For developers

AI detection API: one POST, one GET, a score you can act on

Submit text, get a request_id, poll the result with long-polling. You get a 0 to 100 document score, a verdict label, a probability and the highest-scoring sentences. One API key is $100 and includes 3,000,000 credits, where one credit covers one word.

Get an API key Read the reference
Keys are created, revealed once and regenerated from the Enterprise tab of the dashboard.

Key takeaways

  • $100 per API key, 3,000,000 credits included. One credit equals one word analysed.
  • Two calls, not one. POST /api/v1/enterprise/detect then GET .../{request_id}?wait=true.
  • Auth is a single X-API-Key header. No OAuth, no token exchange.
  • 50 to 10,000 characters per request. Narrower than the 50,000 the web app accepts, so chunk long documents.
  • Credits deduct after the job completes. A key can go negative rather than blocking, and overage is invoiced Net-30.
$100Per API key
3,000,000Credits included
1Credit per word analysed
Net-30Invoicing on overage
Reference

The whole surface in two endpoints

Base URL: https://detector-api-server-tsxmk.ondigitalocean.app

POST /api/v1/enterprise/detect

Submit text for analysis. Returns immediately with a job handle and the credit usage for the submission. Body is { "text": "string" }, between 50 and 10,000 characters.

GET /api/v1/enterprise/detect/{request_id}

Fetch the result. Append ?wait=true and the server holds the connection open until the job completes, up to roughly 30 seconds, which removes the need for a polling loop in most integrations. 

# 1. Submit text for analysis
curl -X POST "https://detector-api-server-tsxmk.ondigitalocean.app/api/v1/enterprise/detect" \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{"text": "Paste between 50 and 10,000 characters of the text you want analysed."}'

# → { "request_id": "ent_a3f9b2c1d4e5", "status": "queued", "usage": { ... } }

# 2. Fetch the result. ?wait=true long-polls until the job completes.
curl "https://detector-api-server-tsxmk.ondigitalocean.app/api/v1/enterprise/detect/ent_a3f9b2c1d4e5?wait=true" \
  -H "X-API-Key: YOUR_API_KEY"

The result payload

{
  "status": "completed",
  "job_id": "ent_a3f9b2c1d4e5",
  "document_score": 82,
  "document_prediction": "AI",
  "document_probability": 0.82,
  "total_sentences": 14,
  "processing_time": 340,
  "sample_sentences": [
    { "sentence": "Furthermore, the results demonstrate...", "score": 91 },
    { "sentence": "This analysis shows...",                  "score": 78 }
  ]
}
FieldTypeMeaning
document_scorenumber 0 to 100Raw AI likelihood score for the document.
document_prediction"AI" | "Human" | "AI + Human"Verdict label derived from the score bands below.
document_probabilitynumber 0 to 1Model confidence in the verdict.
total_sentencesintegerHow many sentences were analysed.
processing_timeinteger, millisecondsWorker processing time for the job.
sample_sentencesarrayTop-scoring sentences with their individual scores.

Score bands: 0 to 50 reads as Human, 51 to 60 as Mixed, 61 to 100 as AI. Apply your own threshold if your workflow needs a different sensitivity, but record which one you used so results stay comparable over time.

Errors

What comes back when something goes wrong

StatusMeaningWhat to do
401Invalid or missing API keyCheck the X-API-Key header. A regenerated key invalidates the previous secret at once.
402Insufficient creditsTop up from the API Keys tab in the dashboard.
422Text too short or missingThe body needs at least 50 characters of text.
404Job not found or expiredRe-submit. Do not assume a job handle is durable indefinitely.
429Rate limit exceededBack off and retry after a few seconds.
5xxServer errorRetry with exponential back-off.

We do not publish a numeric rate limit or an uptime commitment, so do not build against an assumed one. If your integration needs either in writing, ask at support@gptone.me before you commit to a design.

Integration notes

Four things worth getting right up front

Chunk at the API limit, not the UI limit

The endpoint accepts 50 to 10,000 characters. The web app accepts 50,000. If you sized your chunker off the marketing number your first long document will fail. Split on paragraph boundaries and aggregate.

Most common mistake

Store the sentence scores, not just the verdict

sample_sentences is what makes a result reviewable by a human later. A stored verdict with no sentence data is not auditable, and moderation decisions eventually get audited.

Auditability

Record the threshold you applied

If you deviate from the 61-and-above default, write the threshold into the record. Otherwise a policy change silently makes last quarter's data incomparable with this quarter's.

Data hygiene

Never auto-reject on the score alone

Our own benchmark puts the false positive rate at 3.6%, and 8.0% on non-native English writing. Route a flag to a person. Automatic rejection at scale means automatically rejecting real writers at a measurable rate.

Design constraint

Keep your key server-side. It is a bearer credential in a header, so anything that can read it can spend your credits. If one leaks, regenerate it from the dashboard: the old secret stops working immediately and your balance carries over.

Common questions

Questions developers ask about the detection API

One API key costs $100 and includes 3,000,000 credits. One credit covers one word of analysed text, so the included balance is 3,000,000 words. Credits are deducted after a job completes, not on submission. A key is allowed to go to a negative balance rather than hard-blocking, and the overage is invoiced at the end of the billing cycle on Net-30 terms. There is no long-term contract.
Send your key in an X-API-Key header on every request. There is no OAuth flow and no bearer token. Keys are created, revealed once, revoked and regenerated from the Enterprise tab of the dashboard. A regenerated key invalidates the old secret immediately and leaves the credit balance untouched.
No. Submission and retrieval are two calls. POST /api/v1/enterprise/detect returns a request_id straight away, and GET /api/v1/enterprise/detect/{request_id} returns the result. Add ?wait=true to the GET and the server long-polls, holding the connection open until the job finishes, so you do not need a tight retry loop.
The API request body takes between 50 and 10,000 characters of text. That is narrower than the web app, which accepts up to 50,000 characters per scan. Chunk longer documents on your side and aggregate the scores, or send the sections you care about.
An enterprise key is provisioned for the detector and the humanizer. The endpoint documented on this page is the text detector. If image detection through the API matters for your integration, ask before you buy rather than assuming, at support@gptone.me.

Before you wire it in

Three pieces on choosing and placing a detection API inside an existing product.

Integration
AI detection API: how to add content checks to your app
Where the call belongs in a submission flow, and what to do with the result.
Evaluation
How to choose and integrate an AI detection API
The questions worth asking a vendor before you pick one, including the ones about false positives.
Workflow
Adding AI content screening to your CMS workflow
Putting a check into an editorial pipeline without turning it into a bottleneck.

Get a key and make your first call

One key is $100 with 3,000,000 credits included, one credit per word analysed. Keys are issued and regenerated from the Enterprise tab in the dashboard.

See API pricing

Integration questions: support@gptone.me