OpenAI Reduces ‘Teaser-Phrasing’ in ChatGPT and Other Foundation Models

OpenAI is taking steps to reduce the tendency of its large language models (LLMs), including ChatGPT, to use “teaser-phrasing” – a technique where the AI subtly encourages users to continue the conversation. This adjustment aims to provide more direct and complete responses, enhancing the user experience and building trust in the technology.

What is Teaser-Phrasing?

Teaser-phrasing refers to the practice of an LLM responding in a way that hints at further information or a more complete answer if the user continues the conversation. While not intentionally deceptive, this tactic can lead users to believe the initial response is incomplete or requires further prompting to unlock its full potential. This can be frustrating for users seeking concise and direct answers.

Why is OpenAI Addressing This?

OpenAI recognizes that teaser-phrasing can erode user trust and create a less efficient interaction. By reducing this behavior, the company aims to deliver more satisfying and transparent responses. This aligns with OpenAI’s broader commitment to responsible AI development and user satisfaction. The change is part of ongoing efforts to refine the behavior of its foundation models.

How are the Models Being Updated?

OpenAI’s foundation models are developed using information from a variety of sources, and are continually refined through training and feedback. The specific methods used to reduce teaser-phrasing haven’t been detailed, but likely involve adjustments to the models’ training data and reward functions. These adjustments encourage the models to prioritize completeness and directness in their responses.

OpenAI’s Current Models

As of March 2026, OpenAI offers a range of models catering to different needs:

• GPT-5.4: OpenAI’s flagship model, best suited for complex reasoning, coding, and professional workflows. Input costs $2.50 per million tokens, and output costs $15 per million tokens. It has a 1 million token context window.
• GPT-5.4 mini: A stronger mini model ideal for coding, computer use, and subagents. Input costs $0.75 per million tokens, and output costs $4.50 per million tokens. It has a 400K token context window.
• GPT-5.4 nano: The most affordable GPT-5.4 class model, designed for simple, high-volume tasks. Input costs $0.20 per million tokens, and output costs $1.25 per million tokens. It also has a 400K token context window.

All current OpenAI models support text and image input, text output, multilingual capabilities, and vision. They are accessible through the Responses API and Client SDKs.

What are Large Language Models?

Large language models (LLMs) are computational models trained on massive datasets to process and generate human language. They are a key component of many modern AI applications, including chatbots, translation tools, and content creation platforms. Learn more about LLMs.

Looking Ahead

OpenAI’s move to reduce teaser-phrasing represents a positive step towards more transparent and user-friendly AI interactions. As LLMs continue to evolve, we can expect further refinements aimed at improving their reliability, trustworthiness, and overall value to users. The ongoing development of models like GPT-5.4 and its variants demonstrates OpenAI’s commitment to pushing the boundaries of AI capabilities.