ChatGPT Prompts Prompts

General-purpose prompts written for ChatGPT that also work in Claude and Gemini: role, context, constraints and output format in one reusable template.

• Prompt Optimization Engine: Transform any rough idea into an expert-level prompt.
• ChatGPT System Prompt Builder: Write a system prompt that holds behaviour across a long conversation.
• Step-by-Step Reasoning Prompt: Prompt a model to reason through a problem rather than leap to an answer.
• Expert Persona Prompt: Give a model a specific expert perspective rather than a generic one.
• Few-Shot Example Prompt: Teach a model a pattern using well-chosen examples.
• Structured Output Prompt: Get consistent, parseable output from a model every time.
• Prompt Debugging Guide: Diagnose why a prompt is producing the wrong output.
• Summarisation Prompt: Summarise long text for a specific reader and purpose.
• Prompt Chaining Workflow: Break a complex task into a sequence of reliable prompts.
• Creative Writing Prompt: Prompt for writing with a specific voice and set of constraints.
• Roleplay Scenario Prompt: Set up a practice scenario for rehearsing a difficult conversation.
• Prompt Evaluation Rubric: Score prompt output consistently so improvements are measurable.
• Context Optimisation Prompt: Fit the information that matters into a limited context.

Best chatgpt prompts prompts, in full

The top 3 by score on the free Verbito Prompt Scorer.

Structured Output Prompt (score 100/100)

You are an engineer integrating a model into software where output must parse. Build a prompt producing reliable [INSERT FORMAT] output for [INSERT TASK].

Produce:

1. The exact schema, with every field, its type and whether it is required
2. The instruction specifying the format, stated unambiguously
3. A complete example of valid output
4. An instruction to return only the structured output, with no preamble or code fences
5. Handling of fields that cannot be determined: null, empty, or an explicit unknown marker
6. Escaping rules for text containing quotes or newlines
7. An enumerated list for any field with fixed options, since free text will drift
8. Validation the consuming code should perform regardless
9. A fallback for when parsing fails
10. The two fields most likely to be returned inconsistently, and how to constrain them

Always validate on the consuming side. A prompt that produces valid output 99 times in 100 still breaks daily at volume, and the code must handle it rather than assuming compliance.

Tone: exact. Ambiguity in the spec becomes variance in the output.

Audience: the backend developers who parse the output. Context: the goal is zero parse failures across [INSERT EXPECTED DAILY CALLS] calls.
Keep the prompt under 300 words and include the exact JSON schema plus one valid example output.

Summarisation Prompt (score 92/100)

You are a prompt engineer building summarisation for real use. Build a prompt summarising [INSERT CONTENT TYPE] for [INSERT READER AND PURPOSE].

Produce:

1. The reader and what they need the summary to enable
2. Target length, stated in sentences or words rather than a vague brief
3. What must be preserved: figures, names, dates, caveats, dissent
4. What may be dropped
5. Whether the summary must be faithful to emphasis or may reorder by importance
6. An instruction not to introduce information absent from the source
7. Treatment of uncertainty in the source, which must not become certainty in the summary
8. Output structure
9. An instruction to note what was omitted, when that matters
10. A verification step checking every claim against the source

The most damaging summarisation error is flattening a hedged finding into a confident one. Instruct explicitly that qualifiers are preserved.

Tone: precise about fidelity requirements.

Few-Shot Example Prompt (score 91/100)

You are a prompt engineer who uses examples instead of lengthy instructions. Build a few-shot prompt for [INSERT TASK].

Produce:

1. The task instruction, kept brief because the examples carry the pattern
2. Three to five examples in consistent format
3. Example selection reasoning: cover the range, including one edge case
4. The formatting convention marking input and output, applied identically throughout
5. An example showing what to do with ambiguous input
6. Why these examples and not others
7. The ordering, since the final example carries disproportionate weight
8. The template for the actual input
9. How to tell from the output whether the pattern was learned or merely copied
10. When few-shot is the wrong approach and an explicit rule would be clearer

Examples that are too similar teach a narrower pattern than intended. Vary them deliberately along the dimensions that should not matter.

Tone: demonstrative. Let the examples do the instructing.

How to use these prompts

1. Copy the prompt that matches your task.
2. Replace every [INSERT ...] part with your own details: the more specific, the better the answer.
3. Paste it into ChatGPT, Claude or Gemini. To tailor it to your exact situation, describe your goal in the Verbito prompt generator instead.

Build a chatgpt prompts prompt for your task · The 100 best prompts in one kit

Create a prompt or contact Verbito.