Short answer: ordinary users currently cannot check ChatGPT text with OpenAI’s official text-watermark detector. OpenAI says access is restricted to approved organizations that apply for it. The watermark it describes is a statistical pattern in word choices—not hidden characters you can reveal by inspecting or copying the text.
What OpenAI’s invisible text watermark is
OpenAI calls its method textGrain. It subtly shifts a model’s choices among possible words or word pieces, creating a statistical pattern across generated text. A detector with the corresponding secret key and settings checks whether that pattern appears more often than expected by chance.
The signal is part of the wording. OpenAI says textGrain “does not add hidden characters, invisible spaces, or unusual punctuation.” You therefore cannot identify it by showing formatting marks, inspecting Unicode characters, or copying and pasting the passage. OpenAI Help Center: Provenance signals in OpenAI-generated content.
What an ordinary reader can do
- Do not search for invisible characters. Their absence or presence is not a way to test for textGrain; the signal is statistical word-choice patterning.
- Do not ask ChatGPT to verify authorship. OpenAI says ChatGPT does not know whether it generated a particular piece of writing and may make up an answer when asked. OpenAI Help Center: Can I ask ChatGPT if it wrote something?
- Distinguish text checks from image and audio checks. OpenAI’s public verifier and Content Provenance API support certain image and audio files, not ordinary text. Text-detector access is application-based and currently limited to approved organizations, including AI research and academic institutions; applications are reviewed case by case. OpenAI API documentation: Content provenance.
- If your organization is approved, use the authorized detector. A meaningful assessment depends on the passage and its context, including length, language, wording flexibility, and any edits. OpenAI does not offer an ordinary-user text upload checker.
- Describe any result narrowly. A positive result is evidence that an OpenAI system likely generated or processed some of the text. A negative result is inconclusive, not proof of human authorship.
Why a detector can miss watermarked text
Detection is probabilistic. Short passages provide less signal to measure; constrained writing, such as mathematics, offers fewer flexible word choices; and editing or translation can weaken the pattern. TextGrain is not file metadata: copying unchanged wording is expected to preserve the signal, but substantial rewriting, paraphrasing, or translation may reduce or remove detectability. Even unchanged text may not be detectable if it is short or has little wording flexibility.
OpenAI’s published figures are its own evaluations, not independently established guarantees for arbitrary real-world documents:
- For psychology-like content, OpenAI reported detection of about 80% of 200-token passages and about 95% of 400-token passages at a 1% target false-positive rate. It reported substantially lower detection for mathematics, where wording is more constrained.
- In 400-token passages, replacing 10% of words with synonyms reduced reported detection from about 92% to 66%; replacing 25% reduced it to 17%.
- In an evaluation of 500 synthetic English prompts translated into the other 23 official EU languages, detection at a 1% false-positive rate ranged from 69.0% for Spanish to 42.2% for Romanian. OpenAI says it can adjust watermark strength for languages with weaker results.
- The European Commission’s Code of Practice on Transparency of AI-Generated Content, as described by OpenAI, does not require watermarks in outputs shorter than 200 tokens (about 150 English words) or in code snippets.
These results were reported by OpenAI in 2026 and apply to the stated evaluation conditions; they do not establish a universal detection rate or false-positive rate for other passages. OpenAI, “Our approach to EU text provenance rules” (October 5, 2026).
Text watermark detection is not the same as an AI-writing classifier
An authorized textGrain detector looks for a particular signal introduced during generation. A third-party AI-writing classifier instead analyzes linguistic patterns and estimates whether text appears AI-generated. A classifier score does not show that the OpenAI watermark is present, and a classifier that does not flag a passage does not settle who wrote it. OpenAI distinguishes these methods in its explanation of text provenance. OpenAI, “Our approach to EU text provenance rules”.
What a positive or negative result means
- Watermark detected: evidence that an OpenAI system likely generated or processed some of the text. It does not establish whether the model wrote all or only part of it, how much a person contributed, or who used the system.
- No watermark detected: inconclusive. The passage might be too short, constrained, edited, translated, from an unsupported model or product path, or created before watermarking was available. OpenAI states: “The absence of a detected watermark does not prove human authorship.”
- Neither result: establishes whether the text is accurate, who owns it, whether its use is lawful, what disclosure obligations apply, or who is responsible for it.
Is all ChatGPT text watermarked?
No. OpenAI’s October 5, 2026 announcement says API customers around the world can opt in to text watermarking for select models, with the feature off by default in the API. It also says eligible ChatGPT and Codex text output in the EU is scheduled to roll out “over the coming weeks.” Availability depends on region, product, model, export path, and rollout timing; these statements do not mean that every ChatGPT passage is watermarked. OpenAI, “Our approach to EU text provenance rules”.
Quick Recap
Rank #4
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.
