Finally, OpenAI Says ChatGPT Will Listen When People Tell It to Avoid the Long Dash

Sam Altman says ChatGPT finally listens when people tell it to drop the long dash (yeh this beast —) that has followed the chatbot everywhere. Users treated that punctuation as a flashing sign that a machine wrote their text. Some called it distracting. Others felt embarrassed when it slipped into work emails or reports even after they tried to rein it in.

OpenAI pushed out a small fix that changes how the model reacts when a user writes a clear instruction inside the personalization panel. Altman framed it as a simple win, and the move arrived shortly after the company rolled out the new GPT 5.1 model. It sounds minor on the surface. Yet people who rely on the tool know how often the bot stubbornly added that long mark even when asked to avoid it.

Writers said the habit broke their tone and made their work stand out for the wrong reasons. Many stopped using the dash in their own writing because they did not want readers to assume a chatbot drafted their text. Complaints piled up across forums where people kept posting examples of the model promising to avoid it, then slipping it back into the very next sentence.

The new behavior only kicks in when the user plants the instruction in the custom settings area. Altman did not promise success every time in regular chats. That fits with the broader reality of LLM behavior. These models shift output by leaning on probability patterns rather than fixed rules. If a user places the instruction in the right slot, the odds of a clean output increase, though nothing becomes absolute.

Some critics pushed the conversation in another direction. They pointed out that if OpenAI struggled for years to control one simple punctuation mark, talk of near term general intelligence feels a bit premature. The model may look sharp on the surface. Yet it still works like a giant pattern engine that tries to anticipate what should come next rather than follow strict commands with mechanical precision.

Older training data also played a role. People have used the long dash for centuries. It showed up across novels, editorials and essays that filled older datasets. Because the model tries to echo the shape of the writing it has seen, the dash became a default move. Once reinforcement learning kicked in and evaluators rewarded responses that felt polished, the preference grew stronger. That gave the model a habit that stuck around even as users pushed back.

OpenAI now says the fix is part of its work to hand people more control. The company already introduced tools that remember user preferences and let people fine tune how the bot behaves across sessions. The long dash update shows that simple choices matter to users just as much as headline features. For many, this is less about punctuation and more about trying to make the output feel like their own voice.

Every change will still depend on how the model handles probabilities in the background. That leaves room for odd behavior to creep back after future updates. Some users already say the fix works inside the settings panel but still fails if you only mention it inside the chat. With a system that keeps learning from new interactions, small shifts can break old tuning in unpredictable ways. Anyone expecting a crisp on off switch will need patience.

Still, for now, people who truly want to avoid the long dash have a practical way to do it.

How To Add a No Em Dash (—) Rule in Custom Instructions

Below is a clear set of steps based on OpenAI’s official customization guide. You only need to do it once. After that, ChatGPT will try to follow the rule in every conversation.

Step 1: Open the Custom Instructions Panel

1. Open ChatGPT in your browser or app.
2. Look for your profile picture in the bottom corner, then go to "Settings" option and then "Personalization" tab (you can also directly access it through this link).
3. Now in the Personalization tab you will be able to see Custom Instructions option.

Step 2: Add Your Style Requirement

You will see two large text boxes. One controls how ChatGPT should respond. This is where you add the rule.

Write something like:

"Do not use em dashes in any response. Replace them with normal punctuation."

or

"Avoid using em dashes unless necessary for clarity or emphasis; otherwise, use standard punctuation."

Keep it short and clear so the model can pull the instruction into every session.

Step 3: Save the Setting

Scroll down and hit Save.

The instruction becomes active across all chats unless you turn the feature off or erase it later.

Step 4: Test the Behavior

Start a new conversation. Ask the model to write a few lines of text.

Scan the output to confirm it avoids the long dash. (or even ask Chatgpt to analyzes and assess the — use)

If it does not, double check the instruction or rewrite it with simpler language.

Step 5: Adjust Anytime

You can change, refine or remove the rule by visiting the same panel.

This helps if you want different styles for different tasks.

Note: This post was edited/created using GenAI tools and proofread/fact-check by human editors.

Read next: 

• ChatGPT Experiments With Real Group Conversations in a Limited Rollout

• 3 Out of 4 Americans Willingly Trade Personal Data For Discounts Despite Privacy Fears

by Asim BN via Digital Information World