OpenAI is launching a new feature for ChatGPT called Pulse. This feature generates personalized reports for users while they sleep. Pulse creates five to ten briefs designed to get users up to speed on their day. The goal is to encourage people to check ChatGPT first thing in the morning, similar to how they might check social media or a news app.
Pulse is part of a broader shift in OpenAI’s consumer products. These products are increasingly designed to work asynchronously instead of just responding to user questions. Features like the ChatGPT Agent or Codex aim to make ChatGPT feel more like an assistant than a chatbot. With Pulse, OpenAI wants ChatGPT to become more proactive.
The new CEO of Applications, Fidji Simo, stated that the company is building AI to make a high level of support available to everyone. She described ChatGPT Pulse as the first step in that direction, starting with Pro users today with the goal of rolling it out to all users over time.
OpenAI CEO Sam Altman indicated that some new compute-intensive products would be limited to the most expensive subscription plan, which is the case for Pulse. The company has previously stated it is limited in the number of servers available to power ChatGPT. OpenAI is rapidly building new AI data centers with partners like Oracle and SoftBank to increase its capacity.
Pulse will roll out for subscribers to the two-hundred-dollar-a-month Pro plan. It will appear as a new tab in the ChatGPT app. The company hopes to launch Pulse for all users in the future, with Plus subscribers gaining access soon, but it first needs to make the product more efficient.
The reports generated by Pulse can be roundups of news articles on a specific topic, such as updates on a sports team. They can also be more personalized briefs based on a user’s specific context.
In a demonstration, an OpenAI product lead showed several reports Pulse created for him. These included a news roundup about the British soccer team Arsenal, group Halloween costume suggestions for his family, and a toddler-friendly travel itinerary for an upcoming trip.
Each report is displayed as a card featuring AI-generated images and text. Users can click on a card to get the full report and then query ChatGPT about the contents. Pulse will proactively generate some reports, but users can also request new automated reports or provide feedback on existing ones.
A core part of the design is that Pulse stops after generating a few reports and displays a message stating that is it for the day. This is an intentional choice to differentiate the service from engagement-optimized social media apps.
Pulse is compatible with ChatGPT Connectors, allowing users to link apps like Google Calendar and Gmail. Once connected, Pulse can parse through emails overnight to surface important messages in the morning or access a calendar to generate an agenda for upcoming events.
If users have ChatGPT’s memory feature turned on, Pulse will pull context from previous chats to improve the reports. For example, the feature might automatically use a known interest in running to create an itinerary for a trip that includes running routes.
Pulse is described as a new functionality for a consumer product. It can use information like dinner reservations on a calendar to find menu items that align with a user’s dietary preferences, such as pescatarianism.
The feature has the potential to compete with existing news products like Apple News, paid newsletters, or traditional journalism outlets. However, the product lead does not expect Pulse to replace the various news apps people use. The feature cites its sources with links, similar to ChatGPT Search.
It remains to be seen if Pulse is worth the computational power it requires. The service can vary tremendously in how much computing power it uses for a given task. Some projects are fairly efficient, while others may require extensive web searches and document synthesis.
Eventually, OpenAI would like to make Pulse more agentic. This could include the ability to make restaurant reservations on a user’s behalf or draft emails for approval. Such features are likely a long way off, as they would require significant improvements to OpenAI’s agentic models before users would trust them with such decisions.