OpenAI's Dots agents, unveiled at DevDay this week, can check connected sources for new information before a user asks — a capability the company calls "proactive research" — but the mode is strictly read-only, according to the help documentation published alongside the rollout.

The help center's privacy and safety page spells out the limits directly: "Your dot can also look for useful updates before you make a request. This is called proactive research. It can read information from permitted connected sources and save private notes, but its research tools cannot directly: Send messages to other people. Change content through plugins. Control a browser or computer."

That quote is OpenAI's own documentation — a confirmed statement, not tool-signal speculation. Every action a dot takes with what it finds runs through the usual safety checks and rules.

Availability, as of the rollout

Dots are now reaching Pro users in various markets, excluding the European Economic Area, Switzerland, and the UK. Business Premium users get access across all supported ChatGPT regions. Enterprise, Edu, and Healthcare workspaces can try a beta version, which ships off by default. OpenAI warns it might take several days for access to appear on an account.

Custom Rules cannot override the restrictions

Custom Rules — per-user settings governing what a dot can do independently — allow adjustments that make a dot more or less flexible. However, OpenAI states these settings cannot override the proactive research restrictions. The dot itself cannot override them either.

The unprompted mode is limited to two things: reading and taking notes.

Scheduled work moves onto an agent

Scheduled tasks are not new to ChatGPT; the platform's help page on Scheduled tasks already covers one-time and recurring tasks, change monitoring, and responses to supported events from Gmail, Slack, or GitHub.

A dot extends that capability across apps a user has connected, and it maintains context between conversations. The getting-started page gives a concrete example: a dot reviewing a calendar every morning and reporting upcoming events. Both scheduled runs and proactive updates happen within the same conversation thread.

What persists between conversations

A dot draws on plugin connections the user has already set up. Plugin permissions are shared across dots, ChatGPT, ChatGPT Work, and Codex, per the privacy and safety page. The launch post also notes the plugin catalog now offers over 4,000 apps.

The dot builds its own memory from these connections. Disconnecting an app stops new access but does not erase what the dot has already saved. Users currently cannot view or edit those individual memories. Resetting the dot deletes its conversations, saved memories, and scheduled tasks.

One notable absence: as of Sept. 30, none of the three Dots help pages mention an analytics or search platform. The examples of connected apps include a calendar, Slack, and a personal email account, and the getting-started page cautions that available apps may vary.

Browsing during assigned work

The restrictions loosen when a user assigns work. While working on an assignment, a dot can operate its own browser and cloud computer. Enterprise admins can enable or disable cloud browser use, cloud network access, and cloud computer use individually.

Dots now sit alongside the August WebMCP deployments from OpenAI, Shopify, and Cloudflare, which gave agents structured methods for operating within websites. Notably, neither the announcement nor the three support pages link Dots to search rankings or request website modifications. What Dots introduce instead is another agent capable of browsing a site on a user's behalf — a traffic pattern site owners may want to account for as agent-mediated visits grow.

What this means for search marketers

Search marketing runs on periodic checks: rankings, competitor pages, campaign settings — each check often starting from scratch. A dot can be assigned monitoring duties, watch the specific sources it is permitted to access, and report changes. Because proactive research is read-only, a dot could detect a change overnight without any write-side risk; any follow-up actions must pass through OpenAI's action rules and safety checks.

Whether a dot fits into a team's monitoring setup depends entirely on which platforms it can read through ChatGPT's plugins. With no analytics or search platform named in the documentation as of Sept. 30, that remains the gating factor.

What to watch next

OpenAI says users will later be able to add more dots and scale each one's output, covering speed and monthly workload, and the Enterprise release remains in beta. The signal to monitor is which analytics and SEO tools a dot can access through ChatGPT's plugin ecosystem — especially since proactive research only pulls from permitted connected sources.