P
AI for Archivist

Custom GPT: Build an Archivist's Writing and Research Assistant

ChatGPT

For Archivists

Tools: ChatGPT | Time to build: 1-2 hours | Difficulty: Intermediate-Advanced Prerequisites: Comfortable using ChatGPT for archival writing tasks. See Level 3 guide: "Use ChatGPT to Write Archival Grant Proposals"

What This Builds

A Custom GPT is a specialized version of ChatGPT pre-configured with your institution's finding aid standards, writing style, and common reference materials. Anyone on your team can open it and immediately get archival-quality output without any setup. Unlike a regular ChatGPT conversation, a Custom GPT remembers its configuration permanently and can be shared with colleagues via a link or made available through your institution's ChatGPT Teams account.

Prerequisites

  • A ChatGPT account with access to Custom GPTs; requires {{tool:ChatGPT.plan}} ({{tool:ChatGPT.price}}) or higher
  • 2–3 example finding aids from your institution in PDF or DOCX format
  • Your institution's processing manual or style guide (if written down)
  • 1–2 hours for initial setup and testing

The Concept

Think of a Custom GPT as a new hire who has already read your entire style guide and processed several practice collections before their first day. When a colleague opens your Custom GPT, they're not starting from scratch. They're working with something that already knows your standards, your voice, and your common tasks. You set it up once; everyone on your team benefits.

Build It Step by Step

Part 1: Gather your configuration materials (20-30 minutes)

Collect these files before starting:

  1. 2–3 complete finding aids from your institution (PDFs or DOCX): these are the best examples of your institutional voice
  2. Finding aid template: Your blank template showing structure and field names
  3. Processing manual or style notes: Written guidance on scope note length, folder title format, subject heading practices
  4. Common reference response templates: If you have any saved email templates

If you don't have a written style guide, take 10 minutes to write down your most important rules (scope note word count, how you handle folder titles, when you include biographical notes, etc.) in a simple text file.

Part 2: Open the Custom GPT builder (30-45 minutes)

  1. Log into ChatGPT at {{tool:ChatGPT.url}}
  2. Click your profile icon (top right) → My GPTs
  3. Click Create a GPT
  4. The builder shows two panels: a chat interface on the left (where you configure by conversation) and a preview on the right

Use the Configure tab for precise setup: Click Configure at the top to switch to direct configuration.

Fill in these fields:

Name: "[Institution Name] Archival Writing Assistant" (or shorter: "Archival Description Assistant")

Description: "Helps archivists at [Institution Name] draft finding aid descriptions, grant narratives, reference responses, and other professional archival writing following DACS standards."

Instructions (paste this, customized for your institution):

Copy and paste this
You are an archival writing assistant for [Institution Name], a [type of institution].

CORE FUNCTION: Help archivists draft DACS-compliant finding aid descriptions, grant application narratives, reference inquiry responses, exhibit labels, policy documents, and volunteer training materials.

STANDARDS: Follow DACS (Describing Archives: A Content Standard). Past tense for historical content. Formal but accessible prose.

COLLECTION FOCUS: [Describe your main collection areas — e.g., "local history of [county], personal papers of community leaders, records of nonprofit organizations"]

WRITING CONVENTIONS:
- Collection-level scope notes: 150–200 words
- Series descriptions: 2–3 sentences
- Biographical notes: begin with name and dates, ~150 words
- Folder titles: use format "[Topic/Name], [Date or Date Range]"
- Subject headings: suggest LCSH terms; flag any that may not exist

GRANT WRITING: When asked to draft grant narratives, ask about funder (NHPRC, IMLS, NEH, private), word limits, and evaluation criteria before drafting.

REFERENCE RESPONSES: When drafting reference responses, ask about relevant holdings, access procedures, and any restrictions before drafting.

ACCURACY: If asked for specific facts (dates, organizations, LCSH headings) that you're uncertain about, say so clearly rather than guessing. The archivist will verify.

UPLOADED FILES: I've uploaded example finding aids and processing guidance. Use these as models for format and voice.

Conversation starters (optional, adds quick-start buttons):

  • "Draft a scope note from my processing notes"
  • "Write a grant narrative section"
  • "Draft a reference response"
  • "Suggest subject headings for this collection"

Part 3: Upload your knowledge files

Still in the Configure tab, scroll down to Knowledge.

Click Upload files and upload:

  • Your 2–3 example finding aids (PDF or DOCX)
  • Your finding aid template
  • Your processing manual/style notes (if in a file)

These files are permanently attached to this GPT; anyone who uses it gets the benefit of these examples without knowing they're there.

Part 4: Test your Custom GPT (20-30 minutes)

Click Save and then View GPT to open it.

Test with three scenarios:

  1. Paste processing notes and ask for a scope note: does the output match your institutional style?
  2. Describe a grant project and ask for a project narrative: does it ask clarifying questions?
  3. Paste a reference inquiry and ask for a response: does it ask about your holdings first?

If the output is off, go back to the Configure tab and adjust the Instructions. Common fixes:

  • Too generic → Add more specific style rules
  • Too formal/informal → Adjust the PROSE STYLE note
  • Doesn't ask clarifying questions → Strengthen the "ask before drafting" instructions for grant and reference tasks

Real Example: Using the GPT on a Monday morning

Setup: Your "Archival Description Assistant" GPT is ready. A colleague opens it for the first time.

Their input: "I finished processing the Henderson Construction Company Records. 15 linear feet, 1952–1998. Contains contracts, project files organized by job site, correspondence with clients and subcontractors, and financial records. They built most of the downtown buildings in this city during urban renewal. Write a scope note."

Output from your Custom GPT: A 175-word scope note that opens with the company name and date range, describes the record types in your preferred order, references the urban renewal context in the way your institutional style suggests for business records, and closes with a note about researcher value. All in language that sounds like your other finding aids.

Colleague's reaction: "Oh, this is already set up for how we write things. I just had to change a couple of dates."

Time saved: 90 minutes of scope note writing → 15 minutes of review and editing.

What to Do When It Breaks

  • Style doesn't match your institution → Upload one more example finding aid and add a specific rule to the Instructions ("Our scope notes always begin with 'The [Collection Name] consists of...'")
  • GPT makes up LCSH headings → Add to Instructions: "Never invent LCSH headings. Suggest candidates and explicitly note they must be verified at id.loc.gov"
  • Colleagues find it too complicated to use → Simplify the Conversation Starters and add a one-paragraph "How to use this GPT" as the first line of the Description
  • Output quality declines in long conversations → Start a new conversation for each new collection; the GPT's configuration persists but long conversations can drift

Variations

  • Simpler version: Use the Instructions block only (no uploaded files) for a lighter-weight assistant that's easier to maintain as your style evolves
  • Extended version: Add your institution's entire finding aid database as uploaded files so the GPT can check for consistency with how you've described related collections

What to Do Next

  • This week: Use the GPT for every piece of archival writing and refine the instructions based on what works and what doesn't
  • This month: Share the GPT link with your full staff via ChatGPT Teams or a shared login
  • Advanced: Create separate GPTs for different task types (one for finding aids, one for grant writing, one for reference responses) if your team's needs diverge

Advanced guide for archivist professionals. Custom GPTs require a ChatGPT paid subscription.