Personalization

At the time of testing, Operator exists as a standalone application, separate from the main ChatGPT environment. Memory, an important feature recently introduced into ChatGPT, is not yet shared across to Operator. This separation means Operator currently cannot leverage previously stored user context such as preferences, history, or inferred habits. In practice, this severely limits global personalization across tasks. For example, even if a user had previously discussed their travel preferences (e.g., “I prefer Marriott hotels” or “I avoid early morning flights”) inside ChatGPT, Operator would still need to independently rediscover or re-request these inputs. Given OpenAI’s pace of integration between its memory systems and agents, we expect this to be an active area of development.

User-Specific Personalization & Basic Preferences

Today, neither ChatGPT nor Operator proactively collect structured user preferences at sign-up or during usage, preferences such as name, age, gender, location, or budget thresholds. As user’s leverage chat conversations in ChatGPT, it can proactively store memory and provide users with the capability to manage that memory, however, this information being collected available in a structured way helps with a fundamental level of personalization. In typical consumer-facing applications across Retail, Food Delivery, and Travel industries, capturing basic user attributes is fundamental to creating frictionless experiences. For Operator, even lightweight persona modeling (e.g., age group, location, marital status, presence of kids) would yield to Better task defaulting, Faster decisions & Reduced clarification rounds during interactions

Vertical Specific Personalization (Use-Case Specific Memory)

Different verticals naturally demand different types of personalization. Operator would benefit from vertical-specific memory structures depending on the use case category. Here’s a breakdown based on tested Operator use cases:

Shopping: Size preferences, color preferences, preferred brands, budget range

Food Delivery: Dietary restrictions, family size, preferred stores or cuisines

Travel: Loyalty programs (e.g., Marriott, Delta), preferred airports, preferred hotel brands, seat / room / vehicle preferences

Events: Seating preferences (aisle, front rows), budget limits

Local Services: Address confirmation, timing preferences, service provider ratings threshold

In-Line Memory Management

While ChatGPT introduced memory management through explicit chat messages, Operator does not yet manage in-chat preference updates in real-time. For example, when a user casually mentions, “I prefer non-smoking hotel rooms” or “I usually shop under a $100 budget”. Operator currently treats it as a one-off instruction rather than dynamically adjusting its memory model or behavior in-session. Inline, lightweight memory capture, automatically remembering such preferences and applying them for the next actions, would drastically improve the agent’s effectiveness without burdening the user to restate preferences repeatedly.

Global Personalization

Global-level personalization allows Operator to adapt choices based on large cohort behaviors, such as preferring Uber over Lyft in cities where it dominates, without needing deep individual user profiles. What makes global personalization uniquely powerful is that it can be informed not just by static datasets but by the broader internet corpus: aggregating real-time service availability, user reviews, regional preferences, and operational reliability from diverse public sources. Operator is particularly well positioned to action on this opportunity because it already combines strong reasoning models with tool use capabilities (e.g., web browsing and search), enabling it to dynamically gather, reason over, and personalize workflows based on large-scale cohort intelligence

App Specific Personalization Settings

Operator today allows users to personalize actions only at the app-specific level, buried deep within settings for each website or tool. While technically available, this approach is cumbersome for users and risks low adoption. Simple nudges during task execution (“Would you like me to remember this for next time?”) or a global preference dashboard across apps would make personalization more proactive, discoverable, and sticky.

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.