Remove AI watermark API: rewriting and records

Looking for a remove-AI-watermark API? Overwride provides a rewrite endpoint with a shareable draft record. Its response does not certify watermark removal. Here is what you can build with it.

Define what “remove” needs to mean

A text rewrite, a watermark detector, and a record lookup are separate operations. A developer should know which result an endpoint actually returns before putting a “watermark removed” message in an interface.

Statistical watermarks can become harder to detect after extensive rewriting, but that is not a universal removal guarantee. Google describes this limitation for SynthID. Google DeepMind: SynthID text watermarking. Overwride returns rewritten text and its record; it does not return a provider-watermark detection result.

Send a draft to the rewrite endpoint

Call POST /api/v1/rewrite from your server with an Overwride key. The draft field is required and accepts up to 1,000 words. An optional title is visible only to the account owner.

POST https://overwride.com/api/v1/rewrite
Authorization: Bearer $OVERWRIDE_API_KEY
Content-Type: application/json

{
  "draft": "We launched dark mode today. Mia fixed the chart colours.",
  "title": "Dark mode announcement"
}

Use an existing key from your Overwride account in place of the environment variable. Keep it on the server; browser code should call your own backend. The full HTTP API reference documents authentication and error responses.

Keep the returned record with the post

This illustrative response shows the shape of a successful result. The sample ID and wording are examples, not a live verification result.

{
  "id": "7Gx2kQm9Pa",
  "rewrite": "Dark mode is live today. Mia fixed the chart colours.",
  "footer": "Overwride rewrite. Check: overwride.com/c/7Gx2kQm9Pa",
  "draft_retained": true,
  "balance": { "credits_mills": 0, "seat_remaining": 998 }
}

Store id beside your post. Show the rewrite for review and offer the returned footer as an optional disclosure line. The saved draft is public through record lookup, so tell authors what will be visible before they submit confidential material.

The response fields describe a rewrite and record. There is no watermark_removed field or verified “human-written” score. Do not infer either from a successful HTTP response.

Checking and reading records

POST /api/v1/check accepts a post or record link. GET /api/v1/c/<id> returns a public record. Both are free and need no key. They look up Overwride records; they do not detect Claude, OpenAI, or xAI watermarks. See the endpoint reference for schemas and limits.

If an author edits the returned text, do not assume an exact-text lookup will still match. Preserve the record ID and use the link to show the saved versions. The current API reference does not document an endpoint for attaching an edited final version.

Plans and cost

API access is included in Pro at €30/month with 1,000 rewrites, and in the API plan at €129/month. Pro uses its allowance first; the API plan uses credits from the first rewrite. Credit-funded rewrites cost €0.018 each. Prices exclude VAT. Failed rewrites are not charged. See pricing.

Overwride uses its own ledger

xAI has scheduled default watermarking for supported Grok API models from 7 November 2026, with an opt-out setting. xAI API docs: output watermarking. Overwride has opted out of Grok’s text watermarking because we use our own ledger. The ledger keeps the submitted draft and rewrite together in a shareable record, with a disclosure link readers can open.

OpenAI’s public Content Provenance API handles media; text verification needs separate approval. OpenAI: Content Provenance API documentation. A general detector score is not a substitute for the relevant provider’s watermark test. Build the interface around the evidence your integration actually has: the draft, the rewrite, and a record readers can inspect.