How to Create a GPT

GPT Prompt Library (Golden Prompts by Use Case)

Prompts that compile. Copy, tweak placeholders, ship results. Designed for Custom GPTs with Instructions + optional Knowledge + Actions.

Last updated:

How to use this library

Reusable Prompt Patterns

Answer → Steps → Example

Answer my question, then show steps and one concrete example.
Format: 
- **Answer:** one or two sentences.
- **Steps:** short bullets.
- **Example:** a realistic snippet.
Question: {your question here}

Self-check (no private reasoning)

Before finalizing, do a quick accuracy check against your sources.
If Knowledge was used, cite the file + heading like: [pricing.md › Discounts].
Then output only the final answer (no draft notes).

Support Prompts

Bug triage (with ticket Action)

User issue: {symptoms}
Goal: draft steps to reproduce and create a support ticket.
If available, call the Action  with {subject, description, priority}.
Output:
- Summary (1–2 lines)
- Steps to reproduce (bullets)
- Next action for user
- If ticket created, append: (ticket URL)

Policy answer from Knowledge

Answer using Knowledge first. Quote the source heading.
Question: “{policy question}”
Format: 
- Short answer
- Excerpt from [file.md › Heading]
- If not in Knowledge, say “not covered” then give best‑effort guidance.

Teacher/Educator Prompts

Lesson explainer

Explain {concept} to a {audience}.
Use the “explain → example → exercise” flow.
Format: 3 sections with short bullets and a quick quiz (3 questions).

Socratic hints (no spoilers)

The learner is stuck on {problem}.
Give 3 progressively stronger hints without revealing the final answer.
Offer a final checkpoint: “Does your result satisfy {property}?”

Analyst/Data Prompts

CSV insight (code tool)

I uploaded {file}. Find {metric} by {dimension} and show a compact table.
Then provide 3 insights and one anomaly check.
If a plot helps, include a minimal chart.

SQL sketch

Given this schema: {tables/columns}, draft a single SQL query to answer: {business question}.
Return only the SQL. Keep it ANSI‑compatible.

Marketing/Writing Prompts

Feature announcement (short)

Write a 120–160 word announcement for {feature}.
Audience: {audience}.
Format: Hook → three bullets of value → CTA with a link.

Rewrite with tone

Rewrite this paragraph in a {tone} tone.
Keep original facts. 90–140 words. Use scannable sentences.
Content: ```{text}```

Developer Prompts

Action usage (force tool)

Use the  Action with {city} and return only:
{
  "tempC": number,
  "summary": string
}
If the Action fails, return:
{ "error": string, "code": string }

Spec review (lint a schema)

Review this OpenAPI snippet for clarity and minimalism.
Call out: missing types, vague descriptions, oversized responses.
Then output a corrected version only.
```yaml
{openapi}
```

Testing & Debugging Prompts

Knowledge usage check

Answer the question using Knowledge. After the answer, list the file + heading you used like [file.md › Section]. If none, say “no relevant Knowledge.”
Question: {question}

Format compliance

Re-run your last answer so it exactly matches this format:
- One-sentence answer
- Bulleted steps (3–5)
- Example snippet (if applicable)

FAQ

Should I put these prompts in Instructions?

Use prompts during conversations. Keep stable rules (tone, format, sources) in Instructions.

How do I keep outputs consistent?

Specify format, length, and sources. Test with “golden prompts.” Tighten Instructions if drift appears.

Can I force an Action call?

Prompts can request it, but the best place is a decision rule in Instructions + a clear tool description.

What if my prompts are ignored?

Shorten the prompt, remove conflicting directions, and ensure Instructions don’t override it.