ChatGPT Pages: build a self-updating engineering status page

ChatGPT Pages is a new document type for people and agents to write, research and visualize together, launched at DevDay on 29 September 2026 for Pro, Business and Enterprise. The ChatGPT Space launch page says pages can stay updated from connected tools and take instructions on what to update, which sources to use and when to check. The Space agents docs add that the Keep Updated option is not available at launch.

This guide builds one engineering status page with release notes and blockers, and shows how to keep it current today.

What can ChatGPT Pages do at launch?

A page is built from blocks. Type / while editing to insert headings, checklists, tables, code blocks, files, links to chats, visualizations, prompts, tasks and agent instructions. Child pages come from the same menu.

You can also start a page from an existing conversation and invite your team afterwards.

Are self-updating pages in ChatGPT available yet?

Partly. The Pages docs list agent instructions as a block type, but Keep Updated is not live yet. The agents docs also warn that writing a cadence in a page or its instructions does not by itself schedule a task.

So a line such as "update every Monday" in your agent instructions does nothing on its own. To get recurring updates, ask ChatGPT in chat to create a scheduled task, then review the saved schedule. The docs tell you to check an actual run and the resulting page update, because these controls depend on your account.

Build the status page

Create a page called "Payments service status" or similar. Add child pages or sections for release notes, blockers and open pull requests. Then insert an agent instructions block from the / menu and write plain rules. This is our example, not an OpenAI template:

Keep this page accurate for the payments service.
Release notes: list merged pull requests since the last update, one line each, with a link.
Blockers: list items labelled "blocked" on the board, with owner and date.
Never delete a human-written line. Propose edits to decisions instead of changing them.
If a source is unavailable, say so at the top of the page.

Keep the rules short. The page instructions tell an agent what good looks like. The scheduled task decides when it runs.

Keep the page current with a scheduled task

The scheduled tasks docs cover the part that actually runs. You can create and update a task from a ChatGPT or Codex chat. Tasks show up under Scheduled in the sidebar with their recent runs.

1. Test the prompt manually in a regular chat first. Check that it reads the right sources and edits the right page sections.
2. Connect GitHub and make sure the connected app can access the repository.
3. Ask ChatGPT to create a task triggered by pull request activity. Filter on the repository and choose merges only.
4. For blockers from Slack, add @ChatGPT to each watched channel. Slack triggers cover new messages, not reactions, edits or direct messages.
5. Choose Run now, open the page and compare the update with the source.

Some limits matter. Event-triggered tasks work in ChatGPT on the web and mobile, not in the desktop app, Codex CLI or IDE extension. One task can use several event triggers, but it cannot mix them with a time schedule. If you also want a Monday summary, create a second task with a schedule. Custom schedules accept an RFC 5545 RRULE such as RRULE:FREQ=MONTHLY;BYMONTHDAY=1;BYHOUR=9;BYMINUTE=0.

In managed workspaces, admins control event triggers with the "Allow event-triggered scheduled tasks" permission. Scheduled tasks run unattended with your default sandbox settings, so start with the narrowest access that works.

Where Codex fits in this workflow

Codex CLI does not provide the Scheduled management interface. It can help you prepare and test the prompt, skill or script before you schedule it on the web. That split works well: write a skill that formats release notes from merged pull requests, test it in the terminal against your repo, then invoke it by name in the task prompt.

Put the release note format in the repo's AGENTS.md as well. Codex then writes pull request descriptions in the shape the task summarizes, and the page stays consistent.

The docs give a related example: a skill that babysits a pull request can set up a scheduled task that checks PR status with the GitHub plugin. The same pattern keeps a status page honest. The skill defines the output, and the task decides when it runs.

For review, use the page's comments. Mention ChatGPT on a wrong line and ask it to check the source. Use Show attribution to see which lines came from people.

What to check after the first week

Compare five release note lines with the merged pull requests they claim to describe. Check that no human-written blocker disappeared. Confirm that the task history under Scheduled matches the updates you see on the page.

If your team wants to build agent-maintained pages like this on its own repositories, our Codex training covers skills, scheduled work and review habits.

Next step: write the agent instructions block for one service and test the prompt in a regular chat before you schedule anything.

Further reading

• Pages in ChatGPT Space, OpenAI docs
• Work with agents in Space, OpenAI docs
• Scheduled tasks, OpenAI docs
• DevDay 2026 recap, OpenAI