The short answer.

To get a better result from a single AI request, write four asks into it: time back (do the routine part in full), fewer misses (check the result against a standard you name), better decisions (set out the options and objections for the choice you make), and work you could not staff (add what a role you do not employ would bring). A request that carries all four gets a finished draft, a checked one, and the thinking behind it, in one reply. Most people write only the first ask, and get back only a draft to rewrite.

The four asks are how FullyPowered AI thinks about a single request. They belong to the first of four stages a piece of work goes through with AI, which we call Ask, Save, Share, Sustain. They work in whichever assistant your company already pays for: ChatGPT, Claude, Copilot or Gemini.

The four asks, one at a time.

Each of the four asks names something you want from the reply. Write each one as its own line, in plain words, with the details only you know.

  1. Time back.Hand over the whole routine part of the job, not a fragment of it. Say what the finished piece is, who reads it and what goes in. "Write the renewal letter to this client, from these notes, in the format below" gives time back. "Help me with a letter" gives a starting point you then rewrite.
  2. Fewer misses.Ask the assistant to check its own result against the standard a careful colleague would apply, and to list what is missing or does not match. Name the standard: the checklist, the policy wording, last year's figures, the client's own request. A busy person skips this check. The assistant does not get tired of running it.
  3. Better decisions.Ask for the options, what each one costs you, and the objections the other side will raise. You still make the decision. The assistant does the preparation before the meeting, so the meeting spends its time on the choice instead of on finding the facts.
  4. Work you could not staff.Ask the assistant to add what a role you do not have would bring: the view of an editor, a finance lead, a pricing specialist, a second reader from the client's side. A company of 30 people has no in-house editor. The request can still ask for an editor's pass.

The order matters little inside the request. What matters is that each ask is there, written plainly, so the reply comes back complete instead of as a first draft that starts a long conversation.

One request, before and after.

Take a 35-person architecture practice. A project lead prepares a fee proposal for a returning client whose project scope has grown. She opens the company's assistant and types the request she has typed many times before.

Illustrative example.

  • Before."Write a fee proposal for this client's new project." She pastes the brief. The reply is a tidy page that repeats the brief. She rewrites it, checks the scope by hand against the last appointment, and sends it with the client's likely questions in her head only.
  • After."Write a two-page fee proposal for the client's finance director, from the brief below: what has grown in the scope since the last project, what that adds to our services, and what it means for them. Check every stage and deliverable against our last appointment with this client, pasted below, and list anything that does not match or is missing. Set out three ways we can structure the fee, with the objection the client is most likely to raise to each. Then read the proposal as the client's finance director would, and tell me which line will draw a question."

The second request carries all four asks. The first line asks for time back on the whole proposal. The second asks for fewer misses, against a named standard. The third prepares her decision on how to structure the fee. The fourth brings in a reader the practice does not employ: the client's own finance director. She still reads every line, corrects what is wrong and chooses which structure to offer. The assistant prepared; she decided.

A four-line template to copy.

Copy these four lines into your next request and fill in the brackets. Paste the material the request needs below them.

  1. Time back.Write [the finished piece] for [who reads it], from [the material below], covering [what must be in it].
  2. Fewer misses.Check it against [the standard: a checklist, a policy, last period's figures, the original request] and list anything missing or not matching.
  3. Better decisions.Set out [two or three] options for [the decision I make], with what each one costs and the strongest objection to each.
  4. Work you could not staff.Then read it as [a role we do not have] would, and tell me what they would change.

The brackets are the part only you know. A request that leaves a bracket vague gets a vague line in the reply, so the time to spend is on the brackets, not on the wording around them.

When to leave an ask out.

Leave an ask out when the piece of work has no use for it. A quick reply to a colleague needs time back and nothing else. A first look at an idea needs the options and objections, and no check against a standard yet. The four asks are a list to run down before you send a request, not four lines every request must carry.

Two asks are worth keeping in almost every request. The check for misses, because the assistant's first reply is a draft a person reviews, and a named standard makes that review faster. And the role you do not have, because it is the ask people forget most and the one that brings in work nobody in the company had time to do. The five ways AI helps with a piece of work, from gathering to repeating, are set out in The five ways AI helps with a piece of work; the four asks are how you phrase any of them in one request.

What is different in a company of under 200 people.

In a company of under 200 people, the fourth ask does the most. A large company has an editor, an analyst and a pricing team down the corridor. A company of 40 has the owner and whoever is free. The request that asks for the view of a role you do not employ is how a small company gets that second reader on every piece that matters.

  • One person teaches the rest.The person who writes the best requests can show the whole office the four lines in one meeting. Nobody needs a course.
  • The standard already exists.The checklist, the house style and last year's figures sit in a shared folder. Pointing the request at them is the whole of the second ask.
  • The good requests are easy to spot.In a small team you hear which request worked. That is the moment to keep it.

Where each person stands with their assistant is set out in The levels of using an AI assistant at work, and where a whole company stands in The stages of AI at work. The four asks lift the first level for everyone, quickly.

Where the larger gain starts.

The four asks make one request better every time you type it, and the larger gain starts at Save, when the request stops being retyped. A request that worked well once, kept and run the same way the next time, becomes a workflow: the same four asks, the same standard, the same checks, whoever is at the desk. Ask, Save, Share, Sustain follows a piece of work up that ladder.

Deciding which requests to keep, for which teams and roles, is the plan. How organizations plan their use of AI sets out the method, and FullyPowered AI Planner is where the plan lives: you set out your teams, their roles and the work between them, and choose which workflow to build first.