AI prompts for staff: a practical starter library

A simple structure for business prompts

Prompting is less about clever tricks and more about giving clear instructions, the way you would brief a capable new hire. Five parts cover most needs:

You do not need all five every time. A quick rewrite might need only the task and the text. But when output disappoints, the missing piece is usually the source material, the format or the constraints.

• Role and audience: who the output is for and what voice to use, for example "a friendly but professional reply to a residential customer."
• Task: one clear job, such as summarize, draft, rewrite, extract or check.
• Source: the text to work from, pasted in or attached, with an instruction to use only that material.
• Format: length, structure and layout, such as "five bullets" or "a table with columns for date, amount and payee."
• Constraints: what not to do, such as "do not include prices," "flag anything you are unsure of," or "do not invent details."

Prompts for front office and admin staff

These suit office managers, coordinators and front desk staff. Remove sensitive details first, in line with your AI policy.

Front office staff often get the quickest wins, because their work is full of short, repetitive writing tasks. Start them with two or three prompts rather than the whole list.

• Thread summary: "Summarize the email thread below in five bullets. List any open questions and who needs to answer each. Use only the thread. Thread: [paste]."
• Payment reminder: "Write a polite, short reminder to a client whose invoice is 15 days overdue. Keep it under 90 words, warm but clear, and do not threaten fees. Invoice details: [number, amount, due date]."
• Voicemail to call log: "Turn this voicemail transcript into a call log entry with caller name, callback number, reason for call and urgency. If any item is missing, write 'not stated.' Transcript: [paste]."
• Rescheduling notice: "Draft a text message under 300 characters telling a customer their appointment on [date] needs to move, offering [two options], and apologizing briefly."

Prompts for sales, estimating and account management

These help people who write proposals, scopes and client updates. The person responsible checks scope and price every time.

Notice that the sales prompts deliberately keep prices out. Pricing should come from your price list or estimating system and be checked by the estimator, not written by an AI tool from memory.

• Call notes to follow-up: "From the call notes below, draft a follow-up email that recaps what the client needs, the next steps we agreed and the date of our next contact. Do not mention pricing. Notes: [paste]."
• Scope outline: "Turn these site visit notes into a scope outline with sections for work included, exclusions and assumptions. Mark any item that seems unclear with [CHECK]. Notes: [paste]."
• Missing items check: "Compare this draft scope with the checklist below and list any checklist items the scope does not address. Do not rewrite the scope. Scope: [paste]. Checklist: [paste]."
• Report commentary: "Write three short paragraphs of commentary for a monthly client report from the figures below. Explain notable changes in plain language. Do not speculate about causes not supported by the data. Figures: [paste]."

Prompts for professional staff

Lawyers, accountants, engineers and clinicians can use AI to prepare work, never to replace their judgment. These prompts are for drafting and organizing; the professional verifies every substantive point against primary sources.

Asking for quotes of the relevant clause or passage is a simple habit that makes review much faster. The professional can jump straight to the source to confirm the point rather than rereading the whole document.

• Document summary: "Summarize this document for an internal reader in under 200 words, then list key dates, parties and obligations in a table. Quote the relevant clause for each obligation. Document: [paste]."
• Client letter draft: "Draft a plain-language letter to a client explaining the next steps listed below. Use short sentences. Do not add advice beyond what is listed. Steps: [paste]."
• Extraction: "Extract every date, amount and party name from the text below into a table. If a value is ambiguous, note it rather than guessing. Text: [paste]."

What reviewers should check

Every prompt in your library should come with a short note on what to check. Common checks include:

• Facts and figures match the source, and nothing was added that the source does not contain.
• Names, dates, amounts and addresses are correct.
• No prices, commitments or advice appear unless they were meant to.
• Tone fits the client and your firm.
• Anything marked [CHECK] or "not stated" has been resolved.

Building and maintaining a prompt library

Keep prompts that work in one shared place, such as a document in SharePoint or Google Drive, or a pinned post in Teams or Google Chat. For each prompt, record the task, the prompt text, which approved tool it was tested in, an example of good output, and the review checklist. Some assistants let you save reusable instructions or projects; use those for the most common tasks.

Give one person ownership. They accept new prompts, retire ones that stop working, and update wording when policies or prices change. Ask staff to add a prompt whenever they save meaningful time on a task, and review the library quarterly.

Track which prompts get used. A library of 15 prompts that people rely on beats a library of 200 that nobody opens.

Store prompts with placeholders in square brackets, as in the examples above, so people know exactly what to fill in. Keep sensitive examples out of the library itself.

Common prompting mistakes

A few habits cause most poor results. Asking for too much in one prompt, such as summarizing, rewriting and pricing at once, produces muddled output; split it into steps. Leaving out the source material forces the tool to guess. Not specifying the format leads to long, generic text. And accepting the first answer misses the easiest improvement: ask for a shorter version, a different tone or a check against a list.

InTheory includes a starter prompt library tailored to each role as part of team training. You can begin with the templates on this page and adapt them as you learn what your team needs.

One more mistake is trusting fluent output. Well-written text feels correct. Treat polish as a reason to check more carefully, not less.