Your recurring prompts, systematized

Foundations gave you the routine as a concept — numbered steps, each an ask, written down under its trigger. This lesson lands that concept in ChatGPT specifically: where routines live in this tool, how to make them start in one action instead of five, and how a working week actually feels when the recurring prompts run themselves.

Give each routine a home. Scattered chats kill routines — you re-explain context every Monday. Use the tool's grouping feature (projects or folders, depending on your plan) to give each routine a dedicated space: pinned instructions for that job, its files, and its history in one place. The weekly-review space knows it's a weekly review; you stop paying the setup tax per run.

Make step one a paste, not an essay. The best routines start with a fixed opener you keep in your promptbook: "Weekly review. Here's this week's material: [paste]. Run the usual: group, flag, table." Because the space carries the standing instructions, the opener stays three lines — the routine's intelligence lives in the setup, not in Monday's typing.

Chain with handoffs, not memory. Between steps, pass the artifact explicitly: "take the table from above and draft the three follow-up messages." Explicit handoffs keep long routine chats from drifting — and they make the routine portable, because each step names its input instead of trusting the conversation to remember.

Scheduled runs, if your plan has them. As of September 2026, ChatGPT can run some tasks on a schedule — a daily digest, a weekly reminder with a prepared prompt. Treat scheduled runs as invitations, not autopilot: the output lands, you give it the thirty-second look, you act. The routine discipline you built — written steps, explicit checks — is exactly what makes scheduled automation safe to accept.

Audit the system monthly. Same rhythm as the promptbook: which routine got skipped twice (kill it or shrink it), which step always needs a manual fix (fold the fix into the opener), which chore still isn't a routine but happened four times (that's next month's candidate). A routine system stays alive the way a promptbook does — by pruning, and by the same honest test: would you rebuild this today if it vanished?

One mistake to skip: building routine cathedrals. A five-space system with color-coded openers, created in one enthusiastic evening, dies by Thursday. Grow it the earned way: one routine, run weekly until it's boring, then the next. Boring is the goal — a routine that still feels exciting is a routine you're still debugging.