TextWatermark Logo TextWatermark.
Sign Up

Use cases

TextWatermarking hides a short payload in any text without changing what people see, and reads it back later. The pages below show what that is useful for, who it is for, the code to add it to an existing workflow, what it costs and where it stops working.

AI app and agent builders

Label AI-generated text at the source

An invisible label on every response, streamed or not, so you can tell later which response a text came from.

Agent builders

Let your agents sign what they write

Reports and emails signed with your channel key; verify origin and unchanged words before trusting them.

Publishers and companies

Trace a leaked document to its recipient

Identical-looking copies with an encrypted recipient id; read the id from a leaked copy.

Writers, agencies, legal teams

Prove authorship and detect edits

Stamp owner, date and a fingerprint of the words; a check says unchanged, edited, cut short or not yours.

Educators and learning platforms

Show where AI tutor answers end up

Label tutor answers and find them in submissions, with rules for using the result fairly.

Which mode do I need?

You wantMode and endpointMinimum text lengthWho can read the payload
A label inside a streamed responsePlain tokens, /tokensnone: insert after the first characteranyone with an API token
A label on a finished textPlain, /encodea few charactersanyone with an API token
A private id or a signatureKeyed: /encode with a channel-bound token(payload bytes + 30) × 8only tokens on your channel
To know when a text was cut short or damagedRobust: /encode-robust (keyed or plain)(payload bytes + 3) × 8; keyed (payload bytes + 30) × 8plain: anyone with a token; keyed: your channel

Keep plain payloads opaque (an id that only means something in your own system), because any TextWatermarking user can read them. Signing channels are part of the Starter and Pro plans.

What holds for every use case

  • The mark survives copy and paste between apps, chat and email. Text added after it does no harm.
  • Rewriting, translation, retyping, OCR, screenshots and Unicode normalization remove it, as does any tool or exporter that strips invisible characters. Test your own pipeline.
  • It is not an AI detector: a missing mark proves nothing about who or what wrote a text.
  • Each mark and each found payload costs 1 credit. Checks that find nothing are free, and so are refused requests.

Frequently asked questions

What does TextWatermarking add to a text?
Invisible Unicode variation selectors attached to the visible characters. The words, spacing and layout stay the same; the selectors carry a short payload, such as a response id, a recipient id or an owner, which the API reads back later.
What survives and what does not?
The mark survives copy and paste between apps, chat and email. It is removed by rewriting, translating, retyping, OCR, screenshots, Unicode normalization and tools that strip invisible characters.
Is it an AI detector?
No. It only finds marks that were added with TextWatermarking, for example by your own app. Text without a mark may still be AI-generated.
What does a check cost when nothing is found?
Nothing. Decoding text that carries no mark, or a mark that your token cannot read, is free. Only a found payload costs one credit.