Knowledge Templates for GPTs (Markdown & CSV)
Give your GPT clean, predictable files. These templates chunk well, quote well, and reduce hallucinations.
Last updated:
Starter Pack (file tree)
/knowledge/ 01-product-overview.md 02-pricing-policy.md 03-api-auth.md 04-api-endpoints.md 05-troubleshooting-faq.md 06-sop-template.md 07-release-notes.md 08-glossary.md 09-data-dictionary.csv 10-changelog-decisions.md
Tip: keep to ≤ 20 files per GPT. Use clear numbers and topics.
Templates (copy/paste)
1) Product Overview (01-product-overview.md)
# Product Overview
## What it does
One short paragraph.
## Who it's for
- Audience 1
- Audience 2
## Key capabilities
- Capability A — one line
- Capability B — one line
## Limitations / Non-goals
- Not for {case}
## Quick links
- Docs: https://...
- Support: https://...
2) Pricing Policy (02-pricing-policy.md)
# Pricing Policy
## Plans
| Plan | Price | Includes |
|---|---:|---|
| Basic | $X | Feature 1, 2 |
| Pro | $Y | Feature 1, 2, 3 |
## Discounts
- Student: X%
- Annual: X%
## Billing rules
- Proration: {rule}
- Refunds: {policy}
## FAQs
Q: {question}
A: {answer}
3) API Authentication (03-api-auth.md)
# API Authentication
## Overview
Short summary of auth method.
## Steps
1. Generate API key at {url}
2. Send header: `Authorization: Bearer `
## Error codes
- 401: Invalid key — rotate key at {url}
- 403: Missing scope — request scope {scope}
## Examples
```bash
curl -H "Authorization: Bearer $KEY" https://api.example.com/ping
```
4) API Endpoints (04-api-endpoints.md)
# API Endpoints
## Conventions
- Base URL: `https://api.example.com`
- Version: `v1`
## Endpoints
### GET /items
**Purpose:** List items
**Params:** `page`, `pageSize`
**Response:**
```json
{ "items": [{"id":"...","name":"..."}], "page":1, "totalPages":10 }
```
### GET /items/{id}
**Purpose:** Get item by id
**Response:** `{ "id":"...","name":"..." }`
### POST /items
**Body:** `{ "name": "..." }`
**Response:** `{ "id":"...","url":"..." }`
5) Troubleshooting FAQ (05-troubleshooting-faq.md)
# Troubleshooting FAQ
## Installation
Q: {symptom}
A: {fix}
## Authentication
Q: {symptom}
A: {fix}
## Data & Results
Q: {symptom}
A: {fix}
6) SOP / Playbook (06-sop-template.md)
# SOP: {Process Name}
## Goal
One line describing success.
## Preconditions
- Access to {tool}
- Permissions: {role}
## Steps
1. Step one (who, what, expected output)
2. Step two
## Checks & Metrics
- Done when {condition}
- Quality metric: {kpi}
## Rollback
If {failure}, do {action}
7) Release Notes (07-release-notes.md)
# Release Notes
## {YYYY-MM-DD}
### Added
- {item}
### Changed
- {item}
### Fixed
- {item}
## {YYYY-MM-DD}
### Added
- ...
8) Glossary (08-glossary.md)
# Glossary - **Term A** — short definition - **Term B** — short definition - **Acronym** — spelled out, one-line meaning
9) Data Dictionary (09-data-dictionary.csv)
column_name,data_type,description,allowed_values,example user_id,string,Unique user identifier,,u_123 plan,string,Subscription plan,[basic|pro|enterprise],pro created_at,datetime,ISO8601 timestamp,,2025-09-27T12:00:00Z price_usd,number,Price in USD,,19.00
10) Changelog & Decisions (10-changelog-decisions.md)
# Changelog & Decisions
## {YYYY-MM-DD} — Change title
**Decision:** {what changed and why}
**Context:** Links to issues/PRDs
**Impact:** What users should expect
**Next:** Follow-up tasks
Instruction Block (force quotes + headings)
Use the uploaded Knowledge as the primary source. When answering, quote the most relevant section and include the file and heading like [04-api-endpoints.md › GET /items]. If the answer isn’t present, say so before general guidance.
Testing: Golden Questions File
# Golden Questions
- Q: {common user question}
A: {expected key points}
- Q: {edge case}
A: {expected key points}
- Q: {troubleshooting}
A: {expected key points}
Keep this alongside your docs so the GPT can explain the “why” of changes.
FAQ
Do I need front matter (YAML) at the top?
Optional. Clear headings matter more than metadata. If you include front matter, keep it short.
Can I mix multiple topics in one file?
Avoid it. Split by topic so retrieval returns focused quotes and fewer hallucinations.
Are tables better than prose?
Use both. Tables for parameters and fields; prose for explanations and steps.
Should I include images?
Images aren’t indexed as Knowledge text. Keep explanations in text; include alt text if you add images.