OpenAI adds memory—and eyes ads

OpenAI is teaching ChatGPT to remember more of your work at the same moment it is building an ad business around that memory. On April 9, Reuters reported that OpenAI told investors to expect about $2.5 billion in ad revenue in 2026 and $100 billion by 2030. (reuters.com) Memory in ChatGPT is simple in concept: instead of treating every chat like a blank page, the system can carry over details from earlier conversations. OpenAI’s Memory frequently asked questions page says ChatGPT can remember preferences, interests, and other useful context between chats. (help.openai.com) OpenAI splits that memory into two buckets. One bucket is “saved memories,” which are details you explicitly ask it to keep, and the other is “chat history,” where it can reuse information from past conversations to make later answers more tailored. (help.openai.com) That is moving from a chatbot toward something closer to a long-running assistant. OpenAI says projects let users group chats, files, and instructions in one workspace so ChatGPT “remembers what matters and stays on-topic” across repeated work like research, planning, and writing. (help.openai.com) Projects matter because they turn memory from a personal preference feature into a work surface. OpenAI’s help page says a project can hold uploaded files, project-specific instructions, and chats in one place, and those instructions override your global instructions inside that project. (help.openai.com) OpenAI is also adding more ways for ChatGPT to know where and how you are using it. In March 2026, OpenAI added optional device location sharing so ChatGPT could give more relevant local recommendations, news, and weather, and said precise location is deleted after it is used for the response. (help.openai.com) Put those pieces together and the product starts to look less like a one-off question box and more like a feed that knows your habits. A system that remembers your diet, your writing style, your files, your location, and your ongoing projects has far more signals for ranking answers than a system that only sees one prompt. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) That is where the ad story changes shape. CNBC reported on February 5 that OpenAI had already said it would begin testing ads with free users and ChatGPT Go subscribers in the United States, and that the company said those ads would be clearly labeled, placed at the bottom of answers, and would not influence responses. (cnbc.com) Even with that promise, conversational ads are a harder design problem than search ads. In a search engine, people already expect a page of links with sponsored results mixed in, but in a chatbot the answer itself is the product, so users have to trust that the model picked a recommendation because it fit the question, not because someone paid for placement. (cnbc.com) (reuters.com) OpenAI’s own controls show it knows this is sensitive. The company says users can ask what ChatGPT remembers, delete individual memories, clear all memories, turn memory off, or use Temporary Chat so no memories are referenced or created. (help.openai.com) The business logic is not hard to see. If OpenAI thinks ads can grow from $2.5 billion this year to $53 billion by 2029 and $100 billion by 2030, as Reuters reported from the Axios account of investor materials, then memory is not just a convenience feature; it is part of the infrastructure for making conversational surfaces more targetable and more commercially valuable. (reuters.com) So the next fight is not whether chatbots can show ads at all. It is whether a product that remembers your projects, files, preferences, and location can keep the line between “helpful personalization” and “paid influence” bright enough that users still believe the answer in front of them. (help.openai.com 1) (help.openai.com 2) (cnbc.com) (reuters.com)