ai-text-check

One-point checker for AI writing tells, grammar, and synthetic typography in any JavaScript or TypeScript project.

It runs the mepritam.dev content-quality stack together with Vale, slop-gate, write-good, retext, is-this-ai-slop, Veldica, and clean-writing-lint, then returns a single scored report. See SOURCES.md for the file-by-file map.

Install

npm install -D ai-text-check

Node.js 20+ is required. The package ships dual ESM and CommonJS builds with TypeScript types. Vale and slop-gate are regular dependencies, not optional peers.

What it checks

| Engine | What it catches | | ------------------- | ------------------------------------------------------------------------------------------------------------------ | | polish | Custom AI tells (delve, emojis, in today's world, antithesis), readability, markdown structure/SEO, typography | | slop-gate | Vocabulary pack (nestled, treasure trove, leverage, …) | | write-good | Weasel words, clichés, wordy phrases, passive voice | | retext | Passive voice, readability grade, simplify, repeated words | | is-this-ai-slop | Clichés, buzzwords, em-dash spam, antithesis, sycophancy | | veldica | Stock LLM transitions and AI-style markers | | clean-writing | Banned constructions, marketing adjectives, filler, intensifiers | | ai-slop-linter | Wikipedia Signs of AI writing rules with line/column findings | | slop-detector | EQBench SLOP score (vocabulary, contrast patterns, trigrams, tropes) | | prose-slop | Vocabulary, punctuation, and document-cadence tells (English) | | vale | Vendored styles: write-good, Google, AiTextCheck, ai-tells, signs-of-ai-writing |

Safe auto-fixes can strip filler openers and collapse extra whitespace without rewriting code fences.

Emoji in prose is treated as an AI tell (ai-emojis). Set ignoreEmojis: true or pass --ignore-emojis to allow them; the ignore flag is off by default.

TypeScript

import { checkText, defineConfig, personalizeText } from "ai-text-check";

const config = defineConfig({
  preset: "default",
  failOn: "warn",
  ignoreRules: ["leverage-utilize"],
  engines: { vale: true },
});

const report = checkText(draft, {
  format: "markdown",
  applyFixes: true,
  config,
});

if (report.score < 70) {
  console.error(report.issues);
}

JavaScript (CJS or ESM)

const { checkText, sanitizeText } = require("ai-text-check");

const report = checkText("In today's fast-paced world, we leverage synergy.");
console.log(report.score, report.issues, report.enginesRun);

const cleaned = sanitizeText("Wait—what? “Hello” …");

import { checkText } from "ai-text-check";

Node file API

import { checkFile, checkFiles, loadConfig } from "ai-text-check/node";

const config = await loadConfig();
const summary = checkFiles(["content", "README.md"], {
  config,
  writeFixes: true,
});

if (summary.failed) process.exit(1);

loadConfig() reads, in order:

1. ai-text-check.config.json / .js / .mjs / .cjs
2. package.json → "ai-text-check"

Unknown keys and invalid types fail with a validation error. Vale findings are merged into the same issues array (use --no-vale or engines.vale: false to skip).

CLI

npx ai-text-check
npx ai-text-check content README.md
npx ai-text-check --fix --min-score 70 docs
npx ai-text-check --preset ai --json
npx ai-text-check --sanitize
npx ai-text-check --no-vale
npx ai-text-check init

| Flag | Meaning | | --------------------------------------- | ---------------------------------- | | --fix | Apply safe automatic fixes | | --sanitize | Normalize AI typography on disk | | --json | Machine-readable report | | --fail-on error\|warn\|info | CI threshold (default warn) | | --min-score N | Fail if any file scores below N | | --preset default\|strict\|ai\|grammar | Rule pack | | --format auto\|plain\|markdown | Input format | | --no-vale | Skip Vale | | --no-slop | Skip slop-gate and is-this-ai-slop | | --no-write-good | Skip write-good | | --no-retext | Skip retext | | --ignore-emojis | Allow emojis (skip ai-emojis) | | --ignore-rule ID | Skip a rule (repeatable) |

Exit code 1 when issues meet the fail threshold.

Config

ai-text-check.config.json:

{
  "preset": "default",
  "failOn": "warn",
  "minScore": 0,
  "extensions": [".md", ".mdx", ".markdown", ".txt"],
  "ignoreRules": [],
  "ignoreEmojis": false,
  "checks": {
    "ai": true,
    "grammar": true,
    "readability": true,
    "structure": true,
    "seo": true,
    "typography": true
  },
  "engines": {
    "polish": true,
    "slopGate": true,
    "writeGood": true,
    "retext": true,
    "isThisAiSlop": true,
    "veldica": true,
    "cleanWriting": true,
    "aiSlopLinter": true,
    "slopDetector": true,
    "proseSlop": true,
    "vale": true
  }
}

Presets:

• default: every check and every engine
• strict: default plus extra slop-gate punctuation
• ai: AI tells + typography (polish, slop-gate, is-this-ai-slop, Veldica, clean-writing, ai-slop-linter, slop-detector, prose-slop, Vale)
• grammar: grammar + readability (write-good, retext, clean-writing, Vale)

Scripts

npm run test
npm run lint
npm run typecheck
npm run format:check
npm run check
npm run build

License

MIT © 2026 Pritam Ghosh.

The checker itself, including the custom AiTextCheck Vale rules, is MIT. Vendored Vale styles keep their upstream licenses; see NOTICE.

vale/styles/signs-of-ai-writing is CC BY-SA 4.0.