# AI Content Detector

AI detector for English text: was it written by ChatGPT? A 0 to 1 score, three bands, and a flag with a stated false-positive rate.

Send English text of a few sentences up to 100,000 characters. You get a score from 0 to 1, a band (human, AI-edited or AI-generated) and a yes-or-no flag set so that about 1 in 100 human texts is flagged. It is an estimate, not proof: it misses most paraphrased or 'humanized' text and text from older models, and it can be wrong about a person's own writing, so never use it alone to accuse someone. English only. The model runs on our own hardware in the EU and your text is not stored.

- The flag is set so that about 1 in 100 human texts is flagged, and we say so. Most detectors only say "high accuracy".
- Three bands, not a yes or no: human, AI-edited, AI-generated, with the score behind them.
- Honest about what it cannot do: paraphrased and humanized text mostly gets past it, and we say so before you build on it.

## Calls

### `POST https://readymadeapi.com/v1/text/ai-score`

Score one English text: 0 for untouched human writing, around 0.3 for a human draft a chatbot polished, 1 for text a model wrote from scratch.

- Price: $0.00024 per text
- MCP tool: `ai_text_score_score`
- Answers in the same call.

Request:

```json
{
  "text": "In today's fast-paced digital landscape, leveraging innovative solutions is essential for organizations seeking to unlock their full potential. By embracing a holistic approach, businesses can foster meaningful growth and navigate the complexities of an ever-evolving market with confidence."
}
```

Answer:

```json
{
  "score": 0.9842,
  "band": "ai_generated",
  "flagged": true,
  "windows": 1,
  "tokens": 48,
  "sampled": false,
  "notice": "An estimate, not proof. It misses most paraphrased text and text from older models, and can be wrong about a person's own writing. English only."
}
```

## Limits

- English text, 200 to 100,000 characters per call.
- Beyond about 16,000 characters the score comes from eight samples spread along the text, and the answer says "sampled": true.

## Questions

**How long can the text be?**

From 200 to 100,000 characters (about 15,000 words) in one call. Up to about 16,000 characters every word is read; longer texts are scored from eight samples spread along their length and the answer says so. One text is one unit at any length.

**Can I use the result to prove someone used ChatGPT?**

No. It is an estimate. It misses most paraphrased or humanized text and text from older models, and it can be wrong about a person's own writing. Use it to decide what to look at, never alone to accuse someone.

**Which languages?**

English only. The language is not checked: text in another language still gets a score, and that score means nothing. Check the language before you call.

**How is a call counted?**

One text is one unit, whatever its length.

## Using it

- Send `Authorization: Bearer <key>` and a JSON body. Keys and credit: https://readymadeapi.com/account
- Every answer carries `X-Request-Id`, `X-Units` and `X-Charged-Usd`.
- An error is a JSON problem with a `title` to act on and a `detail` in plain words; nothing is charged for it.
- Full schema: https://readymadeapi.com/openapi.json
