Most people treat a good prompt like a good bottle. They use it once, it's great, and then it sits on the shelf until they remember it exists.
The prompt that caught a client about to churn in August? Never ran again. The one that turned your call notes into a tidy follow up list? Buried in a chat from three weeks ago, somewhere between a recipe question and a draft of a hard email.
Scheduling fixes that, and this fall almost every tool you already pay for added a timer.
On October 5, Perplexity added Automations to Perplexity Computer. They run on a schedule or when something happens in a connected app, and they pause for review before anything consequential. ChatGPT rebuilt its scheduled tasks back in June with a dedicated Scheduled page in the sidebar for Go, Plus, Pro, Business and Enterprise users. Littlebird shipped an October 23 update with more flexible schedules for its Routines. And Rize rolled out an October 20 overhaul with Routines that run its agent on a set schedule, plus a Meetings page in beta that imports transcripts from Fathom.
So you've got the timer. What you probably don't have is a short list of prompts worth putting on it, or a few rules that stop the scheduled version from slowly turning into a confident liar.
Where each kind of job should live
Before the prompts, a quick map. Where a prompt runs changes what it can see.
ChatGPT Scheduled. Best for anything that's mostly web research or thinking over things you've already told it. Open Scheduled in the sidebar, write the request, pick the time and how often it repeats.
Perplexity Automations. Use this when the job is watching the outside world and checking sources. Perplexity says they can build on earlier runs, which is exactly what you want for a "what changed since last week" report. The review pause is a nice touch for anything that might take action.
Littlebird Routines. Best when the answer lives in your own stuff: email, calendar, docs, meetings. Reports come to your inbox with the full write up in the body, and you can reply to the email to keep the conversation going.
Rize Routines. The pick for anything about where your time actually went. Rize already knows which client and project each hour belonged to, and with Fathom connected it can pull the call transcript in too.
One more option deserves a mention. The Gemini app got a Tasks toggle on October 13 that lets it keep working while you're away, for AI Pro and Ultra subscribers. Google hasn't said anything about recurring schedules for it, so I use it for long one off jobs, not the weekly stuff below.
The four rules for any prompt on a timer
A prompt you run by hand has a babysitter. You read the answer, you notice when it's off, you push back. A scheduled prompt has nobody watching at 6am on a Monday. These four rules go into every one of the seven prompts below, and into anything you schedule after that.
Scope the dates every time. "Recent" means nothing to a model. Write "the last 7 days, ending yesterday" or "since the first of last month." Otherwise you get the same news story three Mondays in a row.
Give it a way out. Every scheduled prompt needs a line that says if there's nothing worth reporting, reply with one sentence saying so. Without it, the model will find something to report, because you asked for a report. That's how you end up "monitoring" a competitor price change that never happened.
Make it show receipts. Every claim gets a source: a link, an email subject line, a date, a file name. If it can't point to where it found something, it doesn't go in the report.
Set a kill date. Put a review date on every scheduled prompt, 60 or 90 days out. On that date you ask one question: did I act on this report at least twice? If not, delete it. Scheduled prompts pile up like browser tabs, and an inbox full of AI reports you skip just trains you to skip them all.
Here are the seven, one for each slot in the week plus two monthly ones.
Monday: the competitor and pricing watch
Where it lives: Perplexity Automations or ChatGPT Scheduled. When: Monday, 6am.
You are my market analyst for [BUSINESS NAME], a [WHAT YOU SELL] business serving [CUSTOMER TYPE] in [AREA]. Check these competitors: [COMPETITOR 1 + URL], [COMPETITOR 2 + URL], [COMPETITOR 3 + URL]. Look only at changes published or visible in the last 7 days, ending yesterday. Report on: price changes, new offers or promotions, new services, new locations, hiring posts, and anything in their reviews that suggests a weakness. For every item, give the source link and the date you found it. Keep it under 200 words. Put the one change most likely to cost me a customer at the top. If nothing meaningful changed, reply only with: "No meaningful competitor changes this week." |
Good output: three to five lines, a link on each, one clear "watch this" at the top. If you get a 1,200 word essay about their brand positioning, tighten the word limit and run it again.
Tuesday: the review and reputation digest
Where it lives: Perplexity Automations or ChatGPT Scheduled. When: Tuesday, 7am.
Find every new public review of [BUSINESS NAME] at [ADDRESS OR WEBSITE] posted in the last 7 days, ending yesterday, on Google, Yelp, Facebook and [INDUSTRY SITE]. For each review give: platform, star rating, date, the reviewer's first name, a one line summary, and the link. Then flag any review under 4 stars and draft a reply for it under 60 words. The reply should thank them, name the specific issue, and give a next step. No corporate apology language. Finally, list any complaint that shows up in two or more reviews this month. If there are no new reviews, reply only with: "No new reviews this week." |
Good output: a short table and maybe one draft reply. The line I care about is the repeated complaint at the bottom. Anybody can have a bad day, but if the same complaint shows up twice in a month, something in how you work is broken.
Don't let this one post replies on its own. Draft only. A human hits send on anything the public can see.
Wednesday: the stale lead sweep
Where it lives: ChatGPT Scheduled or Littlebird, fed by an export. When: Wednesday, 8am.
This one needs data. If your leads live in Go High Level, export opportunities with their stage, last activity date and owner to a Google Sheet. Then a small Make.com scenario can refresh that sheet every Tuesday night so the prompt always reads current numbers.
Read the sheet [SHEET NAME] in my connected Google Drive. Find every open opportunity where the last activity date is more than [10] days before today and the stage is not Won or Lost. Sort by deal value, highest first. Show the top 10 only. For each, give: name, stage, value, days since last touch, owner, and a two sentence follow up message the owner can send today. The message should reference the last known conversation point if the sheet has notes, and ask one simple question. Do not invent any detail that isn't in the sheet. If a field is blank, say "unknown." If no opportunities qualify, reply only with: "Pipeline is clean this week." |
Good output: ten names, ten short messages, done in about four minutes over coffee. The "do not invent" line matters here more than anywhere, because a follow up that mentions a conversation that never happened is a fast way to look like a bot.
FROM THE AI NEWSROOM The AI Workflow Blueprint The exact systems behind everything in this issue. The audit sheets, the routing logic, the templates and the review cadences, built out step by step so you can copy them straight into your own stack. One time, forty seven dollars. Get the Blueprint for $47 |
Thursday: the cash and receivables nudge list
Where it lives: Littlebird Routine, or ChatGPT Scheduled with your accounting export. When: Thursday, 7am.
Thursday is on purpose. A Thursday morning nudge lands while people are still at their desks and gives them a day to pay before the weekend, and it leaves you Friday to follow up by phone.
Using my connected email and the latest accounts receivable export in [FOLDER], list every unpaid invoice that is past due or due within the next 7 days. For each: client name, invoice number, amount, due date, days overdue, and the date of the last email between us about payment. Group them: Due this week, 1 to 30 days late, 31+ days late. For the first two groups, draft a short, friendly reminder under 70 words that includes the invoice number and amount and a payment link placeholder [LINK]. For the 31+ group, don't draft an email. Write one line suggesting whether I should call, and why, based on our recent emails. End with the total dollars outstanding in each group. If nothing is due or overdue, reply only with: "Receivables are clean." |
Good output: three groups, three totals, and a handful of drafts you can send in five minutes. The 31+ group gets a phone call. By day 31 they've already ignored the polite emails.
Friday: what did I promise this week
Where it lives: Littlebird Routine, or Rize Routine with Fathom connected. When: Friday, 2pm.
If I could only keep one of the seven, it'd be this one. A forgotten promise is the kind of thing clients don't mention. They just start shopping around. Monday's issue covered catching promises one email at a time. This is the weekly sweep that catches the ones you made out loud on calls.
Review my meeting transcripts from Fathom and every email I sent from Monday through today. List every commitment I made to someone else: anything where I said I'd send, check, call, introduce, fix, or follow up. For each: who I promised, what I promised, when I said I'd do it, the source (meeting name and date, or email subject line), and whether there's evidence I already did it. Put the open items with the nearest deadline at the top. Separately, list anything someone promised me that I haven't received yet. Don't include vague pleasantries like "let's grab coffee sometime" unless a date was named. If there are no open commitments, reply only with: "No open promises. Enjoy the weekend." |
Good output: a list of five to fifteen items, sorted by deadline, with receipts. I picked 2pm so you've still got three hours to close loops before the weekend instead of discovering them Monday.
The second list, what other people owe you, turns out to be just as useful. It's how you remember to chase the testimonial, the signed contract and the intro somebody offered on a call and forgot about.
First of the month: the time audit
Where it lives: Rize Routine. When: The 1st, 9am.
Look at my tracked time for last month, the 1st through the last day. Break it down by client and project, and compare each client's hours to what they pay me: [CLIENT A $ per month], [CLIENT B $ per month], [CLIENT C $ per month]. Calculate my effective hourly rate per client and rank them highest to lowest. Flag any client whose hours went up more than 20 percent compared to the previous month. Then list my three biggest categories of time that weren't billable to anyone, with total hours for each. End with one specific suggestion for next month, based only on these numbers. |
Good output: one ranked table, and probably a number you won't love. Almost every service business has a client paying the least and taking the most. You probably already suspect who. This makes it a number you can bring to the renewal conversation.
I left the escape hatch off this one on purpose. There's always time data. If the report comes back thin, the problem is that tracking was off, and Rize's October 27 update now tells you when tracking has stopped, so turn that alert on.
First of the month: the vendor price and terms check
Where it lives: Perplexity Automations or ChatGPT Scheduled. When: The 1st, 10am.
Every software tool you use changed something this year. Credits, seat prices, which plan includes which feature, what they do with your data. This prompt catches those changes before the invoice does.
Here is my software stack: [TOOL 1, plan], [TOOL 2, plan], [TOOL 3, plan], [TOOL 4, plan]. For each tool, check its official pricing page, changelog, blog and terms of service for changes published in the last 31 days, ending yesterday. Report only: price changes, plan or feature changes that affect my plan, new usage limits or credit systems, and changes to data use or AI training terms. For each change, give the effective date, what it means for my plan in one sentence, and the source link. Put anything with an effective date in the next 30 days at the top. If nothing changed for a tool, leave it out. If nothing changed at all, reply only with: "No vendor changes this month." |
Good output: usually two or three items, each with a date and a link. The effective date sort is the important bit. If a price change lands in three weeks, you can still do something about it. If you find out after it hits your card, you can't.
Set it up in one sitting
Don't build all seven tonight. Here's the order I'd go in.
Start with Friday. The promises prompt pays off the first week, and it's the one you'll notice missing if you turn it off.
Add Wednesday next. Stale leads are deals you already paid to get. Get the Go High Level export working first, then schedule the prompt.
Run each prompt by hand twice before scheduling it. If the manual output isn't useful, the scheduled one won't be either. Fix the prompt while you can still watch it.
Make a "Timers" note in whatever you use for notes. One line per scheduled prompt: name, tool, schedule, start date, kill date. Thirty seconds now saves you a confused afternoon in February wondering what's emailing you every Tuesday.
Put the kill dates on your calendar. Ninety days out, a 15 minute block called "Timer review." Keep the ones you've acted on, kill the rest without guilt.
A note on cost. Scheduled runs use the same allowance as your normal usage on most of these tools. Seven prompts a week won't break anything, but if you start scheduling hourly checks, keep an eye on your usage page for the first couple of weeks.
FROM THE AI NEWSROOM The AI Business Accelerator For operators who would rather build it with someone than build it alone. The full stack, installed alongside you, with the decisions made in the room instead of guessed at. Ninety seven dollars. Join the Accelerator for $97 |
This week
Pick the Friday promises prompt and run it by hand today against this week's emails and calls.
Schedule it for Friday at 2pm in Littlebird, or in Rize if your meetings already flow in from Fathom.
Export your open opportunities and run the stale lead sweep once by hand before you automate it.
Add the escape hatch line and a date scope to any prompt you already have scheduled.
Start your Timers note and put the first kill date on the calendar 90 days from today.
Jordan
The AI Newsroom | Practical AI for people with a business to run.