Structured LLM prompt builder — type a topic, pick persona / task / format / tone / constraints, 🎲 per field, or one-click 15 presets. Outputs You are X. TASK. TOPIC. ... multi-section structure ready for ChatGPT, Claude, and Gemini.
PRESETS · Quick presets
Topic / task contentWhat you want the model to do — plain English or whatever language fits. e.g. 'Review the function below for race conditions.'
PersonaA concrete persona (not 'expert' or 'assistant') unlocks tacit domain knowledge a generic prompt won't.
Task typeVerb-led task type — gets dropped into the TASK section.
Output format
Length
Tone00Pick 1–3.
Output language
Constraints00Pick 1–3 — the more specific the constraint, the more it bites.
Recommended modelClaude Opus 4.7 = strongest, Sonnet 4.6 = balanced, Haiku 4.5 = fast/cheap. Same logic for GPT / Gemini families.
HOW TO USE
Type a topic — what you want the model to do (English or whatever language fits)
Pick a concrete persona (e.g., 'senior ML engineer') — abstract roles like 'expert' produce blander output
Pick a verb-led task type (review / analyse / brainstorm)
Pick output format + tone + constraints, or hit '🎲 Randomize all' for a baseline
Pick recommended model — just a reminder of which platform / model variant to use
Keep the prompt ≤ 800 chars for tightest results
Copy and paste into ChatGPT, Claude.ai, or Gemini; switch the platform to the model variant you noted
FAQ
Q01Is this generator free? Do I need an account?+
Completely free, runs in your browser (no API calls, nothing stored). No signup — open and use.
Q02Does the output work in ChatGPT, Claude, and Gemini?+
All three. We compose a structured 'You are X. TASK. TOPIC. OUTPUT FORMAT...' prompt — every modern LLM accepts that shape. The only difference is which UI you paste it into, and which model variant (pro / sonnet / opus / haiku, etc.) you select.
Q03How should I pick a Persona?+
Pick a concrete one — not 'expert' or 'assistant'. e.g., 'senior full-stack engineer with 15 years' brings out senior-level judgment instincts; 'assistant' gives you baseline answers. We've curated 35 personas, all concrete.
Q04What's the difference between Task and Topic?+
Task is the verb (review / analyse / translate); Topic is the noun. Task 'Code review' + Topic 'the function below' is a clear instruction. Topic alone, no Task → the model picks a task itself, often not the one you wanted.
Q05There are a lot of constraints — which ones should I check?+
1–3 max. Most useful: 'Don't pad', 'Lead with the main point', 'Avoid buzzwords'. For code: 'Wrap identifiers in `backticks`', 'If unknown, say so'. For research: 'Cite sources where relevant'.
Q06Does 'Recommended model' get baked into the prompt?+
No. The prompt text itself is identical regardless. The model field just reminds you which platform to paste into and which model to switch to. We can't toggle the model for you across services.
Q07Why is ≤ 800 chars recommended? I see 3000-char prompts all the time.+
Technically there's no limit (GPT-4 / Claude take 200k+ tokens). But the longer the prompt, the more diluted the constraints — tone gets lost in the noise. Empirically 800 is the sweet spot. Past that, consider splitting into multi-turn: turn 1 = role + task, turn 2 = drop the source material, turn 3 = ask for the artefact.