Document AI score
ai_percentage reports how much of the input is covered by AI-labeled sentences, so you can sort a queue, set a threshold or trigger a second review on one field.
AI Detector API
AI Detector API for ChatGPT, GPT-5, Claude and Gemini text
Add AI content detection to your own product. Send text to one REST endpoint and get back a document score with sentence-level evidence as JSON, built on the detection models behind detecting-ai.com.
Plans from $15 a month for 100,000 words. No SDK to install: any language that can send HTTPS works.
Illustration of the AI Detector API: a document goes in, and a JSON response with a 61 percent AI score and labeled sentences comes back
- 99%
- detection accuracy on diverse datasets
- <1%
- false positive rate on human writing
- 1B+
- human and AI texts tested
- $15
- a month for 100,000 words
What you get back
Everything the free AI checker shows, returned as structured data
The response is designed for products, not screenshots. One number for sorting and thresholds, plus the sentence-level detail that lets your interface show users why.
Sentence-level evidence
Every analyzed sentence comes back with character offsets, an AI or Human label and a confidence score from 0 to 1. Highlight the exact passages inside your own editor or report.
Coverage for current models
Designed to review text from ChatGPT and GPT-5, Claude, Gemini and mixed human-edited drafts, and retested as new models ship.
Calibrated, low false positives
A feature-based classifier and a fine-tuned transformer are calibrated into one score, keeping wrong flags on human writing under 1% in Detecting AI's published benchmarks.
Use cases
Where an AI detection API earns its place
Any product that accepts writing it did not watch being written. The API supplies the signal and the evidence; your workflow keeps the judgment.
Learning platforms and LMS
Score assignment submissions as they arrive and show the sentence evidence next to drafts, revision history and the institution's own policy.
Moodle plugin or direct REST integration
Content marketplaces and freelance platforms
Check delivered articles, product descriptions and translations before they reach the client, and route unusual cases to a human editor.
Flag deliveries above a threshold for review
Publishers and newsrooms
Add an authorship signal to contributor intake so editors read the flagged passages first instead of rereading every submission.
Highlight passages inside the CMS
SEO and content agencies
Run every deliverable through one QA step before it is published, and keep the processed word count for each client report.
Batch-check drafts from the server
Admissions and hiring
Review essays, cover letters and written assessments at scale, with the score opening a conversation rather than closing one.
Sort applications by review priority
Trust and safety teams
Triage reviews, comments and generated spam at volume, then send the uncertain middle to moderators together with the highlighted sentences.
Moderation queue with sentence context
How it works
Three steps from API key to highlighted result
The integration is one HTTPS request. The full reference, including authentication, errors and retries, lives in the API documentation.
- 1
Create a key and choose a plan
Prepare an API key in the dashboard and store it in a server-side secret. The subscription supplies the word balance; the key only identifies your account.
- 2
POST the text
Send UTF-8 text with the detector version to api.detecting-ai.com. Usage is counted from the input words, and an Idempotency-Key makes retries safe.
- 3
Render the evidence
Map ai_percentage to a verdict band and use the chunk offsets to highlight sentences in your own interface. Keep a person on consequential decisions.
Example request
curl --request POST \
--url https://api.detecting-ai.com/api/detect/ \
--header "Content-Type: application/json" \
--header "Idempotency-Key: detector-review-001" \
--header "X-API-Key: YOUR_API_KEY" \
--data '{
"text": "Paste the text you want to analyze.",
"version": "v3"
}'Example response, trimmed
{
"success": true,
"data": {
"details": {
"chunks": [
{
"text": "A sentence from the input.",
"startChar": 0,
"endChar": 26,
"type": "AI",
"score": 0.87
}
],
"ai_percentage": 42.5
},
"version": "v3",
"words_processed": 120
}
}Authentication, error codes, safe retries and the MCP server are documented on one page.
Read the full API referenceBuilt for production
Predictable usage, safe retries, explicit versions
The contract is small on purpose. The details below are the ones that matter once the integration runs inside a real review queue.
Safe retries
Send an Idempotency-Key with every request. An exact replay within 24 hours returns the stored result without running the detector or charging the words again.
Predictable usage
Usage is counted from whitespace-separated input words, up to 160,000 characters per request. Every response reports the words processed.
Explicit versions
Pin v3 in the request body so results stay reproducible as the detector improves. v1 and v2 remain available for existing integrations.
Keys stay on your server
Call the API from your backend, an edge function or a worker. Keys are shown once, can be replaced without interrupting the active key, and never need to reach a browser.
Detection engine
Sentences first, then the score
Detecting AI pairs a feature-based classifier with a fine-tuned transformer and calibrates the two into one score. The engine has been tested on more than a billion human-written and AI-generated samples, and it is retested as new models ship.
Read the methodology- 01
Sentence-level analysis
Eligible sentences are scored independently and labeled, so the document signal is explainable passage by passage.
- 02
Measured in both directions
How much AI text gets caught and how often human writing is wrongly flagged are both tracked, which keeps the false positive rate under 1%.
- 03
A signal, not a verdict
Short, translated, heavily edited or formulaic text carries more uncertainty. The evidence is returned so a person can check the call.
Three ways to integrate
REST, MCP or no-code
One subscription and one word balance, reachable from your backend, from an AI assistant or from an automation tool.
REST API
A JSON endpoint for any backend, edge function or batch job. Pin the detector version and store the processed word count.
Open the API docsMCP server
Give Claude, ChatGPT, Codex or any MCP client the detect_ai_text tool so an assistant can check text mid-conversation.
See the MCP serverZapier
Add a detection step to a no-code workflow and pass the result to a human review step when the decision matters.
View the Zapier appPricing
Simple monthly plans, usage counted in words
Start with the detector alone or combine all three APIs under one key. Checkout rechecks the live price and allowance before payment opens.
AI Detector API
$15per month
100,000 detector words
- AI Detector REST endpoint
- detect_ai_text MCP tool
- Detector v1, v2 and v3
- Feature-specific word balance
API Bundle
$35per month
100,000 words for each included API
- AI Detector REST + MCP
- AI Humanizer REST + MCP
- Plagiarism Checker REST
- One API key and dashboard
Compare API plans
An API key is free to create. Requests need an active subscription and remaining words.
API FAQ
Questions developers ask before integrating
Straight answers about accuracy, output, pricing and limits, so the first request behaves the way you expect.
What is an AI detector API?
An AI detector API is a web service that analyzes text for patterns associated with AI writing and returns the result as data instead of a web page. The Detecting AI endpoint accepts a JSON body with the text and a detector version, then responds with a document-level ai_percentage, sentence-level chunks with labels and confidence scores, and the number of words processed.
How accurate is the AI Detector API?
Detecting AI's detection models detect AI-generated text with 99% accuracy across diverse datasets and keep the false positive rate under 1%. Long, unedited samples are the easiest case. Short, heavily edited or translated passages carry more uncertainty, which is why every response includes the sentences behind the score.
Which AI models can the API detect?
It is designed to review text from the GPT family including ChatGPT and GPT-5, Claude, Gemini and mixed human-edited drafts. It does not attribute text to one specific generator with certainty.
How much does the AI Detector API cost?
The AI Detector API plan costs $15 per month and includes 100,000 detector words. The $35 API Bundle adds the AI Humanizer and Plagiarism Checker APIs with 100,000 words for each. Checkout verifies the live price and allowance before opening payment.
Is there a free AI detector API?
Creating an API key is free, but requests need an active subscription and remaining word balance. To evaluate detection quality before subscribing, use the free AI detector on detecting-ai.com, which shows the same kind of sentence-level evidence.
Does the API return sentence-level results?
Yes. Detector v3 returns a chunk for each analyzed sentence with startChar and endChar offsets, an AI or Human label and a confidence score from 0 to 1. Scores above 0.5 receive the AI label. ai_percentage is the percentage of input characters covered by AI-labeled sentences, which differs from the averaged document score shown by the website detector.
Can I call the API from browser JavaScript?
No. A private API key must never be shipped in browser code, a mobile binary or a public repository. Send requests from your server, an edge function or a worker and return only the result your interface needs.
Does a high AI score prove the text was AI-written?
No. AI detection is a probabilistic review signal, not proof of authorship or misconduct. Use the score and the highlighted sentences to focus a review, then consider drafts, sources, writing history and context before any consequential decision.
How fast is the AI Detector API?
Response time depends on text length. One measured sample on 2026-10-08, a 305-word English text on the public web detector including browser overhead, completed in 2.75 seconds. That is a single observation, not a guarantee, so measure with your own representative documents and keep the Idempotency-Key so a timed-out request can be retried without a second charge.
Can the AI Detector API run on premise?
No. The API is hosted at api.detecting-ai.com and there is no self-hosted package, container or model download. Your server sends text over HTTPS and receives JSON; nothing needs to be installed beyond an HTTP client.
Can I use the detector from an AI assistant or a no-code tool?
Yes. The same subscription powers the detect_ai_text tool on the remote MCP server for Claude, ChatGPT, Codex and other MCP clients, and the Detecting AI app on Zapier. MCP and Zapier usage draws from the same detector word balance as the REST API.
Start building
Make your first detection request today
Create a key, send one representative sample with Detector v3 and inspect the full JSON before designing the workflow around it.