I built PrecisionAudit because I got tired of the default "turd polishing" reflex of Large Language Models.
Whenever you ask ChatGPT to improve landing page copy or a cold email, its default behavior is to inject words like "revolutionize", "seamlessly", or worse—invent fake round numbers like "boosts ROI by 300%". It generates "fake confidence."
PrecisionAudit is a strict, dual-agent system designed to do the exact opposite: strip away the BS, penalize unverified claims, and force B2B copy to be mechanism-driven.
How it works under the hood:
Agent 1 (The Brutal Evaluator): It scores your text based on strict mode-aware rubrics (SaaS Landing Page, Cold Email, Pitch Deck, etc.). We implemented a "Hard Ceiling" rule: if your input text lacks verifiable real-world metrics or case studies, your "Specificity" score is hard-capped at 7/10. It actively flags generic percentages as unsupported_claims.
Agent 2 (The Rewriter): It has strict system prompts forbidding it from strengthening fake claims. If it sees "increases speed by 50%" without proof, it neutralizes it into a defensible outcome like "removes manual data entry friction". It also strips out all Markdown formatting to give you raw, copy-ready text.
Mode-Awareness: It knows that extreme brevity and a soft CTA in a Cold Email is a feature (10/10), but the same brevity in a Pitch Deck is a flaw.
Tech Stack & Anti-Abuse:
Built with Next.js, Firebase Admin SDK, and OpenAI (gpt-4o-mini with low temp and fixed seeds for determinism). We handle credit deductions using db.runTransaction to prevent race conditions from concurrent requests. There's also a hidden "sandwich watermark" injected into the rewritten text to prevent users from accidentally burning credits by re-auditing already optimized copy.
You get 5 free credits via Google Auth to play around with it.
I’d love for you to throw your worst, most hyped-up, fluff-filled marketing copy at it. See if you can bypass the guardrails or if the system successfully neutralizes the BS.
Hi HN,
I built PrecisionAudit because I got tired of the default "turd polishing" reflex of Large Language Models.
Whenever you ask ChatGPT to improve landing page copy or a cold email, its default behavior is to inject words like "revolutionize", "seamlessly", or worse—invent fake round numbers like "boosts ROI by 300%". It generates "fake confidence."
PrecisionAudit is a strict, dual-agent system designed to do the exact opposite: strip away the BS, penalize unverified claims, and force B2B copy to be mechanism-driven.
How it works under the hood:
Agent 1 (The Brutal Evaluator): It scores your text based on strict mode-aware rubrics (SaaS Landing Page, Cold Email, Pitch Deck, etc.). We implemented a "Hard Ceiling" rule: if your input text lacks verifiable real-world metrics or case studies, your "Specificity" score is hard-capped at 7/10. It actively flags generic percentages as unsupported_claims.
Agent 2 (The Rewriter): It has strict system prompts forbidding it from strengthening fake claims. If it sees "increases speed by 50%" without proof, it neutralizes it into a defensible outcome like "removes manual data entry friction". It also strips out all Markdown formatting to give you raw, copy-ready text.
Mode-Awareness: It knows that extreme brevity and a soft CTA in a Cold Email is a feature (10/10), but the same brevity in a Pitch Deck is a flaw.
Tech Stack & Anti-Abuse: Built with Next.js, Firebase Admin SDK, and OpenAI (gpt-4o-mini with low temp and fixed seeds for determinism). We handle credit deductions using db.runTransaction to prevent race conditions from concurrent requests. There's also a hidden "sandwich watermark" injected into the rewritten text to prevent users from accidentally burning credits by re-auditing already optimized copy.
You get 5 free credits via Google Auth to play around with it.
I’d love for you to throw your worst, most hyped-up, fluff-filled marketing copy at it. See if you can bypass the guardrails or if the system successfully neutralizes the BS.
Brutal feedback is highly appreciated!
Link: https://www.precisionaudit.app