Framework: Prompt Audit Checklist
A 15-point checklist for evaluating any prompt before putting it into production. Catches the most common prompt failures: vague instructions, missing constraints, absent error handling, and untested edge cases.
Use cases
Operations & Workflow, Strategy & Planning, Development & Code
Platforms
Model-Agnostic
Jump to a section
Workspace-ready resource
Copy the raw material, then check the setup, output contract, and failure modes.
# Prompt Audit Checklist Run this checklist before deploying any prompt into a production workflow,client deliverable, or recurring task. Each item is pass/fail. --- ## Clarity (Does the model know what to do?) ☐ 1. ROLE IS SPECIFIC The prompt defines who the model is in concrete terms. Pass: "You are a B2B SaaS copywriter" Fail: "You are a helpful assistant" ☐ 2. TASK IS UNAMBIGUOUS A human reading the prompt could complete the task without asking clarifying questions. If they would need to ask "but what format?" or "how long?" — the prompt fails this check. ☐ 3. OUTPUT FORMAT IS DEFINED The prompt specifies what the output should look like: length, structure, format (JSON, markdown, prose, table), and any required sections or fields. ☐ 4. SUCCESS CRITERIA ARE STATED The prompt includes at least one measurable or verifiable criterion for what "good output" looks like. Pass: "The summary should be under 100 words and capture all three key arguments" Fail: "Write a good summary" --- ## Constraints (Does the model know what NOT to do?) ☐ 5. BOUNDARIES ARE EXPLICIT The prompt defines what is out of scope, what topics to avoid, what assumptions not to make, or what format not to use. ☐ 6. BANNED PATTERNS ARE LISTED If there are specific words, phrases, or structures the output should not contain, they are explicitly listed. ☐ 7. TONE IS CONSTRAINED The prompt specifies tone (or provides examples), not just a vague adjective. "Formal" is too vague. "Formal: no contractions, no first person, no questions" is a constraint. --- ## Robustness (Does the prompt handle edge cases?) ☐ 8. AMBIGUOUS INPUT IS HANDLED The prompt tells the model what to do when the user's input is unclear, incomplete, or contradictory. Pass: "If the brief is ambiguous, ask one clarifying question before proceeding" Fail: No instruction (model will guess) ☐ 9. UNCERTAINTY IS HANDLED The prompt tells the model what to do when it does not know something or is not confident. Pass: "If unsure, say so and explain what additional information would help" Fail: No instruction (model will fabricate) ☐ 10. LENGTH EDGE CASES ARE HANDLED The prompt works with both very short and very long inputs. Test with a one-sentence input and a 2000-word input. Does it still produce useful output in both cases? --- ## Quality (Does the prompt produce output worth using?) ☐ 11. SELF-CHECK IS INCLUDED The prompt includes a verification step where the model reviews its own output before presenting. At least 3 specific, binary checklist items. ☐ 12. EXAMPLES ARE PROVIDED (if output format matters) For tasks where format is critical (structured data, specific styles, precise layouts), at least one example of the desired output is included in the prompt. ☐ 13. NO PREAMBLE INSTRUCTION The prompt tells the model not to include meta-commentary like "Sure, here's your..." or "I'd be happy to help with..." Just deliver the output. --- ## Production Readiness (Is it ready to deploy?) ☐ 14. TESTED WITH 3+ REPRESENTATIVE INPUTS The prompt has been tested with at least three different inputs that represent the range of real-world use cases. Not just the happy path — include an edge case. ☐ 15. TESTED ON TARGET MODEL The prompt has been tested on the specific model it will run on in production. A prompt that works on Claude may behave differently on GPT or Gemini. Test where it will actually run. --- ## Scoring Count your passes:- 15/15: Production ready. Deploy with confidence.- 12-14: Solid. Fix the gaps before deploying to a client or critical workflow.- 9-11: Needs work. The prompt will produce inconsistent results.- Below 9: Rewrite. The prompt is a draft, not a production asset. --- ## Quick Reference: The Five Most Common Failures 1. No output format defined (model guesses, output varies every time)2. No uncertainty handling (model fabricates instead of flagging gaps)3. No self-check (errors pass through uncaught)4. Untested on target model (works on Claude, breaks on GPT)5. Vague role definition (model defaults to generic assistant behaviour) If you fix only five things, fix these five.Workspace translation
Turn this resource into an inspectable run.
Best next step
Use it as a review standard for the next output you save.
From resource to system
Use it once
Use the framework as a checklist, rubric, or decision aid.
Make it reusable
Attach it as project knowledge so future threads inherit the same criteria.
Decide after the run
Use it as a review standard for the next output you save.
Quality bar
Before using this, check the contract.
What input does this require?
What output should it produce?
Where can it fail?
What should a human review?
Recommended path
When to Use This
Run this checklist on any prompt before it goes into production. "Production" means: a prompt used in a client deliverable, an automated workflow, a customer-facing chatbot, a recurring business process, or any context where the prompt runs repeatedly and the output matters.
Not needed for: one-off questions, casual exploration, brainstorming sessions, or any prompt you will use once and discard.
This is also a useful teaching tool. If you are training a team to write prompts, this checklist makes quality concrete and measurable rather than subjective.
Why It Works
Binary pass/fail forces honest assessment. Checklists with subjective scales ("rate your prompt's clarity from 1-5") produce inflated self-assessments. Binary pass/fail ("does the prompt define the output format? yes or no") is harder to fudge.
The four categories (Clarity, Constraints, Robustness, Quality) cover the full failure surface. Most prompt failures fall into one of these categories. Clarity failures mean the model does not understand the task. Constraint failures mean the model does things you did not want. Robustness failures mean the prompt breaks on edge cases. Quality failures mean the output is not good enough.
The scoring threshold provides a deployment decision. Rather than leaving the evaluation subjective ("I think this prompt is good enough"), the score gives a clear signal. Below 9: rewrite. 12-14: fix the gaps. 15: deploy. This is especially useful for teams where "good enough" means different things to different people.
The "Five Most Common Failures" section is the 80/20. If someone does not have time for the full 15-point checklist, fixing these five issues addresses the majority of real-world prompt failures. It is a practical concession to time pressure without abandoning quality.
How to Customise
Add domain-specific checks. If you work in a regulated industry, add: "☐ 16. COMPLIANCE REQUIREMENTS MET — The prompt includes instructions about [data handling / disclaimer language / prohibited claims]." If you work with code, add: "☐ 17. CODE OUTPUT IS TESTABLE — Any generated code includes error handling and can be run without modification."
Create a simplified version for low-stakes prompts. Not every prompt needs 15 checks. For internal, non-critical prompts, a 5-item version (items 1, 2, 3, 8, 14) covers the essentials.
Add it to your workflow tooling. Turn this into a template in Notion, a checklist in Linear, or a form in your project management tool. Make it part of the process rather than something people remember to do.
Limitations
A checklist ensures minimum quality, not maximum quality. A prompt can pass all 15 checks and still produce mediocre output if the underlying task design is wrong or if the model simply is not capable of the task at the required quality level. The checklist catches preventable errors. It does not make a fundamentally flawed approach work.
The "tested with 3+ representative inputs" requirement is the minimum. For high-stakes prompts (customer-facing, revenue-impacting), test with 10-20 inputs across the full range of scenarios.
Model Notes
This framework is model-agnostic. It applies to prompts for any LLM. The key model-specific check is item 15: always test on the model you will actually use in production. Cross-model behaviour differences are real and often significant enough to break a prompt that works perfectly on a different model.
Related Resources
Browse FrameworksMeta-Prompt: Generate Custom System Prompts
A prompt that generates system prompts. Describe what you need an AI to do, and this meta-prompt produces a structured, production-ready system prompt following best practices.
Pattern / playbook seed
Operations & Workflow · Strategy & Planning
System Prompt: Content Writer
A production-ready system prompt for configuring any LLM as a content writer with tone control, format awareness, and a built-in self-check.
Pattern / playbook seed
Content & Writing · Marketing & Growth
Framework: AI Tool Evaluation Matrix
A structured decision matrix for evaluating AI tools before committing. Scores tools across seven weighted criteria to cut through marketing hype and make informed choices.
Knowledge / rubric seed
Strategy & Planning · Operations & Workflow
Related Guides
Need this operationalized?
Turn the pattern into a workspace system.
Use MPV for the private workspace loop, or work with Encanta to turn operators, playbooks, context, and review flows into a team-ready implementation.