AI Signal Check
Spot AI patterns in text — fairly and transparently, without selling false certainty.
SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.
AI Signal Check is an AI-writing-pattern check skill: it analyzes text for AI-typical language, structure and style patterns and explains, fairly, *why* it reads artificial, generic or too smooth. Not an "AI detector": no authorship verdict, no percentage. Instead an assessment category, signal strength 1–10, confidence, false-positive risk, flagged passages with an alternative explanation, ranked origin hypotheses and an editorial recommendation — for website, cover letter, LinkedIn, support and editorial use.
What this skill does
- Detects text type and channel and picks the fitting lens (email, LinkedIn, website, cover letter, support, SEO, social, Discord, academic, legal)
- Gives an assessment category (7 rungs from "human-feeling" to "likely AI-generated") — always a hypothesis, never a verdict
- Rates signal strength 1–10 as *pattern strength*, not AI probability, plus confidence and false-positive risk, separately
- Scans eight signal layers and weights them (strong/medium/weak) via an evidence ladder — no single word as "proof"
- Quotes specific passages with signal type, rationale, alternative explanation and editorial direction
- Names rhetorical AI constructions (negative parallelism "not just X — it's Y", rule-of-three, "stands as a testament", sycophancy) with the line and a fix — the patterns experts trust more than single words
- Ranks origin hypotheses (AI-generated, AI-assisted, AI-smoothed, template, translation, purely human) by plausibility
- Explains, for every finding, what argues against a hard AI verdict
- Treats short, formal, translated and non-native text with extra caution
- Calibrates both ways — a concrete, human text earns a low signal, no manufactured problems
- Ends useful: recommendation, next options and a handoff block for a natural rewrite
Description
AI Signal Check is a skill for the fair analysis of AI-typical writing patterns — it explains, transparently, why a text reads artificial, generic or too smooth, without accusing anyone of using AI. Its value is **orientation, not certainty**: not "AI yes/no" but "which signals point which way — and how sure is this read?" It is built cautiously on purpose, because AI detection is unreliable: OpenAI shut down its own classifier in 2023 for low accuracy, and detectors misflag roughly 6 in 10 texts by non-native writers as AI (Stanford 2023). So AI Signal Check gives **no authorship verdict and no percentage** like "87% AI" — percentages fake an objectivity they don't have. Instead it returns an assessment category (from "human-feeling" to "likely AI-generated", always a hypothesis), a signal strength 1–10 (not an AI probability but pattern strength), a confidence and a false-positive risk — reported separately. It works with signal weighting and an evidence ladder on the principle "one word is not the problem, the pattern is": a single buzzword or an em-dash is weak; only several co-occurring signal types over enough text carry weight. It scans eight layers (structure smoothness, language/phrasing, concreteness, tone, rhythm/variance, context-fit, template residue, lexical diversity), quotes specific passages, and gives every flag an **alternative explanation** (corporate template, translation, non-native author, SEO, legal/academic form, deliberately plain language). Channel lenses for email, LinkedIn, website, cover letter/HR, support, social, SEO and community tune what counts as a signal — and where the false-positive risk is high because the channel is naturally formulaic. Modes range from a quick check through a detailed report and editorial review to an academic caution mode (no accusation, check the writing process not just the text), brand voice, HR and a natural rewrite. Two principles set it apart: it **calibrates both ways** — a genuinely concrete, human text earns a low signal and a clean pass instead of being talked down; and it **never helps evade detectors** or optimize deception, making text more real (concrete, clear, authentic), not more sneaky. It is the diagnosis skill: check, understand, flag — and, on request, hand off to a rewrite step via a structured block.
Examples
What it does not do
- Gives no final authorship verdict ("this is AI" / "this is human") and no percentage
- Accuses no one and recommends no academic, legal, employment or disciplinary consequences from this check alone
- Doesn't over-read short, formal, translated or non-native text — flags the false-positive risk instead
- Won't help evade AI detectors, make text "undetectable", launder plagiarism or optimize deception
- Doesn't treat a single word or one list as proof — strength comes only from co-occurring patterns
- Doesn't produce a wall of text; leads with the assessment
Compatibility & tech
- Tested (internal)
- 6 scenarios
- Modes
- Quick signal check · Detailed signal report · Editorial review · Academic caution · Brand authenticity · Cover letter / HR · Natural rewrite (authenticity, not camouflage) · Compare (two texts: which reads more AI-typical?) · Section heatmap (which paragraphs read most AI-typical?)
- Inputs
- Text (any kind) · Channel / audience / context (optional) · Brand texts / sample texts (optional, for brand comparison) · Comparison text (original vs. revised version)
- Output format
- Structured signal report (assessment, signal strength, confidence, false-positive risk, short verdict, notable patterns, flagged passages, origin hypotheses, what argues against a verdict, recommendation, next options); on request a dimension table, an exportable report (AI_SIGNAL_REPORT.md) and a rewrite handoff block
- Subcategory
- AI text signals & authenticity review
- License
- Proprietär
Security profile
Runs entirely on your machine with your own AI — no external runtime, no running costs.
Only instructions, templates and references — no executable scripts.
Works offline with what you provide — does not call external services on its own.
What you get
- ai-signal-check-1.0.0/10 files
- .claude-plugin/marketplace.json
plugins/ai-signal-check/skills/ai-signal-check/10 files
- SKILL.md
- manifest.json
references/8 files
- ai-signal-check-channels.md
- ai-signal-check-checklist.md
- ai-signal-check-constructions.md
- ai-signal-check-evidence.md
- ai-signal-check-false-positives.md
- ai-signal-check-modes.md
- ai-signal-check-output-templates.md
- ai-signal-check-signals.md
- .agents/skills/ai-signal-check/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
- LICENSE.txt
Installation
Also works as a chat prompt
No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.
This skill ships no scripts, so the prompt carries its full method.
Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.
Unlock to copy the ready-to-paste prompt — then in “My Skills”.
Reviews
Note
Not an AI detector and not proof of AI use — an editorial signal-check. An assessment never replaces a conversation, a look at the writing process, or a professional review; never use it alone as grounds for academic, legal or employment consequences. Strongest with channel/audience as context; for very short texts it gives only rough hints.
Changelog
- v1.0.030.06.2026First release: a fair AI signal-check to the AUTHORING v2 standard — assessment category + signal strength 1–10 + confidence + false-positive risk (instead of fake percentages), flagged passages with an alternative explanation, ranked origin hypotheses, an evidence ladder + signal weighting, channel lenses (email/LinkedIn/website/cover letter/support/SEO/Discord/academic/legal), modes (quick/detailed/editorial/academic-caution/brand-voice/HR/natural-rewrite/compare/section-heatmap), detection of rhetorical AI constructions (negative parallelism, rule-of-three, sycophancy) and a handoff block. 8 references (modes & categories, signal catalogue, rhetorical constructions, false positives & safety, channels, evidence base with current sources, checklist, output templates), realistic examples incl. a calibration case and internal evals. Abuse-resistant: won't help evade detectors.
Frequently asked questions
What does AI Signal Check do?
AI Signal Check is an AI-writing-pattern check skill: it analyzes text for AI-typical language, structure and style patterns and explains, fairly, *why* it reads artificial, generic or too smooth. Not an "AI detector": no authorship verdict, no percentage. Instead an assessment category, signal strength 1–10, confidence, false-positive risk, flagged passages with an alternative explanation, ranked origin hypotheses and an editorial recommendation — for website, cover letter, LinkedIn, support and editorial use.
How do I tell if a text sounds like AI — without falsely accusing anyone?
AI Signal Check analyzes language, structure, style and context patterns and tells you why a text reads AI-typical, generic or too smooth — with an assessment category, signal strength 1–10, confidence and false-positive risk. It never gives an authorship verdict and never a percentage.
Is AI Signal Check an AI detector like Turnitin or GPTZero?
No. AI detectors are unreliable and frequently misflag non-native writers as AI. AI Signal Check sells orientation, not certainty — a transparent, editorial signal-check with alternative explanations instead of a yes/no claim.
Why doesn't AI Signal Check output a percentage like "87% AI"?
Because percentages fake a certainty that doesn't exist (even Turnitin hides low scores). AI Signal Check uses signal strength, confidence and false-positive risk in words — and always states what argues against a hard verdict.
Can I check cover letters, LinkedIn posts or website copy with AI Signal Check?
Yes. AI Signal Check has channel lenses for email, LinkedIn, website, cover letter/HR, support, social, SEO and community — and flags generic passages with concrete editorial direction, without devaluing the person.
Does AI Signal Check help me get past AI detectors?
No. AI Signal Check never helps make text "undetectable" or optimize deception. It helps make a text more concrete, clearer and more authentic — more real, not more sneaky.
What do I do with the result of AI Signal Check?
AI Signal Check ends useful: flagged passages, origin hypotheses, an editorial recommendation and next options — including a handoff block for a rewrite (more concrete and human), e.g. with Pain-to-Punchline or Newsletter-Schmiede.
Which AI tools does AI Signal Check work with?
Claude (Projects & system prompt) · ChatGPT (Custom GPT & custom instructions) · Any LLM · Native UI (buttons/chips) where available, Markdown fallback otherwise · Optional: web tool for current detection research
How do I use AI Signal Check?
AI Signal Check is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.