AI Prompt Refiner

How It Works

Under the hood, the refiner runs on the same large language model stack as the rest of MintAI Tools — Llama 3.3 70B on Groq by default, with fallback to Gemini 1.5 Flash and OpenRouter's DeepSeek endpoint. What makes it special is the system prompt.

The system prompt instructs the model to act as a prompt engineer and to produce output in a strict three-part structure: the refined prompt itself, an explanation of *why* it works, and an estimated improvement percentage. The model analyzes your input prompt, identifies what's missing (no role, no context, no format spec, no constraints), and fills those gaps.

Notably, the model does not add information you didn't ask for. If your prompt is about dogs, the refined prompt stays about dogs — it just gets more specific about angle, audience, length, and structure. The output is a prompt you can copy-paste directly into any AI tool and expect dramatically better results.

Who Is This Tool For?

• Bloggers & content marketers — Turn 'write a blog post about X' into a detailed prompt that produces outline-ready, on-brand copy the first time.
• Developers using AI coding assistants — Refine vague 'write a function that does X' prompts into specs the model can actually implement cleanly.
• Students & researchers — Get structured, sourced, thoughtfully-organized explanations instead of generic summaries.
• Small business owners — Generate product descriptions, emails, and marketing copy that match a specific tone and audience.
• Anyone learning prompt engineering — Study the before/after examples to internalize what makes a prompt work.
• Teams standardizing AI use — Save refined prompts as templates so everyone on the team gets consistent output quality.

How to Use the AI Prompt Refiner

1. Type the raw version of what you want into the input box. Don't worry about being polished — 'write a cold email to a CTO about our SaaS product' is fine.
2. Optionally add the target AI tool and any context you have: 'for ChatGPT,' 'keep it under 200 words,' 'formal tone,' 'for a LinkedIn audience.'
3. Click ✨ Generate. In 1–2 seconds you'll get the refined prompt, an explanation of the improvements, and an estimated quality lift.
4. Read the 'Why This Works' section — over time you'll start writing better first-draft prompts because you recognize the patterns.
5. Copy the refined prompt into ChatGPT, Claude, Gemini, or wherever you work. Paste it exactly as-is; it's already structured for that model.
6. If the output still isn't quite right, come back and refine again with an added constraint like 'make it shorter' or 'add examples.'

Real Examples

write a blog post about dogs

You are a content writer for a pet-lifestyle blog aimed at first-time dog owners in the US.

  Write a 1,200-word blog post titled "10 Things I Wish I Knew Before Getting My First Dog." Structure it as:
  - A 2-sentence hook addressing a common fear new owners have
  - 10 numbered sections, each 100–150 words, with a bolded sub-heading
  - Each section includes one specific, actionable tip and a brief personal-sounding anecdote
  - A closing paragraph with a supportive, encouraging tone and a single call-to-action (subscribe to the newsletter)

  Tone: warm, honest, conversational — like advice from a friend who's been through it. Avoid: salesy language, exaggerated claims, "in conclusion" phrasing, and any mention of specific brands.

  Format: markdown with H2 for the title and H3 for each numbered point.

  Before returning, verify the article reads naturally out loud and every tip is genuinely useful for a beginner.

write a function that sorts a list

You are a senior Python developer writing production-quality code for a code review by a staff engineer.

  Write a Python function that sorts a list of dictionaries by a specified key, with the following requirements:
  - Function signature: sort_records(records: list[dict], key: str, descending: bool = False) -> list[dict]
  - Handle missing keys gracefully (records without the key go to the end, order preserved)
  - Handle mixed types within the same key gracefully (fall back to string comparison)
  - Include type hints throughout
  - Include a Google-style docstring with Args, Returns, Raises, and two usage examples
  - Include 3 unit tests using pytest

  Constraints: no external dependencies beyond the standard library, O(n log n) time complexity.

  Before returning, mentally run the code against these inputs: [], [{'a': 1}], [{'a': 2}, {}], [{'a': 1}, {'a': 'x'}].

cold email to a potential client

You are a B2B sales copywriter who specializes in short, high-reply-rate cold emails for SaaS founders.

  Write a cold outreach email to a marketing director at a mid-market e-commerce company (500–2,000 employees). The goal is to book a 15-minute call — not to sell in the email.

  Requirements:
  - Subject line under 7 words, non-clickbaity
  - Opening line references something specific and plausible about their role (not flattery)
  - One-sentence value proposition framed as a question or observation
  - One concrete, verifiable proof point (use a placeholder like [metric])
  - Single clear CTA: ask for a 15-minute call, offer two specific time windows
  - Total length: 90–130 words
  - Sign-off: "— [Your Name]"

  Tone: peer-to-peer, direct, respectful of their time. Avoid: "I hope this finds you well," "quick question," excessive punctuation, and any phrase that sounds like a template.

  Before returning, check that the email could plausibly get a reply from a busy executive.

Technical Details

Primary engine: Llama 3.3 70B via Groq (response time typically 0.9–1.8 seconds). Fallback: Google Gemini 1.5 Flash, then OpenRouter (DeepSeek). Temperature is set to 0.7 to encourage creative restructuring. Max output: 2,048 tokens — enough for even complex prompts with multiple constraints.

The system prompt is deliberately format-strict: it forces the model to output the three labeled sections (refined prompt, why it works, expected improvement) so the result is immediately usable. Requests are rate-limited per IP (10/min default) and inputs are not stored.

Frequently Asked Questions

Why Use MintAI Tools for This?

Most prompt-improvement tools either charge a subscription, require a signup, or produce output that's just a longer version of your original prompt without the structural improvements that actually matter. This one is genuinely free, keeps everything in your browser until you copy it out, and produces prompts that consistently work well in ChatGPT, Claude, and Gemini — because the system prompt encodes the same principles used by professional prompt engineers.