AI detection + humanizer workflow

Search engines and platforms are getting increasingly sensitive to AI-typical patterns: uniform sentence lengths, "It's important to note", over-structured lists. Rankion gives you two tools — an AI detector that scans text for AI probability, and a humanizer that actively makes it more natural.

Single-shot detection

Quick check for a single piece of text (e.g. a freshly written paragraph or an article delivered by a client):

curl -X POST https://rankion.ai/api/v1/ai-scanner/detect \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"text": "It is important to note that artificial intelligence..."}'

Cost: 2 credits. The response gives you an ai_probability (0–100) and a sentence-level heatmap showing which sentences look particularly AI-typical.

From ai_probability ≥ 70 you should humanize. Below 40 is usually fine.

Article humanizer

If you want to push a complete article (e.g. a freshly generated one from the AI Content Editor) through the humanizer:

curl -X POST https://rankion.ai/api/v1/ai-scanner/humanize \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Your full article text here...",
    "tone": "natural",
    "language": "en"
  }'

Cost: 5 credits. Synchronous, responds directly with the rewritten text. Duration: 5–15 seconds.

The humanizer varies sentence lengths, builds in imperfections (short sentences, question marks, inversions), swaps typical LLM filler phrases for everyday wording — and stays semantically faithful.

Bulk humanizer (async + batch polling)

For larger volumes (10+ articles, full content pipelines) there's the asynchronous bulk humanizer:

# Start a batch:
curl -X POST https://rankion.ai/api/v1/humanize \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "texts": [
      {"id": "art-1", "content": "..."},
      {"id": "art-2", "content": "..."},
      {"id": "art-3", "content": "..."}
    ],
    "tone": "natural"
  }'

Cost: 8 credits per text. Response: HTTP 202 with a batch_id.

Then you poll:

curl https://rankion.ai/api/v1/humanize/{batch_id} \
  -H "Authorization: Bearer YOUR_TOKEN"

As long as status: "processing", it's still running. On status: "completed", results[] is populated with the humanized versions per input ID. Typical runtime: 30 seconds to 5 minutes depending on batch size.

Unsupported language

AI detection and the humanizer understand German, English, Spanish, and French. Enter text in another language, Polish for example, and the scanner in the UI doesn't even start — it shows you a hint instead, and no credits are spent. Over the API, POST /api/v1/ai-scanner/detect without a language parameter responds with 422 and names the detected language (detected_language) plus the supported codes. Send such a text to the humanizer and it aborts the document with status "failed": the text stays unchanged, and again no credits are spent.

When to use — when not

Use it when:

• The article comes straight from the AI Content Editor and needs to be publish-ready on money pages
• Texts are submitted to platforms with AI detection (universities, some marketplaces)
• You want to build a content series with a consistently natural voice

Don't use it when:

• The text has already been manually edited — the humanizer can introduce style breaks
• The content is heavily technical / factual (code, numbers, definitions) — the humanizer can loosen things that shouldn't be loosened
• The tone should be "expert / formal" — the humanizer default leans towards colloquial

Quality gate before publish

Recommended workflow for a final article:

1. Generate (Your first article with the AI Content Editor)
2. Score & optimize (Optimize content for Google + AI)
3. AI detection — check ai_probability
4. If ≥ 70 → humanize
5. Re-detection → must be < 40
6. Manual final edit (5 minutes is usually enough)
7. Publish

Module details and limits in AI Detector and Humanizer. API reference: AI Tools API (Detector + Humanizer).