Foggetti Studio IT & StoreSoftware · AI · Automazioni
ChatGPT prompt guide for entrepreneurs and managers

Blog

AI & AutomationCornerstone guide

ChatGPT prompt guide for entrepreneurs and managers

Clear guide to ChatGPT at work: allowed activities, rules, instructions and a two-week pilot.

10 minute read

by Giuseppe Foggetti
Software Engineer & AI Solutions Developer — Foggetti Studio

Tuesday morning in a services SME. The salesperson just finished a site visit and needs to send the follow-up. They open ChatGPT on a personal phone, paste the notes as written — names, addresses, a few figures — and ask for «a professional email». In two minutes they have a decent draft. Nobody in the company knows that text left through a private account, or which data ended up in the chat.

It happens more often than people admit. Not because staff are reckless on purpose: because the tool is useful and nobody has said where yes and where no. The message, without spinning: ChatGPT can save time, but employees must not use it freely with confidential data. The company must decide which activities to use it for, provide ready-made instructions, have a person check the results, and verify that time savings are real.

Why DIY does not hold for long

When everyone sets their own rules, three problems arrive together. First: you no longer know which activities the team already uses the model for. Second: without shared instructions, the same customer gets different tone and detail depending on who writes. Third: price lists, margins, staff data or entire contracts land in non-company chats, with no useful trail.

In practice the company gains scattered hours and loses control. That is not an argument against AI. It is an argument against improvisation. A small test, with a clear owner and human review before send, beats twenty personal accounts left free.

Start from activities, not from the tool

The solid way to introduce ChatGPT is simple and often skipped: list the repetitive tasks that waste time. Follow-up emails after a visit. Meeting summaries. Pre-job checklists. First read of a support request. For each one write, in plain language, what goes in, what must come out, who re-reads it and what happens if the result is wrong.

Approve only low-risk activities. Internal drafts and summaries are fine. Negotiated prices, contractual commitments and decisions that bind the company stay out — or must pass an expert. Three to five allowed activities (and a clear forbidden list) beat any generic AI course.

One-page rules, not a fifty-page policy

You need a rule people will actually read. One page is enough. Write what must not be pasted: unnecessary personal data, customer codes, confidential attachments, non-public price lists. Write what is allowed: anonymised text, generic drafts, rephrasing. Name official company accounts and end the season of personal logins on work PCs.

Add one line on accountability: if a message goes to the customer, a named person checked it. That is not bureaucracy. It is the simplest way to stop «ChatGPT said so» becoming an excuse when something goes wrong.

Ready-made instructions (the prompt library)

Without shared instructions, every colleague reinvents the world. Prepare eight to twelve ready prompts in a sheet or Notion. Each card says what to obtain, which minimum inputs to give, the limits («professional tone, max 120 words»), the output format and a good example next to a bad one.

Keep them current. A prompt that works today and is written nowhere becomes personal knowledge within a week. Name an owner, mark the current version and spend thirty minutes a week fixing three real cases. That beats an eight-hour course done once.

A two-week trial that tells the truth

Pick one flow — for example the post-visit email draft — and three to five people. They use only approved instructions. Every text going to the customer passes a person. Log times, corrections and moments when the model invents details nobody put in the notes.

If time really drops and errors stay few, add a second activity. If corrections are too many, improve instructions and inputs — do not buy another tool. Note what must stay manual. Only when instructions are stable does it make sense to connect drafts to email or CRM notes — and in month one, no automatic send to the customer.

An example that comes up often

A services SME with about a dozen people was losing six to eight hours a week on follow-ups and visit summaries. They limited ChatGPT to three activities: follow-up draft, note summary, pre-job checklist. Clear rule: no full price lists or complete customer records in prompts. Ten library instructions, mandatory check by the owner or senior salesperson. After three weeks writing time was down about a third, with zero wrong sends — precisely because a person always re-read before send.

What to leave out (and why)

Resist «make AI do everything». If the error costs reputation or money, human control is not optional. Legal or medical advice, sensitive staff data, contractual commitments: out of month one, full stop. Expanding later with numbers in hand is healthy. Expanding on enthusiasm is like pouring fuel on an unstable process.

From instructions to everyday work

When the library is stable, you can bring ChatGPT closer to the message templates the team already uses. The goal is not «more AI»: it is less effort between draft and send, always with approval on external messages. Keep measuring time and corrections. If both rise, you sped up disorder: step back and clean inputs and rules.

A light weekly ritual keeps improvement alive: three real outputs on the table, what worked, what to fix in the instructions. That is how ChatGPT use becomes team competence, not one person's talent.

Training that sticks

Short sessions on the team's real cases. A checklist within reach. A super-user per area for the first fifteen days. Avoid catalogue courses that end in a certificate and zero change the following Tuesday.

How you know you are improving

Do not multiply dashboards. Pick a few indicators tied to the process you are touching — on ChatGPT prompt guide for entrepreneurs and, cycle time, output quality and adoption of the official flow usually suffice. Set baseline in week zero. Review at thirty days. If numbers do not move, change process and inputs before you change the tool.

After the pilot: expand without losing the thread

Widen only if the primary KPI improved, adoption beats the agreed threshold, the exception backlog is under control and the owner is still in charge. Otherwise reduce scope or strengthen training. A meeting every two weeks is enough: numbers, top exceptions, decisions. No status theatre.

Document the mapping or happy-path rules on a living page. When someone new joins, that page saves three weeks of oral tradition. Light governance is not bureaucracy: it is operating memory.

Vendors and boundaries

If an external vendor enters, clarify who owns configurations, prompts, mappings and logs. Avoid opaque dependencies. A good partner leaves the company more autonomous at ninety days, not more tied to endless tickets to change one rule.

The right tone toward the customer

Even with excellent prompts, the model can sound generic or too confident. Human review is not only error control: it is care for the relationship. If you would not sign a sentence, it does not leave. Over time prompts refine on these edge cases — and that is where the library becomes company asset.

A detail that makes the difference

Block calendar time for the pilot. Without dedicated slots the project stays «between other things» and never really starts. The operating owner updates status every week in five minutes: done, blocked, next step. It is not elegant. It works.

In short: the operating framework

Short version to print or paste into a meeting.

Steps

  1. Choose 3–5 repetitive activities + a forbidden list (day 1–2).
  2. Write 1-page rules: forbidden/allowed data, accounts, who checks.
  3. Prepare 8–12 ready instructions (goal, limits, example).
  4. Trial 2 weeks on one flow, 3–5 people, mandatory human check.
  5. Measure, fix instructions/inputs, expand only if savings are real.

Quick checklist

  • Allowed and forbidden activities communicated to the team.
  • Company accounts active (no personal logins at work).
  • Instruction owner named.
  • A person checks every outbound customer text.
  • Week-zero times logged (for comparison).

KPIs (max three)

  • Average time to write a draft (minutes, before/after).
  • How often a person corrects the output.
  • Data incidents (prompts with forbidden info): target zero.

Practical value (and the next step)

To recap: ChatGPT can save time, but employees must not use it freely with confidential data. The company must decide which activities to use it for, provide ready-made instructions, have a person check the results, and verify that time savings are real. Practical value is not «having AI»: it is fewer hours on repetitive drafts, fewer mistakes toward the customer, and less risk of sensitive data leaving through a personal account.

This week, in half an hour: write three to five allowed activities and what is forbidden, name who checks outbound messages, and pick one flow to trial for two weeks. Bring the schema to the meeting and treat it as an operating checklist.

If you want to set rules, instructions and a pilot on your real workflow, Foggetti Studio can help starting from your team's actual process.

ChatGPT prompt guide for entrepreneurs and managers | Foggetti Studio