Civil War Pension Affidavit Analysis Prompt Pack

This prompt pack is designed for Claude, ChatGPT, Gemini, or Perplexity. The prompts work best when you provide clean input, verify output carefully, and treat the AI as a research assistant rather than an autonomous decision-maker. 

How to Use This Prompt Pack

 

 

Step 1: Start with Prompt 1 if your source is raw OCR. Otherwise skip to Prompt 2.

Step 2: Use Prompt 2 on every affidavit in the pension file to build a master extraction table.

Step 3: Combine all extracted tables from multiple affidavits into one spreadsheet, then run Prompt 3 for cluster analysis.

Step 4 (optional): Use Prompt 4 when you need a neighbor-focused research list for census, land, and probate follow-up. 

Prompt 1: Clean Transcription from OCR

When to use: You have OCR output or a rough transcript that needs cleanup before analysis.

Copy and paste this prompt with your OCR text:

text
I have OCR output from a Civil War pension affidavit. Please clean and format this transcript for genealogical research:

1. Correct obvious OCR errors (common letter substitutions like "rn" for "m", "cl" for "d")
2. Preserve all original spelling of names and places exactly as written
3. Keep all dates, numbers, and legal phrases intact
4. Format as readable paragraphs with line breaks between sections
5. Flag any unclear words with [unclear] rather than guessing
6. Do not modernize language or correct period grammar
7. Preserve all signatures and notary statements

Here is the OCR text:

[PASTE YOUR OCR TEXT HERE]

Output the cleaned transcript only, with [unclear] markers where needed.

---

Prompt 2: Extract All Named People and Relationships

When to use: After you have a clean transcript and want structured extraction of every person mentioned.

Copy and paste this prompt with your clean transcript:

text
Extract every person named in this Civil War pension affidavit into a structured table. For each person, provide:

- Full name as written (preserve exact spelling)
- Residence stated in affidavit
- Relationship term used (neighbor, comrade, physician, minister, relative, official, witness, or unknown)
- Years of acquaintance if stated (e.g., "known 15 years")
- Event witnessed or testified about (marriage, death, disability, residence, reputation)
- Exact phrase from affidavit supporting the relationship (quote directly)
- Label: Choose ONE only from this list:
  * NEIGHBOR (explicit: uses words "neighbor," "lived near," "resided next to")
  * POSSIBLE NEIGHBOR (indirect: same town/county, local acquaintance, community reputation testimony)
  * RELATIVE (explicit or very likely family)
  * COMRADE (military service connection)
  * MEDICAL (physician, surgeon, nurse)
  * RELIGIOUS (minister, priest, church official)
  * LEGAL (notary, clerk, official)
  * UNKNOWN (insufficient information)

Rules:
- Do not infer relationships beyond what the text states
- Quote the exact phrase that justifies each label
- Mark uncertain cases as POSSIBLE NEIGHBOR, not NEIGHBOR
- Include the affiant themselves as the first row
- If multiple people appear in one sentence, create separate rows
- Do not hallucinate names or details

Output as a markdown table with these columns:
Name | Residence | Relationship Term | Years Known | Event Witnessed | Supporting Quote | Label

Here is the affidavit transcript:

[PASTE YOUR CLEAN TRANSCRIPT HERE]

---

Prompt 3: FAN-Club Cluster Analysis

When to use: After extraction, to identify research-worthy clusters and connections across multiple affidavits.

Copy and paste this prompt with your extracted name table from Prompt 2:

text
Analyze this list of people extracted from Civil War pension affidavits and identify research-worthy clusters and patterns.

For the analysis, provide:

1. **Network Clustering:** Group people into these networks:
   - Civilian/Local Network (neighbors, local residents, non-military)
   - Military Network (comrades, regiment connections)
   - Family Network (relatives, likely in-laws based on surnames)
   - Professional Network (doctors, ministers, lawyers, officials)
   - Unknown/Unclassified

2. **Pattern Detection:** Flag these research clues:
   - Repeated surnames (possible family connections)
   - Shared residence locations (tight neighborhood clusters)
   - People who appear multiple times across different affidavits
   - Long acquaintance stated before migration (e.g., "knew him 20 years" when affidavit is 10 years after arrival)
   - Witnesses present at key life events (marriage, birth, death)
   - Surnames matching children's middle names in the pension file

3. **Priority Research List:** Rank the top 5-10 people most worth researching further and explain why for each:
   - What records to check (census, land, probate, marriage, their own pension file)
   - What questions they might answer
   - What connections they might reveal

4. **Geographic Migration Clues:** If any affidavits mention prior residence or acquaintance in another location, list those with estimated timeline.

Here is the extracted data:

[PASTE YOUR TABLE FROM PROMPT 2 HERE]

Additional context (optional):
- Veteran's name:
- Widow's name:
- Known residence locations:
- Known military unit:

---

Prompt 4: Neighbor-Specific Deep Dive (Optional Advanced)

When to use: When you want to focus specifically on neighbor relationships and prepare for cluster research.

text
From this list of people extracted from a pension affidavit, identify everyone who is either explicitly a neighbor or possibly a neighbor based on the evidence.

For each neighbor or possible neighbor, provide:

1. Name and residence
2. Evidence category:
   - EXPLICIT: affidavit uses "neighbor" or "lived near"
   - PROXIMITY: same town/county, local witness
   - REPUTATION: testified about household reputation or character
   - LONG ACQUAINTANCE: knew family 10+ years in same locality
   - LIFE EVENT: present at marriage, birth, or death
   - UNCERTAIN: some neighbor signals but weak evidence

3. Estimated start of association (calculate from "known X years" if present)
4. Possible kinship clues (shared surname, present at family event, migrated together)
5. Next research actions (specific records and why)

Rules:
- Only include people with at least some neighbor evidence
- Exclude pure military comrades unless they're also post-war neighbors
- Exclude officials acting only in official capacity
- Flag anyone who might be family disguised as neighbor

Format as a research-ready table:
Name | Residence | Evidence Category | Assoc. Start Est. | Kinship Clues | Next Records

Here is the extracted data:

[PASTE YOUR TABLE FROM PROMPT 2 HERE]

---

Best Practices

• Always verify AI output against the original affidavit image before citing.

• Copy AI output into a spreadsheet immediately for sorting and filtering.

• Add a citation column with affidavit page/image number for each extracted person.

• Flag any hallucinated names during verification and remove them.

• Save your prompts and outputs in your research log for reproducibility.

• Treat AI extraction as a first pass, not finished genealogy.

---