AI Prompt Library

Free AI Prompt Library for Business: 30 Copy-Paste Prompts

Copy-paste prompts for customer support, sales, marketing, operations, meetings, research and hiring. Each one says what it is for and tells the model to use only the facts you give it. They work with any chat model, such as ChatGPT, Claude, Gemini or a local model.

  • 30 prompts in 7 groups
  • No sign-up and no paid placements
  • Last updated October 26, 2026

01Start here

How to use these prompts

These are starting points I wrote, not benchmarked or guaranteed. A prompt saves you the blank page, but you still own the result.

  1. Type your details into the fill-in boxes below to replace the most common [BRACKETS] in every prompt at once. Replace any that are left, and each {{variable}}, with your own text.
  2. Run it on three real examples before you rely on it.
  3. Check names, numbers and dates against the source. A model can sound sure and still be wrong.
  4. Do not paste confidential customer or employee data into a tool until you have read its data policy.
  5. The Open in ChatGPT and Open in Claude buttons start a new chat with the prompt already typed in. Replace any {{variables}} before you send it.
  6. Keep a person responsible for anything that is sent, signed or decided.
Download the AI Starter Kit (ZIP)

02Personalise

Fill in your details once

Type your details below and every matching [BRACKET] in all 30 prompts updates on this page as you type. Anything you leave empty stays as a placeholder. What you type stays in your browser.

Replaces [COMPANY]
Replaces [TONE]
Replaces [AUDIENCE]
Replaces [PRODUCT OR SERVICE]
Replaces [PRICE]
Replaces [ROLE]

03Customer support prompts

Customer support prompts

Use these to sort and draft, never to send. A person should read every reply before it reaches a customer.

Support email triage

Sorts incoming email into categories and flags anything a person must read.

You are a support triage assistant for [COMPANY]. Read the email below and return JSON only, with these keys:
- category: billing, bug, how-to, sales, spam or other
- urgency: low, normal or high
- summary: at most 25 words
- needs_human: true or false

Set needs_human to true if the email mentions legal action, a refund above [AMOUNT], account deletion, a safety issue, or anything you are unsure about. Do not invent facts about our product.

Email:
{{email_body}}

Reply draft for human approval

Drafts a reply that a person reviews before anything is sent.

Draft a reply to the customer message below. A person will review it before it is sent.
Tone: [TONE, for example friendly and direct].
Use only facts from the knowledge snippets. If the answer is not in them, write NEEDS ANSWER: followed by the question for the team instead of guessing.
Keep it under 120 words.

Customer message:
{{message}}

Knowledge snippets:
{{snippets}}

FAQ draft from repeated questions

Turns a batch of real customer questions into FAQ entries.

Below are real customer questions from the last [PERIOD]. Group the ones that mean the same thing. For each group write one FAQ entry: a short question in the customers own words, and an answer of at most 60 words using only the facts in the notes.
If the notes do not answer a group, write NEEDS ANSWER: and the question for the team.

Questions:
{{questions}}

Notes:
{{notes}}

Escalation handoff summary

Summarises a long support thread for the teammate taking over.

Summarise the support thread below for a teammate who has not seen it. Write: what the customer wants, what has been tried, how the customer feels (one line, based only on their words), and the next step you recommend.
Do not add facts that are not in the thread. If something is unclear, list it under Questions.

Thread:
{{thread}}

Calm reply to an upset customer

Drafts a steady reply for a person to review before sending.

A customer is upset. Draft a reply that a person will review before it is sent.
Acknowledge the specific problem in one sentence without admitting fault we have not confirmed. Do not promise a refund, date or fix unless it appears in the confirmed facts. Offer one clear next step. Keep it under 120 words.

Customer message:
{{message}}

Confirmed facts:
{{facts}}

04Sales and lead prompts

Sales and lead prompts

Base every answer on notes or facts you supply, and check anything that sounds like a promise before you send it.

Lead qualification from form text

Rates an enquiry against your ideal customer and lists what is missing.

Read the enquiry below and return JSON only with: fit (strong, possible or poor), budget_signal (stated, implied or none), urgency (low, normal or high), reason (at most 20 words), and missing_info (a list of questions to ask).
Our ideal customer: [DESCRIBE IDEAL CUSTOMER]. Base the answer only on the text. If it is unclear, use possible and say what is missing.

Enquiry:
{{enquiry}}

Follow-up email after a call

Drafts a short follow-up for a person to edit.

Write a follow-up email after a sales call. A person will edit it before sending.
Include: a one-line thanks, a recap of the three needs the prospect mentioned, the agreed next step with its date if one was given, and one question. Use only the call notes. Keep it under 150 words, plain and friendly.

Call notes:
{{notes}}

Objection answers from product facts

Prepares honest answers to the objections buyers raise most.

List the five objections a buyer is most likely to raise about [PRODUCT OR SERVICE]. For each, write a two-sentence answer using only the facts below. If the facts do not support a good answer, say so and suggest what we would need to find out.

Facts:
{{facts}}

Proposal outline from discovery notes

Builds the structure of a proposal, not the final text.

Turn the discovery notes below into a proposal outline with these headings: Goal, Current situation, Proposed approach, Scope (included and not included), Timeline, Price (leave as [PRICE]), Risks and assumptions.
Use only what the notes say. Mark every gap with ASK CLIENT: and the question to ask.

Notes:
{{notes}}

05Marketing and content prompts

Marketing and content prompts

These give you drafts and options. Add real examples and numbers yourself, and keep only the claims you can back up.

Content brief

Gives a writer a clear brief before drafting.

Create a content brief for an article on [TOPIC] for [AUDIENCE]. Include: the one question the article must answer, a suggested title and three alternatives, an outline of H2 headings, what the reader should be able to do afterwards, and a list of facts or figures that need a source before publishing.
Do not state any statistic yourself.

Turn an article into social posts

Creates short posts from an article you wrote.

Using only the article below, write [N] social posts for [PLATFORM]. Each must make one point from the article, stay under [LIMIT] characters, and avoid hype words. Do not add claims or numbers that are not in the article. End each with a plain call to read it.

Article:
{{article}}

Email subject line variants

Gives options to test, not a promise of results.

Write ten subject lines for the email below: five plain and descriptive, five curiosity-led. Each under 55 characters. No clickbait that the email body does not deliver on, and no all caps.

Email:
{{email}}

Meta title and description writer

Drafts search snippets within the length limits.

Write a meta title (under 60 characters) and a meta description (under 155 characters) for the page below. Put the main phrase [KEYWORD] near the start of the title, describe what the reader gets, and do not promise anything the page does not contain. Give three options and count the characters of each.

Page content:
{{content}}

Plain-language edit

Tightens writing without changing its meaning.

Edit the text below for clarity. Shorten long sentences, replace jargon with plain words, and cut filler. Keep every fact, number and name exactly as written, and keep my voice. Return the edited text, then a short list of what you changed and anything you were unsure about.

Text:
{{text}}

06Operations and finance prompts

Operations and finance prompts

The prompts that extract or calculate are only as good as the data you paste in, so spot-check every number.

Receipt or invoice extraction

Pulls structured fields out of messy receipt text.

Extract these fields from the receipt text below: vendor, date (YYYY-MM-DD), currency, subtotal, tax, total, payment_method.
Return JSON only. Use null for any field you cannot find. Do not guess or calculate missing values.

Receipt text:
{{text}}

Weekly KPI summary

Writes the wording of a weekly update from numbers you have already calculated.

You are writing a weekly update for [AUDIENCE]. Using only the numbers below, write five bullets: what changed, by how much (absolute and percent), the likely reason only if the data supports one, and one thing to check next week.
If a number looks like a data error, say so instead of explaining it.

Data:
{{table}}

AI feature cost check

A sanity check to run next to the AI API cost calculator above.

Here is my planned AI feature: [DESCRIPTION].
Average input tokens per request: [N]. Average output tokens per request: [N]. Requests per month: [N].
Price per 1M input tokens: [PRICE]. Price per 1M output tokens: [PRICE].

Calculate the monthly model cost step by step, list three assumptions that could make the real cost higher, and suggest one way to cut cost without hurting quality.

Standard operating procedure from rough notes

Turns messy notes into a numbered procedure.

Turn the rough notes below into a standard operating procedure for [TASK]. Use numbered steps, name who does each step and which tool they use, add a check after each risky step, and end with what to do if something goes wrong.
If a step is missing or unclear, write ASK: and the question instead of inventing it.

Notes:
{{notes}}

Vendor quote comparison table

Lines up quotes on the same terms.

Compare the vendor quotes below in a table with one row per vendor and these columns: price, what is included, what costs extra, contract length, cancellation terms, and anything unclear.
Use only the text provided. Write NOT STATED where a quote is silent. Finish with three questions to ask each vendor.

Quotes:
{{quotes}}

Polite payment reminder

Drafts a firm but friendly reminder for review.

Write a polite payment reminder for invoice [NUMBER], amount [AMOUNT], due [DATE], now [N] days overdue. Tone: [TONE]. Mention how to pay and who to contact. Do not threaten, add fees or invent terms that are not listed below. Keep it under 100 words.

Payment terms:
{{terms}}

07Meeting and admin prompts

Meeting and admin prompts

Paste real notes or threads, and keep names, dates and decisions exactly as written.

Meeting notes to action items

Turns a transcript into decisions, owners and open questions.

From the transcript below, list:
1. Decisions made
2. Action items, each with an owner and a due date (write unassigned if either is missing)
3. Open questions

Keep dates exactly as spoken, and do not add tasks nobody mentioned. Quote nothing longer than 10 words.

Transcript:
{{transcript}}

Meeting agenda

Builds an agenda with time boxes.

Create an agenda for a [LENGTH]-minute meeting about [PURPOSE] with [ATTENDEES]. For each item give the question to settle, who leads it and a time box. Put decisions first and updates last. End with five minutes for actions. Flag any item that could be an email instead.

Long email thread summary

Summarises a thread and pulls out what you owe.

Summarise the email thread below in this order: the decision or question at the centre, what each person wants, what is agreed, what is still open, and anything I need to reply to or do (with who asked and any date they gave).
Quote nothing longer than 10 words and add no facts.

Thread:
{{thread}}

Decision log entry

Records why a decision was made.

From the notes below write a decision log entry with: the decision, the date, who decided, the options considered, why this one was chosen, what would make us revisit it, and the review date. Use only the notes and mark gaps with UNKNOWN.

Notes:
{{notes}}

08Research and decision prompts

Research and decision prompts

A model can list what to check, but it cannot confirm that something is true. Open the sources.

Claims to verify

Lists what to check before you repeat an article.

Read the text below and list every factual claim that could be checked: numbers, dates, names, quotations and cause-and-effect statements. For each, say what kind of source would confirm it. Do not say whether it is true.

Text:
{{text}}

Compare options against my criteria

Scores options on what matters to you.

I am choosing between [OPTION A], [OPTION B] and [OPTION C] for [PURPOSE]. My criteria, in order of importance: [CRITERIA]. Make a table with one row per option and one column per criterion, using only the information I provide below. Write UNKNOWN where it is missing, then list what I should check first.

Information:
{{info}}

Pre-mortem for an automation

Finds failure points before you build.

We plan to automate [PROCESS] using [TOOLS]. Imagine it failed badly three months from now. List the ten most likely reasons, ordered by likelihood times damage. For each give an early warning sign and one safeguard, such as human approval, a spending cap or a log. Mark which risks involve customer data.

09Hiring prompts

Hiring prompts

Use the same questions and scorecard for every candidate, and keep a person responsible for every decision.

Job description from outcomes

Writes a job description around results.

Write a job description for [ROLE] at [COMPANY]. Start from the outcomes this person should deliver in the first 90 days: [OUTCOMES]. Include a short overview, responsibilities, must-have skills, nice-to-have skills, and what we offer from the facts below. Avoid unnecessary requirements and exclusionary wording. Do not invent pay or benefits.

Facts:
{{facts}}

Screening questions

Creates consistent questions for every candidate.

Write eight screening questions for [ROLE]. Each should test one skill from this list: [SKILLS]. For each question give what a strong answer includes and what a weak answer sounds like. Use the same questions for every candidate and avoid questions about age, family, health or other protected topics.

Interview scorecard

Keeps scoring consistent across interviewers.

Create an interview scorecard for [ROLE]. Use 5 to 7 criteria taken from the outcomes below. For each, give a 1 to 4 scale with a plain description of each score, and a space for evidence quotes. Add a reminder to score before discussing with other interviewers.

Outcomes:
{{outcomes}}

10About this library

An AI prompt library for business is a set of reusable instructions that you paste into a chat model so it does the same job in the same way every time. This page is a free AI prompt library for business with 30 copy-paste prompts for customer support, sales, marketing, operations, meetings, research and hiring. Below I explain how the prompts are built, where to start, and the checks that keep the output safe to use.

What is an AI prompt library for business?

Most people write a fresh prompt every time they open ChatGPT, Claude or Gemini. The result changes with their wording and with how much context they remembered to include. A prompt library fixes that by saving the instructions that already worked, so a new teammate, or you on a busy Friday, can reuse them without starting from a blank page.

A library does not make a model more accurate. What it does is make your inputs and your checks consistent, which is where most avoidable mistakes start. A good business prompt says what the task is, which facts the model may use, what the answer should look like and what to do when the facts are missing. I wrote the prompts above to follow that pattern, and each one says what it is for and tells the model to stay with the facts you supply.

Which prompts should you try first?

If you are new to this, pick the job that eats the most time each week and start there. A rough guide:

How I structure a good business prompt

Whether you use these prompts or write your own, five parts make the biggest difference:

  1. The task in one line. Say plainly what you want, for example “Summarise the support thread below for a teammate who has not seen it.”
  2. The facts it may use. Paste the real notes, thread or figures and label them. Without them the model fills the gaps with guesses.
  3. The limits. Set a length, a tone and what it must not do, such as promising a refund or a delivery date that you have not confirmed.
  4. The output format. Ask for a table, a numbered list or a fixed set of headings, so the answer is quick to check.
  5. An exit when facts are missing. Tell the model to write a marker such as NEEDS ANSWER instead of inventing something. A marker is much easier to spot than a confident wrong answer.

The model makers publish their own guidance on writing prompts, and it is worth reading once. See Anthropic’s prompt engineering overview and OpenAI’s prompt engineering guide. The details differ, but both stress clear instructions, supplied context and testing on real examples.

How to keep an AI prompt library for business safe

The prompts are the easy part. The habits around them decide whether they save time or create problems.

Protect customer and employee data

Do not paste confidential customer or employee details into a tool until you have read its data policy. When in doubt, remove names, emails and account numbers first, or use made-up examples while you test. My guide to AI agent governance for small businesses covers who should be allowed to use which tools, and with what data.

Keep a person responsible

Use the output to draft, sort and summarise, not to send, sign or decide on its own. A person should read every customer reply and check every number against the source before it goes anywhere.

Test before you trust

Run each prompt on three real examples, including one awkward one, before you rely on it. Models can sound sure and still be wrong, and the same prompt can give different answers on different days or on different models. If you are still choosing a model, my comparison of ChatGPT, Claude and Gemini for business is a good place to start.

Make this AI prompt library for business your own

The prompts here are starting points, so change them. Replace the [BRACKETS] and {{variables}} with your details, add your house style and remove anything that does not fit your business.

Then keep your edited versions somewhere your team can find them, such as a shared document, with a short note on what each prompt is for, which model you tested it on and the date. When a prompt gives a bad answer, work out why and fix the prompt, rather than just correcting that one output. You can download the whole set with the buttons near the top of this page.

From prompts to automation

Once a prompt works and you use it every day, you can automate it. A prompt that triages support email, for instance, can run inside a workflow instead of being pasted by hand. The free n8n workflow templates in the AI Toolkit are one place to start, and how to build a no-code AI agent pipeline walks through the steps.

Before you commit, check the running cost with the AI cost and ROI calculators, and read how to use AI to automate a small business for the wider picture. Automating a prompt does not remove the need for review, so keep a human check on anything that reaches a customer.

Frequently asked questions about this AI prompt library

Is this AI prompt library for business really free?

Yes. There is no sign-up and no paid placement, and you can copy any prompt straight from the page or download the full set as a Markdown file.

Which AI model works best with these prompts?

They are written to work with any chat model, including ChatGPT, Claude, Gemini or a local model. Results vary between models, so test on your own examples and keep the one that gives the most reliable answers for your task.

Can I change the prompts to suit my business?

Yes, that is the idea. Replace the bracketed details, add your tone of voice and trim anything you do not need. Keep the instruction to use only the facts you supply, because it stops most invented details.

Will a prompt give the same answer every time?

No. Models produce slightly different answers on each run, and they change over time. That is why I suggest testing each prompt on a few real examples and checking names, numbers and dates against the source.

Is it safe to paste customer information into an AI chatbot?

Not by default. Read the tool’s data policy first, remove personal details where you can, and avoid pasting anything confidential until you know how it is stored and used.

How often is this AI prompt library for business updated?

The “last updated” date near the top of the page shows when it last changed. I revise or replace a prompt when I find one that does not behave, and I add new ones as I need them.

Start small: pick one prompt from this AI prompt library for business, run it on three real examples, then keep it, rewrite it or drop it. If you want more free tools to go with it, the AI Toolkit has calculators, templates and a directory of self-hosted AI tools.

11More free resources

More free AI tools and templates

These prompts are one part of the AI Toolkit. The rest is below.