Active development is paused. The audit tool, credit purchases, and existing balances stay operational — no plans to take it down. Credits remain valid until used.
How It Works
Not just another AI detector: a pattern-level rewriter.
How detection works
Most AI detectors guess. They run text through a classifier and spit out a percentage. Wrong half the time. Useless for real editing.
avoid-ai-writing works differently. It identifies 43 categories of structural patterns that AI writing produces (like em dash overuse, ironically), hollow intensifiers, formulaic transitions, uniform paragraph lengths, hedging language, and more. It maps these against a 109-word replacement table to flag exactly what sounds like AI and why.
Two-pass rewrite
Your text
↓
Pattern scan43 categories, 109-word table
↓
RewriteFlag AI patterns, replace with natural alternatives
↓
Second passRe-scan rewritten text, catch survivors
↓
Clean text + diff summary
The output isn't a score. It's clean text that reads like a human wrote it, with a detailed breakdown of every change and why it was made.
The burn mechanic
Audits paid in $avoid permanently burn the tokens on Solana. Not locked. Not staked. Burned. Removed from the total supply forever.
The burn amount adjusts dynamically to target ~$0.25 USD per audit. Price feeds from DexScreener and Jupiter update every 60 seconds. If token price goes up, you burn fewer tokens. If it drops, you burn more. The dollar cost stays roughly the same either way.
The web app
This site is a Next.js app on Vercel that ties the detection engine to Solana's token burn mechanic. The detection engine is MIT-licensed and open source.
You
→
Next.js frontendPaste text, connect wallet, fetch live burn price
↓
SolanaSPL burn instruction, on-chain verification
→
Vercel API routeVerify burn amount + tx, reject replays
↓
Claude APISKILL.md as system prompt, two-pass audit
→
Audit resultFlags, rewrite, diff summary
Tokens are burned using Solana's native SPL Token burn instruction, not transferred to a dead wallet. The burn amount adjusts to live token price, targeting ~$0.25 per audit. The API independently verifies the burn (correct mint, minimum amount, not replayed) before running the audit. Text capped at 5,000 words. Rate limited to 10 audits/hr per IP.
Open source
The detection engine is open source. Fork it, run it locally, or plug it into your own tools. No API keys, no rate limits, no lock-in.