OpenAI CEO Sam Altman confirmed earlier via his personal "X" platform account that ChatGPT has fixed a long-standing "bad habit" that users have complained about. If users can now request ChatGPT not to use long dashes (—, i.e., Em Dash) in their "custom instructions," it will reduce the problem of AI-generated content being easily identifiable by humans.
The overuse of dashes has become the biggest flaw in AI writing.
Prior to this update, ChatGPT seemed to ignore user requests in the prompts, consistently using excessive long dashes in its generated text. The frequent use of excessive long dashes in content has become a criterion for many to determine whether content is AI-generated.
Therefore, once long dashes are used frequently or excessively in an article, even if not all content using dashes is generated by AI, this unique "article writing style" has already become a basis for people to question whether it is generated by AI.
Training data may be the main reason; it becomes more difficult to identify issues after updates.
It is unclear why generative AI models tend to use a large number of long dashes, but it is speculated that since large language models are trained on a large amount of books and online content (such as scientific papers, public forum posts or news articles), it is likely because long dashes appear frequently in the training materials of these models, and the AI training process did not mark them as symbols that should be avoided.
However, as OpenAI corrects this problem, it will obviously be more difficult to determine whether content is generated by AI by identifying whether a large number of long dashes are used. Therefore, users may need to examine online (and even printed) content more carefully in the future.
Small-but-happy win:
If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it's supposed to do!
- Sam Altman (@sama) November 14, 2025
