A Free Personal AI Prompt Library Template You Can Actually Maintain

Both files are plain text. Import the CSV into Google Sheets, Airtable, or Notion; open the Markdown in any editor.

Table of contents

• Pick a format (Markdown, CSV, spreadsheet, Notion)
• Recommended fields
• Naming conventions
• Tags that actually help you find things
• Variables and placeholders
• Version numbers
• Model notes and test results
• Privacy notes
• Example entries
• Migrating from chat history or scattered docs
• A 15-minute weekly routine
• FAQ

Pick a format

The format matters less than the discipline. Pick the one you will actually open on a Wednesday afternoon.

The CSV template imports cleanly into all four hosted options above. The Markdown template is best if you already keep notes in Obsidian, VS Code, or a Git repo.

Recommended fields

Twelve fields cover almost every real-world case. Do not add a thirteenth without a specific reason.

• ID — stable identifier, never reused (e.g. p-0001).
• Name — short, action-first ("Draft weekly client email").
• Category — one of a small closed list you commit to.
• Tags — 2 to 5 lowercase, comma-separated.
• Prompt — the full text with {{variables}}.
• Variables — the list used, in order.
• Model — the model you last verified this on.
• Version — semantic-ish (1.0, 1.1, 2.0).
• Last tested — ISO date, YYYY-MM-DD.
• Test result — one line: what worked, what broke.
• Privacy — Safe / Redact / Fill at runtime.
• Notes — pairings, warnings, related prompt IDs.

Naming conventions

Pick one convention per field and hold the line.

• IDs use a prefix and a zero-padded number: p-0001, p-0042. Never reuse a retired ID.
• Names start with a verb: "Draft…", "Summarize…", "Explain…", "Rewrite…", "Extract…".
• Never put a date or a model name in the name — those have their own fields.
• Categories stay lowercase and singular ("writing", not "Writings").

Tags that actually help you find things

A tag is only worth adding if it changes what shows up when you filter. "ai" as a tag is useless in an AI prompt library. Good tags describe the job (email, outline, summary), the audience (client, teammate, exec), or the constraint (short, longform, json-output). Cap yourself at five per prompt.

Variables and placeholders

Wrap variables in double braces: {{client_name}}, {{tone}}. Keep the names short, lowercase, and specific enough that you would not confuse two of them at 3pm. Store the template unfilled; fill at runtime. If you find yourself creating "customer-A version" and "customer-B version" of the same prompt, that is a signal to add a variable, not a second entry.

Version numbers

• Start every prompt at 1.0.
• Minor bump (1.0 → 1.1): wording or small structure changes.
• Major bump (1.1 → 2.0): new output shape, new required variables, or a swap to a different model class.
• Never overwrite a major version. Keep the previous entry in an archive tab or folder with the same ID and a -archived suffix in the name.

For the full workflow — branches, comparison tests, rollback — see the companion guide, How to Version and Test AI Prompts.

Model notes and test results

Record the model you tested against and one sentence of what happened. Not a research report — a sticky note.

• "Kept structure; needed to remove one filler sentence."
• "Broke JSON output on Claude; reverted to gpt-4o-mini."
• "Owner detection accurate on 8/10 samples."

A prompt without a "last tested" date is a prompt you cannot trust. Empty fields are permission to redo the test now.

Privacy notes

• Never paste API keys, passwords, or full customer records into a saved prompt body.
• If a prompt needs sensitive input, mark it Privacy: fill at runtime.
• Assume anything saved in a shared workspace is visible to everyone with access. Design accordingly.
• For a shared team library, add ownership and approval on top of these rules — covered in building a shared prompt library for a small team.

Example entries

Three sample entries from the downloadable template:

Full prompts, variables, and privacy flags live in the CSV and Markdown files.

Migrating from chat history or scattered docs

1. Set a filter: only prompts you actually re-used in the last 60 days. Skip everything else.
2. Paste each survivor into the template. Assign an ID, a category, and a version of 1.0.
3. Re-run each one once against its current model. Fill in "last tested" and "test result" as you go.
4. Do not migrate prompts you cannot describe in one sentence. If it is not clear now, it will not be clearer in three months.
5. When you cross ~25 entries in a single file, promote the library to a spreadsheet or Notion database. Keep the ID scheme.

A 15-minute weekly maintenance routine

1. Open the library. Sort by last tested ascending.
2. Re-run the top three oldest prompts on their listed model.
3. Still works? Update the date. Done.
4. Drifted? Bump the version, log a one-line note, fix.
5. Archive anything untouched for 90 days.

This is the whole point of having a library at all: a small, trusted set of prompts you know still work, plus the discipline to keep them that way. If a tool does not make this routine easier, it is not the right tool yet.