What Prompertize does
Prompertize turns a rough request into a structured AI prompt before you copy it. It helps you define the task, background, voice, audience, format, guardrails, and evidence boundaries so ChatGPT, Claude, Gemini, Copilot, and other AI tools have a clearer brief.
Prompertize never sends your prompt anywhere by itself. Copy places the prompt on your clipboard, and you paste it into the AI tool of your choice. The Open in buttons below the preview send the prompt to that provider only when you click them: ChatGPT, Claude and Perplexity open with the prompt in the link (short prompts), while Gemini and long prompts copy the prompt and open the site for you to paste. You still decide when to send.
Fast workflow
- Use Quick mode when you have a rough idea and want Prompertize to shape it.
- Use Pro mode when you want full control over each framework field.
- Pick a Framework, Target AI, and optional Recipe.
- Fill the starred fields. Optional fields improve precision but are not required.
- Review Readiness, Safety, and the Before And After preview.
- Click Copy, then paste the prompt into your AI tool.
Modes
Easy Mode
Best for less technical users. It hides advanced controls, keeps one large request box, and lets the user choose a simple recipe. Prompertize still adds structure behind the scenes.
Quick Mode
Best for everyday use. Paste a loose request, notes, source text, or an unfinished prompt. Prompertize combines it with your selected framework, recipe, target AI, voice, and response format.
Pro Mode
Best for important work. You fill the framework fields directly, which gives you tighter control over context, objective, constraints, examples, output shape, and system-level behavior.
Token impact
A structured prompt may use more input tokens than a rough request. The savings usually come from fewer clarification turns, fewer rewrites, and more focused model responses.
Prompt Tokens
The rough request is often shorter. The Prompertized prompt adds structure, field labels, constraints, examples, and output instructions, so it may be longer.
Iteration Savings
The value is avoiding back-and-forth. Clear context, objective, audience, response format, and guardrails reduce the chance that you need to correct or rerun the answer.
Efficiency: Standard, Lean, Ultra
Standard keeps the full structure. Lean leaves out generic text: the framework label, the default guardrails (replaced by one short line) and long model-tuning notes, and adds a short "answer first" line. Ultra also writes the prompt as plain "Label: value" lines and adds a brevity line. Guardrails, limits and clarifying-question lines that you or a recipe wrote are never shortened. The token line under the selector shows the size of your request, the size of the prompt at this level, and how that compares with Standard. Estimates cover the prompt only, not the reply, and AI tokenizers differ. Reply length (under Advanced) adds one line asking for a shorter reply; the AI may not follow it exactly.
How To Read The Metric
The token line under the Efficiency selector compares your rough request, the prompt at the chosen level, and Standard. It does not estimate retries avoided. Treat it as guidance because each AI uses its own tokenizer.
Best Way To Save Tokens
Keep context relevant, remove repeated notes, use a precise response format, and ask for the exact deliverable instead of asking the model to infer what matters.
Prompt frameworks
CO-STAR
Context, Objective, Style, Tone, Audience, Response. Use it when you need a complete prompt brief with a clear situation, task, voice, reader, and output format.
RACE
Role, Action, Context, Expectation. Use it for professional tasks where the AI should act in a specific role and satisfy a concrete quality bar.
RISEN
Role, Instructions, Steps, End goal, Narrowing. Use it when the task needs careful boundaries, ordered reasoning, exclusions, or compliance-style constraints.
CRAFT
Context, Role, Action, Format, Tone. Use it for deliverables such as reports, emails, plans, summaries, briefs, and polished business writing.
TAG
Task, Action, Goal. Use it for short prompts when you need speed and only want to specify what to do and what success looks like.
Custom
Define your own framework name and rules. Use it for team templates, compliance templates, research methods, teaching formats, or any repeatable prompt structure.
NIST SP 1353 IPD uses CO-STAR in illustrative AI prompt examples for CSF 2.0 analysis and reporting. NIST states those examples are not endorsements.
Core fields
System Prompt
Persistent instructions for how the AI should behave. Use it for policies, evidence rules, refusal boundaries, persona, or output discipline. Save reusable versions in System Prompt Memory.
Context
Background the model needs: source text, facts, business situation, constraints, assumptions, definitions, stakeholders, and what has already happened.
Objective, Task, Action, Or End Goal
The exact work you want done. Strong objectives use action verbs and define success: summarize, compare, draft, classify, explain, generate, audit, plan, transform, or critique.
Role
The expert lens the AI should use, such as project manager, teacher, cybersecurity analyst, product designer, reviewer, editor, or data analyst.
Style
The writing or presentation style: executive memo, technical documentation, plain-language explainer, academic, social post, creative direction, or your custom style.
Tone
The emotional and professional register: concise, friendly, formal, persuasive, empathetic, critical but fair, humorous, technical, or creative.
Audience
The people who will read or use the response. Examples include executives, beginners, customers, developers, students, designers, or technical experts.
Response
The final output format. Be specific when the answer must be machine-readable, visual, printable, presentation-ready, or ready to paste into another tool.
Response types
Text And Structured Outputs
Use Markdown report, bulleted list, JSON object, table, step-by-step plan, email draft, social post, or checklist when the final answer is primarily text or structured data.
Image Generation Prompt
Use for prompts that will be pasted into an image model. Include subject, composition, style, color palette, lighting, aspect ratio, mood, detail level, and negative constraints.
Logo Image Prompt
Use for brand marks. Include brand name, symbol concept, typography direction, colors, background, usage context, variants, and a rule to avoid existing logos or trademark confusion.
Media Prompt
Use for broader creative assets such as banners, thumbnails, product shots, ads, mockups, posters, or campaign visuals. Specify platform, dimensions, visual hierarchy, and required text.
Video Prompt
Use for video generation or production briefs. Include scene order, camera movement, pacing, characters, setting, transitions, narration, duration, aspect ratio, and brand constraints.
PDF-Ready Report
Use when the output should become a document. Ask for title, summary, section headings, evidence, tables, recommendations, appendix notes, and clean Markdown that exports well.
Slide Deck Outline
Use for presentation planning. Ask for slide titles, key messages, speaker notes, charts or visuals, transitions, and a clear recommendation or closing action.
Infographic, Storyboard, Design Spec
Use these when another person or design tool will turn the response into a visual. Ask for layout, sections, hierarchy, callouts, colors, assets, constraints, and acceptance criteria.
Advanced controls
Redaction
Use redaction before copying prompts that contain names, project codes, credentials, customer IDs, or other sensitive values. Redaction changes the generated prompt preview, not the source fields you typed. Redaction rules can contain the very values they hide, so they are kept only while this page is open and are never saved or exported.
P2 Paste Scrub
Turn on P2 scrub before using personal or sensitive source text. Prompertize masks common patterns locally before preview, copy, and vault save, including emails, greeting or self-introduction names, phone numbers, SSNs, card-like numbers, IP addresses, street-address-like text, DOB fields, and secret/token fields.
Exact Phrases
Best first choice. Add one word or phrase per line, such as a company name, unreleased product, person name, account number, or private project codename.
Key And Key=Value
Add key names such as password or api_key to redact values after separators like colon or equals. Add an exact pair such as customer_id=12345 when only one value should be masked.
Regex Patterns
Use only for advanced matching. Prompertize runs regex locally, limits pattern size, skips invalid patterns, and warns you when a pattern is skipped. To keep the page responsive, patterns that can backtrack badly are refused: repeated groups containing | or another repeat, backreferences, lookarounds, and more than one open-ended + or * (use bounded repeats such as {1,64}). Regex rules do not run on text longer than 20,000 characters; you will see a warning, and exact phrases and keys still apply.
Delimiter Format
Markdown headings are easiest to read. Hash sections are compact. XML tags are useful when you want very explicit boundaries around context, instructions, examples, or data. Ultra efficiency ignores this setting and uses plain "Label: value" lines.
Guardrails
Add rules for what the AI must not invent, what sources it can use, what topics are out of scope, and when it should say it does not have enough information.
Ask When Details Are Missing
Use this when wrong assumptions are costly. The generated prompt tells the AI to ask clarifying questions instead of guessing when required details are missing.
Step-By-Step Process
Use this for multi-part work. Provide the process you want followed, such as inspect source, extract facts, compare options, identify risks, then produce final recommendations.
Variable References
Use named intermediate outputs like [KEY_FACTS], [RISKS], [OPTIONS], or [FINAL_RECOMMENDATION] when you want the AI to organize work before the final answer.
Dataset Analysis Notes
Use this for spreadsheets, exports, logs, surveys, or tabular data. Include column names, definitions, filters, missing-value rules, and calculation limits.
Ask For Data Separately
Use this when the instruction and the dataset should stay separate. The generated prompt asks the AI to wait for the data before analyzing it.
Examples
Add good input-output examples, style examples, or counterexamples. Examples help the model match format and quality more reliably than abstract instructions alone.
Saving and reusing work
System Prompt Memory
Use for reusable behavior rules, such as evidence-only answering, JSON output discipline, brand voice, or review standards. Selecting a memory loads it automatically.
Follow-up
After the AI answers, open the Follow-up panel. The one-click follow-ups (shorter, longer, simpler, ask clarifying questions, change tone, check against original requirements) create a short prompt you paste into the same chat. To refine a reply, paste the AI's reply and say what should change: you get a structured revision prompt with the reply as delimited context, so it works in any chat. If a reply is pasted, the one-click follow-ups include it too. Your P2 scrub and redaction settings apply, and the sensitive-data check asks before you copy or open anything. Nothing in the panel is saved, and nothing is sent until you click Copy or an Open button.
Prompt Vault
Use for finished prompts you explicitly want to keep. A vault entry stores the generated prompt text (after any P2 scrub or redaction you had on), its name and tags, and the builder settings you used. It does not store the source text you typed or your redaction rules. Copy puts the saved prompt on your clipboard, and Use loads the saved settings while keeping what you are currently typing. Export or import the vault as JSON for backup.
Readiness And Safety
Readiness flags missing required fields. Safety flags possible sensitive data locally, such as secrets, emails, phone numbers, SSNs, credit cards, IP addresses, addresses, and confidential labels, before you paste into an AI tool.
Privacy and security
Prompertize stores UI settings, system prompt memory, and explicit vault items in this browser's local storage on your device. Working request text and redaction rules are not kept in the autosaved draft or in vault entries, and the request text clears after a successful copy. Use Clear all local data to remove everything this page has stored.
This builder page includes no analytics, telemetry, ads, tracking pixels, remote prompt processing, or network calls. Its Content-Security-Policy blocks them. Text goes to an AI provider only when you click an Open in button for that provider; the sensitive-data check asks you to confirm first if it finds anything. A reply you paste into the Follow-up panel is kept in the page only, is never stored, and is treated the same way.
P2 scrub and custom redaction run locally. They are helpful safeguards, not a guarantee; review the generated prompt before sending it to any AI service.
Because prompts can contain sensitive information, review the Safety panel before pasting text into any third-party AI service. Prompertize can help structure the prompt, but the destination AI site controls what happens after you send it.
