OpenAI’s new text watermark, called textGrain, embeds a statistical signal in the word choices of some AI-generated text. It is not a visible label or hidden-character string. OpenAI says API customers worldwide can opt in for selected models, while eligible ChatGPT and Codex text output in the European Union is due to receive the watermark over the coming weeks. The detector is not publicly available at launch.
What is textGrain?
textGrain is OpenAI’s method for embedding an invisible statistical signal in generated text. Rather than adding a tag to a file, the system subtly influences the model’s choice of words or word pieces. Across a sufficiently long passage, those choices can form a pattern that a detector can assess. OpenAI says it does not insert invisible spaces, hidden characters, or unusual punctuation; the signal is in the wording itself. (OpenAI’s October 5, 2026 announcement; OpenAI Help Center)
As an Amazon Associate I earn from qualifying purchases.
This differs from metadata-based credentials, which attach information about a file’s origin or history, and from classifier-style detectors, which infer likely AI authorship from text patterns without checking for a specific embedded signal. OpenAI says its broader provenance work uses C2PA Content Credentials for supported images; those image systems are not textGrain. (OpenAI’s provenance overview)
Where is OpenAI adding the watermark?
OpenAI announced two rollout tracks on October 5, 2026:
#1 Best Overall
- API: Customers worldwide can opt in for selected models. Watermarking is off by default.
- ChatGPT and Codex: OpenAI said eligible text output in the European Union would receive the watermark over the coming weeks. The announcement does not specify that all models, account types, or outputs are eligible.
OpenAI cited the EU AI Act and its commitments under the EU Code of Practice on Transparency of AI-Generated Content as its rationale for the EU rollout. That statement describes the company’s reasoning; it does not by itself resolve how every legal obligation applies. (OpenAI announcement; OpenAI Help Center)
Who can check whether text has the watermark?
OpenAI is accepting applications for detector access from researchers and expert organizations. The detector is not publicly available at launch. OpenAI’s API documentation describes its content-provenance checks as covering supported OpenAI signals, not as a general-purpose detector for text from every AI system. (OpenAI announcement; OpenAI Developers documentation)
How well does textGrain detect marked text?
The figures below are results OpenAI reported from its own evaluations on October 5, 2026. They are not independent validation, and performance in those evaluations does not guarantee reliable results on arbitrary text.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems| OpenAI-reported evaluation | Reported result |
|---|---|
| 200-token psychology passages, at a 1% target false-positive rate | About 80% detection |
| 400-token psychology passages, at a 1% target false-positive rate | About 95% detection |
| 400-token passages after replacing 10% of words with synonyms | Detection fell from about 92% to 66% |
| 400-token passages after replacing 25% of words with synonyms | Detection fell to 17% |
| Mathematics passages | Substantially lower detection; OpenAI attributed the difficulty to less flexible word choice |
OpenAI said textGrain matched or exceeded other approaches it tested, including SynthID for text. That is a comparison reported by the company, not evidence of independent replication. OpenAI also cautions that strong results in ideal conditions may not translate to everyday detection. (OpenAI announcement)
Rank #3
- Used Book in Good Condition
Why can a watermark be missed or weakened?
Detection depends on the passage having enough flexible language for the watermark pattern to emerge and remain intact. OpenAI identifies several limits:
- Short passages: They may not contain enough text for a reliable result. OpenAI’s Help Center cites the EU Code of Practice as not requiring watermarks for outputs shorter than 200 tokens—about 150 words in English—or for code snippets.
- Code and constrained writing: Code and mathematics offer fewer plausible word or token choices, leaving less room to embed the signal.
- Editing: Replacing words with synonyms can weaken detection, as OpenAI’s reported evaluation figures illustrate.
- Translation or unsupported generation: A passage may be translated, produced by an older or unsupported model, or generated by another company’s system and therefore lack a detectable OpenAI watermark.
These limitations mean a detector result should not be treated as a complete account of a passage’s origin. (OpenAI Help Center; OpenAI announcement)
Rank #4
What does a detection result prove?
A positive result is evidence that an OpenAI system likely generated or processed some of the passage. It does not tell you how much text came from the model, how much a person edited it, who used the system, who owns the text, whether its use was lawful, who is responsible for it, or whether its claims are true. OpenAI states: “A watermark does not measure human contribution.” (OpenAI announcement)
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchA negative result does not prove that a person wrote the passage. The watermark may be absent because the text is short, constrained, edited, translated, generated by an unsupported or older model, or produced by a different AI system. OpenAI’s own warning is explicit: “The absence of a detected watermark does not prove human authorship.”
That distinction matters in schools, workplaces, publishing, and moderation: textGrain is a provenance signal, not an authorship verdict or a truth-checking tool. OpenAI describes text watermarking and detection as early technologies with significant limitations. (OpenAI announcement)
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
