Paste the post into GrokDetect. A long post, an X Article or several replies joined together work. A single short reply is too little for a reliable answer, because the check needs at least 100 words.
Steps
- Copy the text of the post, or join the replies from one account into a single block, oldest first.
- Run the check and read the AI verdict, the highlighted passages and the family similarity.
- Treat a Grok-style result as a reason to look closer, not as a finding. Repeated structure across many replies is stronger evidence than one score. The habits to look for are listed in how to tell if Grok wrote it.
What does not work
- Looking for a watermark. Grok's text watermark is off by default and covers API text only, so posts on X carry none to find. See Grok watermark status.
- Judging from one line. Short fragments of human and AI writing look alike.
Screening at scale
Brands and moderators who receive many replies can run them through an API instead of one at a time, and group the results by account.
The same engine also runs for other models, such as ChatGPT.