OpenAI updated ChatGPT's memory system in June, shifting from manual "remember this" commands to an automated process the company calls dreaming. The chatbot now builds and edits a profile about you over time, deciding on its own what to retain. This affects every response you get, and for professionals in fields like marketing, legal, healthcare, and customer support, those assumptions can quietly steer outputs in directions you did not intend.
If you spent a week researching a vacation spot, ChatGPT might assume you live there. A deep dive into home repair could tag you as a DIY expert. The bot uses this stored context to personalize answers, but it will not always make correct assessments. Understanding how this works matters because memory now operates automatically, and its influence on responses is easy to miss.
How the memory upgrade changed
When memory launched in April 2024, users had to explicitly tell ChatGPT "remember this." Those memories were static. If you mentioned an upcoming party, the bot treated it as a future event indefinitely, even after the date passed. The June update introduced flexible memories that update over time. The party moves from future to past without a manual correction.
OpenAI describes three goals for the feature: carrying forward useful context, following preferences and constraints, and staying current as facts change. The trade-off is control. The chatbot now sifts through conversations and saves what it deems important. You no longer have to issue commands, but you also do not get to approve what goes into the profile it builds about you.
Finding and editing what ChatGPT remembers
You can inspect the memory profile in the web app by clicking your profile icon in the bottom left, then choosing Personalization and scrolling to the Memory section. On mobile, tap the menu button, then your profile icon, then Memory. A toggle switch lets you disable the feature entirely. Turning it off stops new memories from forming but preserves existing ones in case you re-enable it later.
Select Manage next to Memory summaries to see the profile. It covers interests, preferences, and how you like answers delivered. You can highlight any entry and choose Make a correction or Don't mention this again. A prompt box at the bottom accepts commands like "forget that I'm a journalist." To wipe everything, tap the three dots at the top, select Delete and turn off memory, and confirm. Past conversations remain, but if you turn memory back on, ChatGPT will re-index those chats and rebuild its understanding of you from scratch.
Another option is the temporary chat feature. Nothing typed in a temporary chat gets committed to memory, and the conversation disappears when you close it. On the web, click the dotted speech bubble icon in the top right to start one. This is useful for sensitive queries or one-off research where you do not want the interaction shaping future responses.
Why this matters for creatives, support teams, and business professionals
For anyone using ChatGPT to draft marketing copy, prepare legal summaries, or handle customer inquiries, a flawed memory profile creates real friction. If the bot believes you are based in a city you only researched for a client, location-specific suggestions will be wrong. If it assumes expertise you do not have, its explanations may skip steps you need. For writers and content teams, a memory cluttered with unrelated projects can muddy tone and style recommendations.
Checking the memory profile should become a regular habit, much like reviewing browser cookies or app permissions. A quick audit after a project wraps up prevents cross-contamination between work streams. For teams exploring ChatGPT Courses or AI Productivity Courses, understanding memory management is a practical skill that directly affects output quality. The feature is convenient, but the responsibility for accuracy still sits with the user.
Your membership also unlocks:
