Write Your Own ‘Act For Me’ Prompts.

← Back to Issue #9

Go Deeper Guide  •  Companion to Issue #9

Write Your Own “Act For Me” Prompts

You already tried the contractor-comparison prompt from this issue. That part was really a search — the kind a plain AI chat could mostly do too. This guide shows you the part that actually needs a tool with “agent” abilities, and how both prompts are built, so you can write your own version for any task.

Why This Is a Good First Task

Finding a plumber, roofer, or landscaper usually means the same tedious routine: search Yelp, cross-check Google reviews, maybe glance at the Better Business Bureau, then try to remember which one actually had the best rating. It's exactly the kind of task that's easy for a person to lose track of — and straightforward for a computer that's good at reading several pages and organizing what it finds.

That makes it a good first task to hand to an AI that can act, not just answer. And unlike tasks that need you to be signed into a personal account, this one runs entirely on public information — nothing to log into, nothing to configure.

Before You Start — Three Things to Know

• It's reading public reviews, not making the call for you. Claude can gather ratings, review counts, and contact info from public sites. It won't message anyone, request a quote, or book an appointment unless you separately ask it to.
• Review counts matter as much as star ratings. A perfect 5.0 with 3 reviews means less than a 4.6 with 200 reviews — ask Claude to show both, and weigh them yourself.
• It can't verify licensing or insurance. For anything structural or safety-related — electrical, roofing, plumbing — always confirm a contractor's license and insurance yourself before hiring. That's not something a review site or an AI summary can substitute for.

Step 1: Get Your Shortlist

Research step

This is the prompt from this issue. It's a research step — Claude searches a couple of public sites and organizes what it finds.

Reminder on getting there: open Claude in Chrome, click its icon ✧ in your toolbar to open the side panel (pin it via the puzzle-piece icon if you don't see it), then paste the prompt into the chat box. See Issue #9 if you need the full walkthrough.

“Search Yelp and Google for plumbers near [your city or zip code]. List the top 5 with their star rating, number of reviews, and phone number. Don't contact anyone — just show me the list.”

Worth knowing: a plain AI chat with web search — Claude.ai, ChatGPT, whatever you use — can probably do a decent version of this step too, since it's really just search-and-summarize. Nothing wrong with that. It's step 2 where things change.

Step 2: Let Claude Take the Next Step

Where agent tools take over

This is the part a plain chat-only AI can't do at all — it takes real typing into a real page, not just generating text for you to copy. It takes a tool with “agent” abilities to pull off. Claude in Chrome works this way, and so does Claude Cowork and ChatGPT's Agent mode — this is a capability spreading across AI tools generally, not one company's trick. This guide walks through it using Claude in Chrome since that's the tool and prompt we've tested, but the idea works wherever you see “agent mode” or similar.

“Open [top pick]'s website and fill out their contact or quote-request form. Use my name, [your phone number], and describe the job as: ‘[a quick description, e.g. leaky kitchen faucet, want it fixed this week].’ Show me exactly what you've filled in before you submit anything.”

Notice the instruction to show you before it submits. Claude in Chrome is built to pause on anything consequential — submitting a form, sending a message, downloading a file — and check it against what you originally asked for. Building that pause into your own prompt is good practice anyway: it means you get a final look before anything leaves your hands.

The Anatomy of a Good Prompt

Every good “act for me” prompt has three parts. Here's the Step 2 prompt broken down — once you can spot these three parts, you can build a version of this for almost any task.

1. The Scope — where to work, and on what
   “Open [top pick]'s website and fill out their contact or quote-request form.”
2. The Ask — exactly what to fill in
   “Use my name, [your phone number], and describe the job as: [a quick description].”
3. The Boundary — what it's not allowed to do
   “Show me exactly what you've filled in before you submit anything.”

Scope, Ask, Boundary. That's the formula, whether the task is a search or a real action. Leave any one of the three out and you'll either get a vague result, an incomplete one, or — worse — Claude submitting something before you've had a chance to check it.

Try the Formula on Something Else

Here's the same three-part formula, built for two other everyday tasks.

Comparing prices on a big purchase

“Search [Retailer A], [Retailer B], and [Retailer C] for a [item, e.g. ‘countertop microwave, roughly 1.2 cubic feet’]. For each site, list the top 3 matches with price, rating, and shipping cost. Don't buy anything — just show me the comparison.”

Planning a trip

“Look up flights from [your airport] to [destination] for the week of [dates]. List the three cheapest nonstop or one-stop options with airline, price, and total travel time. Don't book anything — just show me the options.”

Notice the shape is identical every time: where to look (Scope), what to hand back (Ask), and what not to do yet (Boundary). Swap in your own task and you've written your own prompt.

Like the plumber search, both of these are still research prompts — a plain AI chat could take a decent swing at them too. To get to a true Step 2 with either one, extend it into a real action: have Claude actually add the flight to your calendar as a hold, or fill out the “notify me” form on the retailer's site for that microwave. That's the same jump from Answering to Acting, applied to a new task.

One more habit worth building: after you get a result, ask a smart follow-up — like “which of these are closest to me” or “which have any complaints about no-shows?” That follow-up is often where the real value shows up, because it connects the list to something only you know or care about.

What to Expect: Example Output

Here's the kind of list Claude in Chrome should hand back for the contractor prompt from this issue:

Business	Rating	Reviews / Phone
ABC Plumbing	4.8 stars	212 reviews — (555) 123-4567
Quick Fix Plumbers	4.6 stars	89 reviews — (555) 234-5678
Neighborhood Plumbing Co.	4.9 stars	34 reviews — (555) 345-6789

(These are illustrative examples, not real businesses — your list will look different.)

What AI Still Can't Do Here

• It can't verify licensing, insurance, or bonding — always confirm that yourself before hiring, especially for anything structural or safety-related.
• It can miss businesses with little online review presence, even good ones — word-of-mouth recommendations still matter.
• It won't negotiate price or book anything for you.
• Treat its list as a strong starting point, not the final word.

Your Turn — Try It This Week

• Run the Step 1 search prompt for a service you actually need right now.

• Try Step 2: have Claude fill out a real quote-request form — and check what it filled in before you let it submit.

• Pick one other task that's been bugging you, and write your own Scope / Ask / Boundary prompt for it — for the research half and the action half.

• Before hiring anyone, confirm their license and insurance yourself.

One Last Thing

Small tasks like this are the best way to build real trust in what AI can and can't do. Once you've watched it handle something small and useful correctly, you'll have a much better sense of what to hand it next — and what to still do yourself.

Talk soon,
Gary
Founder, AI Unlocked

← Back to Issue #9

AI Unlocked  •  aiunlocked.us  •  Your trusted guide on the AI journey

Helping adults 55+ navigate AI with confidence — no hype, no jargon.