a value is partly legible or ambiguous, enter what appears and write "uncertain" in Notes, quoting the text you relied on. - For email chains, code the top message and list earlier messages in Notes with their dates and senders. After the table, list coding questions for the review lead, such as documents that fit no listed type or names that may refer to the same person. This is a first pass for reviewer and attorney quality control.
Example Output
A single coding table with date, document type, author, recipients, CC, BCC, subject, people and entities for each document, with BLANK for missing fields and a list of questions for the review lead.
Illustrative example — names, figures, and facts are fictional.
OBJECTIVE CODING TABLE: Alderwood Capital v. Penrose Data Systems (fictional)
Doc 1 | PEN0001022 | PEN0001023 | 2024-04-17 | Email | Rivas, Tomas | Holt, Andrea | Ng, Calvin | BLANK | RE: Q2 migration timeline | Rivas, Tomas; Holt, Andrea; Ng, Calvin; "Priya" | Penrose Data Systems; Alderwood Capital | Earlier message in chain: 2024-04-16 from Holt, Andrea. "Priya" appears without a surname.
Doc 2 | PEN0001024 | PEN0001031 | 2024-03-01 | Agreement | BLANK | BLANK | BLANK | BLANK | Statement of Work No. 3 | Holt, Andrea (signatory); Ostrowski, Lee (signatory) | Penrose Data Systems; Alderwood Capital | "Dated as of March 1, 2024." Signature dates 2024-03-04 and 2024-03-06 listed per convention.
Doc 3 | PEN0001032 | PEN0001032 | BLANK | Memo | BLANK | BLANK | BLANK | BLANK | Migration risks (draft) | Ng, Calvin | Penrose Data Systems | No date or author on the face of the document. Header reads "DRAFT: internal."
Doc 4 | PEN0001033 | PEN0001035 | 2024-05-2? | Invoice | BLANK | BLANK | BLANK | BLANK | Invoice 7781 | BLANK | Penrose Data Systems; Alderwood Capital | Date uncertain: the last digit is illegible in the text ("05/2_/2024").
QUESTIONS FOR REVIEW LEAD
1. Doc 3 fits "Memo," but it may be notes. Should it be "Other"?
2. "Priya" in Doc 1 may be Priya Sandhu, who appears in other batches. Confirm before linking names.
3. Doc 4: should a partly legible date be coded with the known parts or as BLANK?
All names and facts are fictional and illustrative.Tips
- •Set the conventions once (date format, name format, document type list) and reuse the same prompt for every batch so the tables merge cleanly.
- •BLANK is a feature. A guessed author or date is worse than an empty field because it looks reliable.
- •Spot-check a sample of rows against the originals each batch; objective coding errors compound when the table feeds a chronology or exhibit list.
- •For subjective calls like relevance, responsiveness or privilege, use document-review-classifier with attorney oversight.
- •This is a first pass for quality control and attorney review, not a final coding record.
Frequently Asked Questions
What is the difference between objective and subjective coding?
Objective coding records facts on the face of a document: date, author, recipients, type, Bates numbers and names. Subjective coding records judgments, such as relevance, responsiveness, issue tags, privilege and hot-document flags. This prompt does only the first. Keeping the two separate makes quality control easier and keeps attorney judgment where it belongs.
Why mark fields BLANK instead of letting Claude infer them?
Inferred values look as authoritative as real ones once they are in a table. If an undated memo is coded with a guessed date, it can end up in a chronology or exhibit list as fact. BLANK tells everyone downstream that the field needs another source, such as metadata, a cover email or a witness.
Can I use this alongside metadata from the production?
Yes. If you have load-file metadata, paste the relevant fields into each document tag and tell Claude whether metadata or the face of the document controls when they differ. Ask it to note any conflicts in the Notes column, since a mismatch between metadata and the document itself can be significant.
How do I keep coding consistent across batches?
Fix the conventions in the prompt and do not change them mid-review: date format, name format, the document type list and the rule for multiple dates. Save the prompt as a template or in a Claude Project. Periodically paste a sample of earlier rows and ask Claude to flag inconsistencies across batches.
Get New Prompts Like This Every Week
Join the free Claude for Lawyers newsletter — weekly prompts, tutorials, and practice-specific guides.