Skip to content

AI Prompt Library

This page is the working library for reusable AI prompts in Archives workflows. Use it together with AI-Assisted Archives Workflows, which defines the review and policy expectations for AI-assisted work.

Purpose

Use this page to collect prompts for:

  • subject heading suggestions
  • scope and content note drafting
  • biographical or historical note drafting
  • transcript cleanup and speaker review
  • metadata extraction from OCR or transcripts
  • formatting and restructuring descriptive drafts

How to Use These Prompts

  1. Treat every prompt result as a draft.
  2. Compare the output to the source material.
  3. Correct names, dates, places, and context before reuse.
  4. Remove invented or uncertain details.
  5. Apply local descriptive standards before saving or publishing.

Human review is required

Prompts on this page are staff aids, not authoritative description. Final subject analysis, scope and content notes, and public description still require human review and approval.

Prompt Template

Use this structure when adding new prompts to the library.

Field What to include
Task The archival task the prompt supports.
Prompt The actual reusable prompt text.
Use With The source material to give the AI, such as OCR text, transcript text, or draft description.
Review Required The facts, standards, and risks staff must check before reusing the output.

Subject Headings

Research Subject Headings and Keywords from an Uploaded File

Field What to include
Task Analyze an uploaded transcript, oral history, finding aid, or similar file and suggest a small set of focused, evidence-based subject headings and keywords.
Use With Oral history transcripts, interview transcripts, finding aids, OCR text, digitized documents, or other descriptive source files.
Review Required Verify every proposed controlled heading against FAST or LCSH, remove unsupported or overly broad terms, and confirm the final list reflects the file rather than assumed historical context.
You are assisting with archival metadata creation for an uploaded file or a provided link, such as an oral history transcript, finding aid, interview transcript, or digitized document.

Your task is to analyze the uploaded file and suggest a small set of focused, valid archival subject headings and keywords.

Rules:

1. Use the document itself as the evidence.

   * Do not infer topics that are not clearly represented.
   * Do not add broad historical, social, political, or geographic themes unless the file substantially discusses them.
   * Prefer the most specific accurate term over a broad umbrella term.

2. Return only 5-10 total recommendations.

   * Fewer is acceptable if the document is narrow.
   * Do not pad the list.
   * Prioritize names, places, organizations, events, and the most important topical subjects.

3. Prefer controlled subject headings.

   * Prefer FAST headings when available.
   * LCSH headings are acceptable when they are clearly valid and more appropriate.
   * Do not invent controlled headings.
   * Do not "clean up" or modernize a controlled heading unless the authority record uses that form.

4. Verify each controlled heading.

   * Check the heading against an authoritative source such as OCLC FAST/searchFAST or the Library of Congress Linked Data Service.
   * If you cannot verify the exact heading, do not present it as FAST or LCSH.
   * If a useful concept cannot be verified as a controlled heading, list it separately as a local keyword.

5. Separate subject headings from keywords.

   * "Subject headings" must be valid controlled vocabulary terms.
   * "Keywords" may be local/free-text terms useful for discovery but should still be conservative and evidence-based.

6. Formatting requirements:

   * For FAST headings, give the heading and label it "FAST."
   * For LCSH headings, give the heading and label it "LCSH."
   * Format LCSH subdivisions with double dashes and no spaces, for example: Agriculture--Georgia.
   * Include a short justification based on the file.
   * Include a confidence rating: High, Medium, or Low.
   * Do not include unverified headings in the controlled-heading section.

Output format:

## Recommended Controlled Subject Headings

| Heading | Vocabulary | Why it fits | Confidence |
| ------- | ---------- | ----------- | ---------- |

## Recommended Local Keywords

| Keyword | Why it fits | Confidence |
| ------- | ----------- | ---------- |

## Optional Genre/Form Terms

| Term | Vocabulary | Why it fits | Confidence |
| ---- | ---------- | ----------- | ---------- |

## Excluded / Not Recommended

Briefly list any tempting but rejected headings and explain why they are too broad, unsupported, duplicative, or unverified.

Before finalizing, double-check that the total number of recommended controlled headings and local keywords is no more than 10.

Scope and Content Notes

Draft a DACS-Aligned Scope and Contents Note from an Uploaded File

Field What to include
Task Analyze an uploaded transcript, interview, digitized document, or similar archival file and draft a concise Scope and Contents note aligned with DACS principles.
Use With Oral history transcripts, interview transcripts, OCR text, digitized documents, reports, correspondence, and similar descriptive source files.
Review Required Confirm that the note reflects only what is explicitly present in the file, remove interpretation or unsupported context, and check that names, dates, places, formats, and topics are accurate and complete.
You are assisting with archival metadata creation for an uploaded file, such as an oral history transcript, interview transcript, digitized document, or other archival material.

Your task is to analyze the uploaded file and write a Scope and Contents note that follows DACS (Describing Archives: A Content Standard) principles.

Rules:

1. Base the note strictly on the document itself.

   * Do not infer, speculate, or add contextual information that is not clearly supported by the file.
   * Do not introduce historical background, interpretation, or analysis unless it is explicitly present in the document.
   * Do not guess names, dates, locations, or topics.

2. Follow DACS guidance for Scope and Contents notes.

   * Clearly describe the nature, content, and subject matter of the material.
   * Identify the type of material (e.g., oral history transcript, interview, correspondence, report).
   * Summarize the main topics, themes, people, organizations, places, and events discussed.
   * Note the structure or format if relevant (e.g., question-and-answer format, chronological narrative).
   * Include dates or time periods only if explicitly stated in the document.
   * Keep the tone neutral, factual, and concise.

3. Do not include:

   * Interpretation, evaluation, or significance statements.
   * Biographical or historical notes unless they are directly part of the document content.
   * Information not present in the file.

4. Length and clarity:

   * Write 1-2 concise paragraphs.
   * Use complete sentences.
   * Avoid redundancy and overly general language.

5. Insufficient information:

   * If the uploaded file does not provide enough information to produce a meaningful Scope and Contents note, do not fabricate content.
   * Instead, clearly state that more information is needed and specify what is missing (e.g., unclear subject matter, missing names, lack of descriptive content).

Output format:

## Scope and Contents

[Write the Scope and Contents note here, or explain why there is insufficient information to create one.]

Biographical or Historical Notes

Draft a DACS-Aligned Biographical or Historical Note

Field What to include
Task Research a person, family, organization, place, event, or other topic and draft a concise DACS-aligned biographical, historical, or administrative note for archival description.
Use With A creator name, family name, organization, place, event, topic, and any supporting uploaded archival file or source notes.
Review Required Verify identities, dates, variant names, relationships, and relevance to the archival material; remove unsupported claims; and confirm that every factual statement is backed by an actual source.
You are assisting with archival description.

Your task is to write a concise Biographical Note or Historical Note following DACS principles for the topic provided by the user.

The user will provide one of the following:

* A person's name
* A family name
* An organization or corporate body
* A place
* An event
* Another topic that may require a biographical or historical note

You may also use an uploaded archival file for context, but do not rely only on the file unless it contains enough verified information.

## Core Rules

1. Do not guess or fabricate.

   * Do not invent birth dates, death dates, occupations, relationships, offices held, locations, organizational histories, dates of operation, event dates, or significance.
   * Do not assume that two people, places, organizations, or events with similar names are the same.
   * If identity is ambiguous, say so clearly.
   * If no reliable information can be found, write: No reliable information found.

2. Research before writing.

   Use:

   * The uploaded file, if provided
   * Wikipedia, if available
   * General web searches
   * Reliable institutional, government, educational, archival, newspaper, encyclopedia, or organizational sources when available

   Wikipedia may be used as a starting point, but important facts should be checked against another reliable source when possible.

3. Follow DACS-style content expectations.

   For a person or family, include only verified information relevant to archival context, such as:

   * Full name and variant names, if verified
   * Birth and death dates, if verified
   * Places of residence or activity, if relevant
   * Occupation, public role, or major activities
   * Family or organizational relationships, if relevant
   * Connection to the archival material, if known

   For an organization or corporate body, include only verified information such as:

   * Official name and variant names
   * Dates of existence or operation, if verified
   * Location
   * Function, purpose, mandate, or activities
   * Predecessor or successor bodies, if relevant and verified
   * Connection to the archival material, if known

   For a place, event, or historical topic, include only verified information such as:

   * What it is or was
   * Where it was located or occurred
   * Relevant dates or time period
   * Key participants, communities, organizations, or historical context
   * Why it is relevant to the archival material, if known

4. Keep the note focused.

   * Do not write a general encyclopedia article.
   * Include only information that helps users understand the archival material, creator, subject, or context.
   * Avoid broad historical background unless directly relevant.
   * Keep the note neutral, factual, and concise.

5. Use cautious language when needed.

   * If a fact appears in only one source, say "According to [source]..."
   * If sources conflict, identify the conflict instead of choosing one without explanation.
   * If the connection to the archival material is uncertain, say "The relationship to the material is not clear from the available sources."

6. Cite sources.

   * Provide citations or source links for all factual claims.
   * Do not cite sources that do not actually support the statement.
   * Do not cite Wikipedia alone for major facts if better sources are available.

## Output Format

## Note Type

State one:

* Biographical Note
* Historical Note
* Administrative History Note
* Insufficient Information

## Draft Note

Write 1-2 concise paragraphs.

If no reliable information is found, write only:

No reliable information found.

## Sources Consulted

List the sources used. Include source title/name and link if available.

## Unverified or Ambiguous Information

Briefly list any facts, identities, dates, or relationships that could not be verified.

## Reason for Exclusion

If you rejected tempting information because it was unsupported, ambiguous, overbroad, or unrelated to the archival material, briefly explain why.

Metadata Extraction

Extract Verified Dublin Core Metadata for Vtext / DSpace Entry

Field What to include
Task Extract verified Dublin Core metadata from an uploaded file, supplied web link, or both for entry into Vtext / DSpace.
Use With Uploaded files, repository records, webpages, OCR text, transcripts, embedded PDF metadata, cover pages, title pages, and related reliable source links.
Review Required Confirm that every field is supported by the file or linked source, remove guessed values, verify subject terms and rights statements, and check that the description and file details match the actual item.
You are assisting with archival metadata creation for entry into Vtext / DSpace.

Your task is to extract verified Dublin Core metadata from the uploaded file, supplied web link, or both.

Use only information that is clearly supported by the file, webpage, embedded metadata, visible text, file properties, or reliable linked source information. Do not guess, infer, or fabricate missing metadata.

## Core Rules

1. Use only verified information.

   * Do not invent titles, creators, dates, publishers, subjects, descriptions, rights statements, locations, or identifiers.
   * If a field is unclear, leave it blank.
   * If multiple possible values exist and the correct one is uncertain, list the ambiguity in the "Unclear / Needs Review" section.
   * Do not assume that a filename is the formal title unless no better title is available. If using the filename as a provisional title, clearly mark it as provisional.

2. Extract as much as possible.

   Check for metadata in:

   * The uploaded file content
   * File name
   * File size
   * File format
   * PDF title/author metadata, if available
   * OCR text or transcript text
   * Cover page or title page
   * Header/footer information
   * Webpage title and visible page metadata
   * Linked repository, institutional, or creator information, if available

3. Follow Dublin Core-style field usage.

   Use these fields when supported:

   * dc.title
   * dc.creator
   * dc.contributor
   * dc.subject
   * dc.description
   * dc.publisher
   * dc.date
   * dc.type
   * dc.format
   * dc.identifier
   * dc.source
   * dc.language
   * dc.relation
   * dc.coverage
   * dc.rights

4. Be conservative with subject headings.

   * Use only focused, document-supported subjects.
   * Prefer verified controlled vocabulary terms when possible.
   * Do not add broad or generic subjects unless they are central to the item.
   * Use no more than 5-10 subject terms total.
   * If a subject is useful but not verified as a controlled heading, label it as a local keyword.

5. Write the description in archival style.

   The description should begin with a physical/digital extent statement in this format:

   [number] electronic record(s) ([format or formats]), [file size in bytes] bytes. [filename].

   If multiple file formats are present, list the formats inside one parenthetical, separated by commas.

   Examples:

   * 1 electronic record (PDF), 77202 bytes. oral-history-transcript.pdf.
   * 3 electronic records (PDF, TIF, JPG), 245901 bytes total. interview-transcript.pdf; portrait.tif; access-copy.jpg.

   After that opening sentence, add a concise description of the item's content, based only on verified information.

6. Dates

   * Use ISO format when possible: YYYY-MM-DD.
   * If only a year is known, use YYYY.
   * If a date is uncertain, leave dc.date blank and explain the issue under "Unclear / Needs Review."
   * Do not derive a date from upload date, file modified date, or web crawl date unless the user specifically says to use that as the item date.

7. Rights

   * Only include a rights statement if it is explicitly provided or supplied by the user.
   * Do not assume public domain, copyright status, Creative Commons status, or institutional ownership.
   * If no rights information is present, leave dc.rights blank.

8. Output clean metadata.

   * Leave unknown fields blank.
   * Do not fill fields with "unknown," "N/A," or guesses.
   * Preserve original capitalization of proper names and titles unless there is an obvious OCR error.
   * Include a brief evidence note for important extracted values.

## Output Format

## Dublin Core Metadata Draft

| Field          | Value | Evidence / Notes |
| -------------- | ----- | ---------------- |
| dc.title       |       |                  |
| dc.creator     |       |                  |
| dc.contributor |       |                  |
| dc.subject     |       |                  |
| dc.description |       |                  |
| dc.publisher   |       |                  |
| dc.date        |       |                  |
| dc.type        |       |                  |
| dc.format      |       |                  |
| dc.identifier  |       |                  |
| dc.source      |       |                  |
| dc.language    |       |                  |
| dc.relation    |       |                  |
| dc.coverage    |       |                  |
| dc.rights      |       |                  |

## Suggested Subject Terms

| Term | Type | Vocabulary | Evidence |
| ---- | ---- | ---------- | -------- |

Type should be one of:

* Controlled subject heading
* Local keyword
* Name
* Place
* Organization
* Genre/Form

Vocabulary should be one of:

* FAST
* LCSH
* Local
* Other verified vocabulary

## File / Source Details

| Detail                     | Value |
| -------------------------- | ----- |
| File name                  |       |
| File format                |       |
| File size in bytes         |       |
| Number of files            |       |
| Source URL, if applicable  |       |
| Access date, if applicable |       |

## Unclear / Needs Review

List any fields that could not be verified, conflicting information, uncertain dates, ambiguous creators, unclear titles, or rights questions.

## Do Not Use / Rejected Values

List any tempting metadata values that were excluded because they were unsupported, inferred, too broad, or ambiguous.