Highlights of the week
The configuration data for week 32 of 2026 shows a continuation of the personalization efforts observed last week, with the definitive replacement of the general onboarding instructions by the guided writing style setup. The advertising infrastructure is expanding its partnerships, adding Expedia to the list of specific product cards alongside the previously observed Temu integrations. In the search and execution domains, new third-party application identifiers for local search and reservations have appeared, which is highly relevant for generative engine optimization. Finally, the user interface is preparing for significant updates regarding rich text editing and local indexing visibility.
Focus by theme: search
The transition toward a personalized writing style setup, initiated last week with the "writing-style beacon", is now fully visible in the system instructions. The configuration confirms that the previous general "ChatGPT Work" onboarding has been entirely replaced by a guided collaborative setup. The system is instructed to learn exclusively from user-authored writing in past chats and to ignore third-party documents or incoming messages to preserve the user's authentic voice.
In the local search and entity recognition domain, the map widget continues to evolve. The map-entity-notes-experiment-enrolled and enable-map-entity-notes-sidecar flags have been activated, and the local home card limit has been set to nine. Furthermore, a new configuration introduces specific application identifiers for Resy, OpenTable, and Yelp. This suggests that OpenAI is actively structuring its local business and restaurant reservation integrations, which represents a significant development for local generative engine optimization.
The shopping search experience shows minor adjustments, with the sidebar_review_alternatives_only flag now enabled. Additionally, the instant answers infrastructure has been updated to include flags for enable-entity-autocomplete and dynamic entity autocomplete configurations, indicating ongoing work on predictive search queries.
Focus by theme: web run
The execution environment shows the activation of the enable_graphable_function_v2_renderer flag within the learning blocks infrastructure. This follows the strict constraints on mathematical visualizers observed in week 31 and suggests that a new version of the graph rendering engine is now active for users.
Another configuration shows that the structured_input_enabled flag has been turned on, which might indicate an improvement in how the web execution environment parses and processes complex user queries or data tables.
Focus by theme: retrieval and grounding
The third-party integration strategy shows a nuanced evolution compared to last week. While week 31 indicated a prioritization of the Google workspace ecosystem and a disabling of Microsoft tools, a new configuration this week specifically adds sharepoint_allowed_sites and sharepoint_site_access_policy to an enterprise connector. This suggests that Microsoft enterprise integrations are being refined with granular access controls rather than being permanently deprecated.
The translation features observed previously are advancing, with the file_translation_enabled and mobile_source_first_enabled flags now active under a redesign_file variant. This points to an upcoming interface overhaul for document translation.
Regarding audio grounding, the dictation infrastructure continues to mature. The Android dictation bar is now enabled by default, solidifying the low-latency audio input features tested in previous weeks.
Finally, local data retrieval shows significant movement. The flags enable_local_indexing and enable_ux have been activated alongside a wave parameter set to 10. This indicates that the user interface elements required for indexing local files and conversation history are being rolled out.
Focus by theme: ads
The advertising infrastructure continues to expand its roster of dedicated agents and card formats. Following the introduction of Temu product cards and personalized ads last week, the configuration now includes specific identifiers for Expedia, labeled as expedia_business_agent and expedia_prod_agent, as well as a homh_prod_agent. This confirms that OpenAI is building custom advertising agents for major travel and e-commerce platforms.
The onboarding screens for the "Ask ChatGPT" ad feature have also been updated. The configuration now specifies English-localized image assets and includes fallback images for scenarios where ad personalization is unavailable, ensuring a consistent transparency experience for the user.
To monitor
A major update to the composer interface appears to be imminent. Flags for rich_text_input, rich_text_message_editing, rich_text_user_turns, and expanded_composer_mode have all been enabled. This suggests that users will soon be able to format their prompts using rich text directly within the input box.
New internal model identifiers have appeared in the configuration, specifically gpt-5.5-cyber-preview, gpt-daybreak-5.6-sol, and gpt-daybreak-red-5.6-sol. While these are based on the publicly available GPT-5.5 and GPT-5.6 models, the "cyber-preview" and "daybreak-red" modifiers suggest specialized testing environments or security-focused variants, potentially related to internal red-teaming efforts.
The "ghostrider" image generation workflow remains under active configuration, with the progress_enabled flag turned off and the addition of a summarized_tools parameter.
Lastly, the text input capacity has been doubled in a specific configuration, with the show-in-text-field-max-chars limit increasing from 25,000 to 50,000 characters, which may affect how users paste large documents directly into the chat interface.
