WildSEO launches tool to rewrite around AI text watermarks

WildSEO has released TextGrain Remover, a free tool that removes hidden Unicode characters from text and rewrites passages intended to weaken statistical AI-watermark patterns. The browser-based service accepts pasted text or public article URLs, handles passages of up to about 1,300 words and does not require an account or credit card. The launch on Oct. 7 arrives days after OpenAI announced textGrain, an invisible watermark based on word-choice patterns rather than inserted characters. As reported on Oct. 5, OpenAI had begun an EU-focused rollout of text watermarking for eligible ChatGPT and Codex users while allowing API customers worldwide to opt in for selected models. TextGrain Remover has two distinct functions. Its Clean Only mode removes zero-width spaces, soft hyphens, directional marks, hidden tag characters, variation selectors and other non-printing material without changing the words themselves. Its Rewrite & Clean mode uses AI to rewrite text sentence by sentence, seeking to disrupt predictable n-gram and token-frequency patterns while retaining names, figures, citations, links, quotations and factual claims. A side-by-side comparison and inspection tool show the wording changed, hidden characters removed and whether concrete details appear to have been lost. WildSEO’s central limitation is that rewriting cannot establish that textGrain has been removed. OpenAI’s detector is not public, with access initially limited to approved researchers and expert organizations. Accordingly, the remover reports the extent of wording changes rather than a percentage of watermark removal. The distinction matters because the two types of signal work differently. Hidden-character cleaning deletes actual characters embedded in a passage. Statistical watermarking instead biases word choices where several continuations are plausible, allowing a detector with the relevant key to identify an unusual pattern across a sufficiently long text. OpenAI’s own tests illustrate how editing can erode such a signal. At a 1% target false-positive rate, it detected watermarks in about 92% of unedited 400-token passages. Replacing roughly 10% of words with synonyms reduced detection to 66%, and changing 25% reduced it to 17%. The company also found detection rates of about 80% for 200-token passages and 95% for 400-token passages in psychology content, while warning that performance is lower for more constrained material such as mathematics. It said substantial paraphrasing or translation can make a watermark undetectable, but the absence of a detection does not prove that text was written by a human. WildSEO said the tool is intended for draft-stage content hygiene, alongside its broader platform for monitoring citations and visibility in services including ChatGPT, Claude, Perplexity and Google AI Overviews. Its founder, Tim, said the aim was to remove “digital noise” while preserving authors’ facts and original voice. The clean-only operation runs locally in a user’s browser and sends nothing elsewhere. Rewrites, however, are processed by AI; WildSEO says it may retain the submitted text, results, model information and quality scores for up to 30 days before deleting the rewritten text. Anonymous scores and timing data may remain, and URL imports transmit both the public URL and extracted article copy for processing. The service advises users not to submit sensitive data. TextGrain Remover is available at WildSEO’s TextGrain Remover page, with hourly rewrite limits intended to keep the service free.