The H.E.L.P. Framework for AI Prompting

Abstract four-step geometric graphic representing the H.E.L.P. framework for working with AI, from Nantucket AI

Most people who tried an AI assistant and walked away disappointed did not use a bad tool. They used a vague ask. They typed one sentence, got back something generic, and decided the whole category was oversold.

The H.E.L.P. framework for AI prompting exists to close that gap. Four steps, in order: Humanize, Evaluate, Leverage, Perfect. It is the method Nantucket AI teaches island businesses, and it works in whatever chat assistant you already have open.

Why a framework beats improvising

When you improvise a prompt, you write the request the way you would say it out loud to someone who already knows your business. The assistant does not know your business. It fills the gaps with the most average version of your industry it can produce, and average is exactly what you cannot use.

A framework fixes three things at once.

It makes you write down context you have never written down before. It forces a review step you would otherwise skip, because the draft looks finished and finished-looking work is hard to argue with. And it makes the prompt reusable, so the second time you do this task it costs ten minutes instead of an hour.

That last part is where the real return sits. One good prompt, saved, used forty times, beats forty clever improvised ones.

H is for Humanize

Humanize means giving the assistant the context, role, audience, and voice it does not have.

Context is your situation. Role is who it should write as. Audience is who reads the result. Voice is how your business sounds.

Here is the difference. A weak ask sounds like this: “Write a description for a rental cottage.”

A humanized ask sounds like this: “You are writing for a family that manages three rental cottages in Siasconset. The audience is a couple in their sixties booking a two week stay in September, not a summer family with kids. The tone is plain and warm, no luxury language. The cottage sleeps four, has an outdoor shower, is a nine minute walk to the beach, and has no air conditioning. Say the no air conditioning part plainly, do not hide it.”

Same task. The second one can only produce something close to usable, because you removed the room the assistant had to guess in.

E is for Evaluate

Evaluate is where you judge the output against what you actually know. This is the step people skip, and skipping it is what makes AI dangerous rather than merely mediocre.

You are checking four things. Is anything here factually wrong. Is anything here invented. What is missing that a reader would need. And where would this embarrass me if a customer read it closely.

An example. A nonprofit asks for a paragraph about its work for a grant application. The draft comes back reading well, and it includes a line about serving “over 400 island families last year.” Nobody gave the assistant that number. It produced a plausible one because the sentence needed a number in it.

If that goes into a grant application, you have a real problem, and the tool will not be the one answering for it. Evaluate catches it. Then you turn the finding into the next instruction: “Remove every statistic. I will supply the real figures. Leave a bracketed placeholder where each one belongs.”

Note that the E used to stand for Expand in older Nantucket AI material. It is Evaluate now. Expanding a draft is the second half of the step. The judgment comes first. If you ask for more without judging what you already have, you get a longer draft with the same problems inside it.

L is for Leverage

Leverage is getting the work out of the chat window and into the business, and then taking more from the conversation you already loaded.

Most people stop at one output. That is the waste. You spent real effort telling the assistant about your restaurant, your season, your regulars, and your menu. That context is sitting there and it is free to reuse.

A restaurant owner writes one good prompt for the fall menu announcement. Before closing the tab, she asks the same conversation for the short version for Instagram, the version for the email list, the version for the front door sign, three answers to the question staff will get about why the lobster roll left the menu, and a two sentence note for the reservation platform.

Six deliverables from one loaded conversation. The second through sixth cost almost nothing, because the expensive part was the setup.

P is for Perfect

Perfect has two halves. Make the work recognizably yours, then save the prompt.

The first half is editing. Every assistant has tics. It over-explains, it reaches for the same three transitions, it adds a closing sentence that summarizes what it just said. Cut those. Put back the one detail only you would have included.

The second half is the part almost nobody does. Take the prompt that finally worked, strip out the specifics, and save it as a template with blanks.

A property manager who nails the monthly owner update writes the working prompt into a document with [MONTH], [PROPERTY], and [ISSUES THIS MONTH] marked as fill-ins. Next month the job is a fill-in-the-blank exercise. So is the month after. And when she hires help in April, the new person produces the same quality on day one, because the standard is written down instead of living in her head.

Who this is for

H.E.L.P. is written for people who are not technical and do not intend to become technical. Inn keepers, restaurant owners, executive directors, office managers, real estate agents, retail staff, and anyone else whose actual job is something other than operating software.

It assumes no coding, no setup, and no math. If you can type a question into a chat box, you can run the method today.

It also assumes you stay in charge. Nothing here hands work over. You write the ask, you judge the answer, and you decide what ships with your name on it. That is the second rung of the trust ladder, the level called Assist, and most businesses should be very comfortable living there for a long time.

Where to go next

The canonical definition of each step lives on The H.E.L.P. Method page. Start there if you want the short reference.

If you want to practice it on your own work, The H.E.L.P. Framework course is free and takes about three hours. Five modules, sixteen lessons, and one exercise per step that you run on a real task from your own business rather than a made up one.

Bring one task you are dreading this week. Run it through the four steps. Keep the prompt.