The CREATE Framework: How to Write Better AI Prompts Step by Step
When an AI prompt leaves important details unstated, the response can miss the mark even when it addresses the right topic.
“Write a marketing plan” is a valid instruction, but it leaves the audience, objective, depth, format, constraints, and standard for a useful answer open to interpretation.
The CREATE Framework gives those decisions a structure. It organizes a prompt into six components: Character, Request, Examples, Adjustments, Type of Output, and Extras.
This guide explains what each component does, why the framework is useful, how the parts work together, and how to build a CREATE prompt without adding unnecessary complexity.
Key Takeaways
- CREATE stands for Character, Request, Examples, Adjustments, Type of Output, and Extras.
- The framework organizes the information that shapes an AI response instead of treating prompting as one loose instruction.
- Request defines the task. The other components add perspective, reference points, constraints, format, and supporting context.
- A strong CREATE prompt is specific where specificity matters and brief where extra detail adds little value.
- Not every task needs all six components. Simple requests can often remain simple.
- CREATE can make a prompt clearer and easier to control, but it cannot guarantee that an AI response is factually correct.
What Is the CREATE Framework?
The CREATE Framework is a six-part method for structuring prompts for generative AI. AI educator Dave Birss teaches the framework as Character, Request, Examples, Adjustments, Type of Output, and Extras.
Its purpose is to separate the main parts of an instruction so they can be controlled deliberately. Instead of mixing the task, tone, examples, formatting, and constraints into one loose prompt, CREATE gives each type of information a defined role.
Request sits at the center because it states the work to be done. Character establishes a useful perspective when one is needed. Examples provide reference points. Adjustments refine execution. Type of Output defines the form of the response. Extras hold relevant directions that do not fit naturally elsewhere.
CREATE is therefore better understood as a briefing structure than a formula for making prompts longer. It helps the user decide which instructions matter before the AI responds.
Why the CREATE Framework Matters
Prompting becomes harder when important expectations remain implicit.
A broad request gives an AI system more room to interpret tone, scope, audience, depth, structure, and priorities. The resulting answer may still be relevant, but it can require substantial correction before it becomes useful.
CREATE brings those decisions into the prompt. Instead of fixing several problems after the first response, the user can define the important requirements in advance.
More detail is not always better. Instructions that do not affect the task can become noise, especially when they repeat or conflict. The practical value of CREATE is selective clarity: state what materially shapes the result and leave out what does not.
This logic is consistent with OpenAI’s prompting guidance, which recommends clear, specific instructions, relevant context, and explicit direction about the desired outcome and format.
What Does CREATE Stand For?
Each CREATE component addresses a different source of ambiguity. Birss defines the six elements as Character, Request, Examples, Adjustments, Type of Output, and Extras.
C: Character
Character defines the perspective or role the AI should use when approaching the task.
“Act as an SEO content strategist” creates a different frame than “act as a high school teacher.” Even with the same subject, the vocabulary, priorities, assumptions, and level of explanation can change.
Keep the role relevant and restrained. A useful Character establishes the perspective needed for the task without adding an elaborate fictional biography.
Some prompts do not need a Character. Straightforward definitions, conversions, or extraction tasks may already be clear without one.
R: Request
Request states the job to be completed.
This is the core instruction, so it should identify both the action and the subject. “Help me with email marketing” is broad. “Create a beginner email marketing plan for a local service business” defines a clearer outcome.
Useful specificity resolves meaningful uncertainty. It does not require packing every possible detail into one sentence.
If the Request is weak, the remaining components may improve presentation while leaving the underlying task unclear.
E: Examples
Examples provide reference points for the response.
They are useful when a desired style, structure, classification, or quality standard is easier to demonstrate than describe. A sample headline can clarify what “specific but not sensational” means. Showing a table structure can establish the exact fields an output should contain.
Choose examples that genuinely represent the target. Irrelevant or conflicting references can introduce new ambiguity.
You can omit this component when the desired result is already obvious.
A: Adjustments
Adjustments refine how the Request should be carried out.
Tone, audience, length, scope, reading level, exclusions, and priorities can all belong here when they materially affect the answer.
For example, “write for first-time business owners, avoid jargon, and keep each recommendation practical” changes the execution without changing the underlying task.
Concrete adjustments are easier to follow than subjective ones. “Shorten the introduction, remove repetition, and explain technical terms in plain language” gives more direction than “make it better.”
T: Type of Output
Type of Output specifies the form the response should take.
A comparison could appear as prose, a table, a scorecard, or a recommendation memo. Stating the required format reduces the chance of receiving useful information in a structure that creates extra work.
The instruction can also define what belongs inside that format. An article request might require an introduction, key takeaways, H2 sections, a conclusion, and FAQs. A comparison table might require fixed columns.
Format should serve the information. Tables work well for side-by-side attributes, while explanations that depend on context are usually clearer in paragraphs.
E: Extras
Extras contain relevant directions that do not sit naturally inside the other five components.
A required keyword, source restriction, business constraint, exclusion, or piece of background context can fit here when it materially changes the answer.
For example, a travel prompt might specify that the traveler will not rent a car. A content prompt might require the writer to avoid unsupported statistics.
Use Extras selectively. If an instruction does not affect the usefulness, accuracy, or usability of the response, it probably does not belong in the prompt.
How the CREATE Framework Works
CREATE works by separating types of control that are often blended together in ordinary prompts.
First, the framework defines the assignment. Request establishes the outcome, while Character can shape the perspective used to approach it.
Next, it narrows interpretation. Examples and Adjustments communicate the expected standard, audience, boundaries, and execution.
Then, it controls delivery. Type of Output determines how the answer should be organized for its intended use.
Finally, Extras capture important context that could otherwise be overlooked.
A long prompt is not required. What changes is the visibility of the decisions influencing the answer, which makes them easier to see, test, and revise.
This structure also makes troubleshooting more precise. When a response misses the mark, you can examine the task, references, constraints, format, and context separately instead of rewriting the entire prompt at random.
CREATE Framework Example
Consider this basic prompt:
“Write a blog post about affiliate marketing.”
The topic is clear, but the assignment is not. Nothing identifies the reader, the specific problem, the angle, the level of detail, or the final structure.
A CREATE version can make those decisions explicit.
Character: Act as an experienced affiliate marketing editor.
Request: Write a beginner-friendly article explaining why affiliate links can receive clicks without generating sales.
Examples: Cover audience mismatch, low buying intent, bot or low-quality traffic, merchant conversion problems, and tracking issues.
Adjustments: Use concise language. Avoid exaggerated claims. Explain each reason in plain language and focus on diagnosis rather than generic traffic advice.
Type of Output: Write a blog article with an introduction, key takeaways, five H2 sections, a conclusion, and FAQs.
Extras: Make each section distinct. Do not repeat the same explanation under different headings.
Combined, the prompt becomes:
Act as an experienced affiliate marketing editor. Write a beginner-friendly article explaining why affiliate links can receive clicks without generating sales. Cover audience mismatch, low buying intent, bot or low-quality traffic, merchant conversion problems, and tracking issues. Use concise language, avoid exaggerated claims, and focus on diagnosis rather than generic traffic advice. Format the response as a blog article with an introduction, key takeaways, five H2 sections, a conclusion, and FAQs. Make each section distinct and avoid repeating the same explanation under different headings.
Length is not the main advantage. The revised prompt provides a clearer assignment, quality bar, scope, and deliverable.
How to Use the CREATE Framework Step by Step
A practical CREATE prompt starts with the task and adds structure only where it improves control.
- Write the Request first. State the outcome you need in one clear sentence. This creates an anchor for every other instruction and exposes uncertainty before more detail is added.
- Choose a relevant Character. Add a role when perspective or expertise should influence the response. Keep it specific to the job rather than building an unnecessary persona.
- Add Examples when the standard is hard to describe. Use a reference to demonstrate a pattern, style, structure, or threshold. One strong example may be enough.
- Set the Adjustments. Define the audience, tone, depth, exclusions, priorities, length, or other constraints that affect execution. Prefer concrete directions over broad adjectives.
- Specify the Type of Output. Choose the format that makes the result immediately usable. When structure matters, state the required sections, fields, or organization.
- Add necessary Extras, then edit. Include remaining context that changes the answer. Remove duplicate instructions, resolve conflicts, and cut details that do not influence the result.
The finished prompt should feel complete, not crowded.
Common CREATE Framework Mistakes
The framework becomes less useful when structure turns into excess.
- Overbuilding the Character: A long persona can distract from the task. Use enough detail to establish perspective and stop there.
- Leaving the Request vague: Formatting, examples, and tone cannot compensate for an unclear objective.
- Using examples that point in different directions: Reference material should clarify the target, not create competing standards.
- Writing subjective Adjustments: Directions such as “make it amazing” or “sound better” are difficult to apply consistently. Define what improvement means.
- Choosing a format that does not fit the information: A table suits comparable attributes, but it is poor for nuanced explanation that depends on context.
- Loading Extras with minor preferences: Too many secondary instructions can bury the requirements that matter.
- Repeating the same rule in several components: State an instruction once in the place where it fits best.
- Treating CREATE as mandatory: The framework is useful when it adds clarity. A direct prompt can be better for a simple task.
When the CREATE Framework Makes Sense
CREATE is most useful when a task has enough moving parts that unstated assumptions could change the result.
It fits situations that require a defined audience, perspective, structure, tone, quality standard, or set of constraints. Content briefs, research tasks, comparisons, analyses, reusable workflows, and complex drafting assignments are natural examples.
Reusable prompts are another strong fit. Separating the task from examples, adjustments, output requirements, and extras makes individual components easier to update without rebuilding the entire instruction.
Simple tasks need less machinery. “Summarize this paragraph in two sentences” already contains a clear Request and Type of Output. Adding four more components may not improve it.
The practical rule is proportionality. Use enough structure to remove meaningful ambiguity, but no more.
CREATE Framework vs. a Basic Prompt
Both approaches can be effective. The right choice depends on how much control the task requires.
| Factor | Basic Prompt | CREATE Prompt |
| Best for | Simple, obvious tasks | Complex or constrained tasks |
| Setup | Fast | Requires more planning |
| Context | Usually limited | Can be deliberately structured |
| Output control | Often minimal | Can define tone, format, examples, and constraints |
| Reusability | Depends on the task | Easier to standardize when components are clear |
| Main risk | Too much left to interpretation | Too many unnecessary instructions |
A basic prompt is not an inferior prompt. If the task is already clear, simplicity is a strength.
CREATE becomes more useful as the number of important decisions grows. The higher the cost of ambiguity, the more valuable a structured brief can become.
Conclusion
The CREATE Framework organizes the information that matters in an AI prompt.
Character sets a useful perspective. Request defines the work. Examples establish reference points. Adjustments refine execution. Type of Output determines the final form. Extras capture relevant context that remains.
Its value comes from control, not complexity.
Use all six components when the task needs them. Leave components out when they add nothing. The strongest prompt is not the one with the most instructions, but the one that makes the important instructions clear.
Frequently Asked Questions
What Does CREATE Stand For in AI Prompting?
CREATE stands for Character, Request, Examples, Adjustments, Type of Output, and Extras.
Who Created the CREATE Framework?
The framework is associated with AI educator Dave Birss, who teaches CREATE as a method for structuring prompts for generative AI.
Do I Need to Use Every Part of the CREATE Framework?
No. Use the components that reduce meaningful ambiguity for the task. A simple prompt may need only a Request and a Type of Output.
What Is the Most Important Part of a CREATE Prompt?
Request is usually the foundation because it defines the task. The remaining components shape how that task should be interpreted, executed, and delivered.
Is the CREATE Framework Only for ChatGPT?
No. The structure can be applied to generative AI tools that accept natural-language instructions.
Does the CREATE Framework Guarantee Better AI Answers?
No. It can help you give clearer instructions and produce a more usable response, but it cannot guarantee factual accuracy, sound reasoning, or a correct result.
How Long Should a CREATE Prompt Be?
There is no ideal length. Include the details that materially affect the task, then remove repetition and unnecessary instructions.
Leave a Reply