Prompting
AI prompt frameworks: how to structure a prompt and test that it works
A prompt framework is a repeatable structure for writing instructions. It helps most when the team also agrees how to test the result.
AI prompt frameworks are repeatable structures for writing a prompt, such as instruction, context, examples and the output you want. Frameworks go by different names, and the vendors document similar parts. Microsoft lists instructions, primary content, examples, cues and supporting content as common prompt components. Anthropic says to settle three things before tuning a prompt: clear success criteria, a way to test against them, and a first draft. A framework gives a team a shared starting shape for every prompt it writes. Testing shows whether the prompt works.
plain answer
What is an AI prompt framework?
A prompt framework is a template for the parts of a prompt and the order to write them in. Many teams use their own named version. Because the names vary and no single one is standard, this page describes the parts the vendors themselves document rather than endorsing one acronym.
Microsoft's prompt engineering guide says that "prompt construction can be difficult" and is "more of an art than a science, often requiring experience and intuition." A framework reduces the guesswork. It does not remove the need to check the output.
framework
The parts Microsoft lists for a prompt
Instructions
Microsoft calls these "likely the most commonly used prompt component": what you want the model to do. Specific instructions leave less to interpretation.
Primary content
The text the model processes or transforms, for example the document to summarize.
Examples
Microsoft describes one-shot and few-shot prompting as including "one or more examples of the desired behavior of the model," typically as input and output pairs.
Cues
Microsoft describes cues as acting "as the jumpstart for the output of the model," often a prefix the model builds on.
Supporting content
Context the model can use to shape the output, such as the date or user preferences, which is not the main object of the task.
Output structure and citations
Microsoft says specifying the output structure can have a significant effect on the quality of the results, and that asking for citations can reduce incorrect responses.
how to
Test the prompt, not just the template
Define success first
Anthropic's prompt engineering overview assumes you have "a clear definition of the success criteria for your use case."
Have a way to test
The same page assumes "some ways to empirically test against those criteria," and points to its guidance on defining success criteria and building evaluations.
Start from a draft
Anthropic then assumes "a first draft prompt you want to improve" and lists techniques such as clarity, examples, XML structuring, role prompting, thinking and prompt chaining in its Prompting best practices guide.
Do not assume it carries over
Microsoft cautions that "just because a carefully crafted prompt worked well for a particular scenario doesn't necessarily mean it will generalize more broadly," and that you still need to validate responses.
Check the model
Microsoft notes its techniques are not recommended for reasoning models such as gpt-5 and the o-series, and that each model behaves differently. Re-test when the model changes.
ai gap
Where prompt frameworks go wrong
One way this goes wrong is treating the framework as the finish line. A team adopts a template, writes every prompt in it and never tests any of them, so a polished prompt produces an unchecked answer.
A second way is prompts that live in one person's notes. When that person leaves or the model changes, nobody knows which prompts exist, who owns them or whether they still work. Keep a shared library with an owner, a purpose and a test date for each prompt.
differentiation
How this differs from our related pages
This page is a guide to prompt frameworks and testing them. Our Prompt Writing and Instruction Design course is a training course on writing instructions. Our AI training for employees page covers training across tools and roles, and our AI governance training page covers governance courses.
kriv role
What Kriv AI does
We help teams choose a prompt structure, set success criteria and tests, and keep a shared prompt library with owners and review dates. We train people to write and test prompts, and we help set the rules for prompts that touch sensitive data. Every engagement is scoped and priced after a short discovery call.
Straight answers
Frequently asked questions about AI prompt frameworks: how to structure a prompt and test that it works
What is a prompt framework, or prompt engineering framework?
A repeatable structure for writing a prompt, covering parts such as instructions, primary content, examples, cues, supporting content and the output format you want.
Is there one standard prompt framework?
No. Teams use differently named versions that share similar parts. The vendor guides from Microsoft and Anthropic describe techniques rather than a single required template.
How do we know a prompt works?
Define success criteria first and test against them. Anthropic assumes you have success criteria, a way to test and a first draft before you tune a prompt.
Do prompt techniques work on every model?
Not necessarily. Microsoft says each model behaves differently and that some techniques are not recommended for reasoning models, so re-test when the model changes.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call