AI Prompts for Business: Examples You Can Use Today

AI prompts that work for business give the tool a role, the background, a clear task, a format for the answer and a check on its own work. Below are copy-and-paste prompts for writing, summaries, spreadsheets, customer replies and meeting prep, with a note on why each one works.
Outputs vary from one run to the next, and AI tools make mistakes. Every result here is a first draft for a person to check. For the bigger picture first, read AI for business. If you want help applying prompts like these across your team, see the AI consulting page.
What is an AI prompt, and what is prompt engineering?
An AI prompt is the instruction you give an AI tool. Prompt engineering is the habit of writing those instructions carefully and improving them after you see the result. Despite the technical name, it is mostly clear writing. Grammarly’s guide to prompts makes the same point with two of its first tips: choose one goal or task, and use clear and precise language.
The two big AI companies say the same thing in their own guidance. Anthropic, the maker of Claude, suggests showing your prompt to a colleague who has little context and asking whether they would be confused. If they would be, the AI will be too. OpenAI’s prompt engineering guide lists the same basics: clear instructions, relevant context, examples, and testing the results.
With Claude, typos do not matter when you prompt it. You do not need perfect grammar or spelling, because it knows what you are trying to do.
Free resource · Guide
Mastering Claude
My guide to setting up Claude and getting real work out of it, from your first prompt to a daily workflow.
Get it free →
The people who most need this are not the experts. The U.S. Census Bureau reported that 19.8% of businesses were using AI as of May 3, 2026, and under 20% of firms with four or fewer employees. Most small teams are still learning, and a short list of prompts that work is a fast way to start.
How do you build a good prompt?
Use five parts: role, context, task, format and check. You do not need all five every time, but a prompt that has all five rarely comes back useless.
| Part | What it does | Example wording |
|---|---|---|
| Role | Sets the point of view and the level of skill | “Act as an experienced bookkeeper.” |
| Context | Gives the background the tool cannot guess | “Our firm has 12 staff and sells to dental practices.” |
| Task | States the one thing you want done | “Write a follow-up email after a sales call.” |
| Format | Says what the answer should look like | “Under 120 words, plain English, no bullet points.” |
| Check | Makes the tool review its own work | “List any assumptions you made and anything you are unsure of.” |
Here is the whole structure as a reusable template. Replace the brackets and delete the lines you do not need.
Act as [role].
Context: [who you are, who the audience is, what has happened so far].
Task: [one clear job].
Format: [length, tone, layout].
Before you answer, tell me what you are assuming and what you would need to know to do this better. After you answer, flag anything that a person should double-check.
This works because it removes guessing. The tool knows whose voice to use, what the situation is, what to produce and how it should look, and it is asked to point out its own weak spots. The last line does not make the answer correct, but it gives you a list of places to look.
Which AI prompts help with writing?
Writing prompts work best when you supply the facts and ask the tool to shape them. Never ask it to invent the facts. Paste in your own notes, and keep the numbers and names that matter in the prompt.
Prompt 1: a follow-up email after a call
Act as a friendly, direct sales professional.
Context: We sell [product or service] to [type of customer]. I just spoke with [first name], who runs [type of business]. They said their main problem is [problem] and they want to decide by [date]. Next step: [next step].
Task: Write a follow-up email.
Format: Under 120 words. Plain English. One clear question at the end. No hype words.
List anything in the email that I should confirm before sending.
Why it works: the email is built from real details, so it does not sound generic. The word limit and the single question keep it short, and the last line flags promises a person must confirm.
Prompt 2: tighten a draft without changing the meaning
Here is a draft: [paste draft]
Rewrite it so it is shorter and easier to read for [audience]. Keep every fact, number and name exactly as written. Use short sentences and plain words. After the rewrite, list every change you made to the meaning, if any.
Why it works: “keep every fact” protects the details the tool is most likely to change. Asking for a list of meaning changes makes quiet edits visible.
Prompt 3: turn rough notes into a one-page document
Act as an operations manager writing for a new hire.
Context: Below are rough notes on how we handle [process]. They are messy and may contradict each other.
Task: Turn them into a one-page checklist with numbered steps. Mark any step that is unclear or that the notes contradict, instead of guessing.
Notes: [paste notes]
Why it works: telling the tool to mark gaps instead of guessing turns hidden problems in your process into a visible list of questions.
Which AI prompts help with summarising?
Summary prompts work when you say who the summary is for and what decision it supports. A summary written for an owner reads differently from one written for a new hire.
Prompt 4: summarise a long document for a decision
Act as a careful analyst.
Context: I am deciding whether to [decision]. Below is a [contract, report, proposal or article].
Task: Summarise it in no more than 8 bullet points for a busy owner. Then list: (1) anything that affects cost, (2) any deadlines, (3) anything unusual or one-sided, and (4) questions I should ask before agreeing.
Quote the exact sentence from the document for each point in sections 1 to 3, so I can find it.
Document: [paste text]
Why it works: asking for the exact sentence gives you something to verify. A summary with no quotes is much harder to trust. This prompt helps you read faster. It does not replace reading the parts that matter, and it is not legal advice, so have a qualified person review any contract.
Prompt 5: summarise a long email thread
Below is an email thread. Write a summary with three headings: What was decided, What is still open, and Who needs to do what by when. If a person's name or a date is not stated clearly, write "not stated" instead of guessing.
Thread: [paste thread]
Why it works: three fixed headings give you the same shape every time. “Not stated” gives the tool permission to admit a gap.
Which AI prompts help with analysing a spreadsheet?
Spreadsheet prompts work best when you describe the columns and the question before you ask for anything. Most AI assistants can read a CSV or Excel file you upload. Check the tool’s privacy terms and remove personal data first.
Prompt 6: find patterns in sales or expenses
Act as a business analyst.
Context: The attached spreadsheet has one row per [sale, invoice or expense]. Columns: [column names and what each means]. The period covered is [dates].
Task: Find the three most important patterns a business owner should know about, such as the biggest change over time, the biggest category, and anything that looks unusual.
Format: For each pattern, give one plain-English sentence, the numbers behind it, and the exact rows or columns you used so I can check them. If something in the data looks wrong or missing, say so before you analyse it.
Why it works: listing the columns avoids misreading. Asking for the rows used lets a person confirm each number. Asking about bad data first stops the tool from building conclusions on a broken file.
Prompt 7: ask for the formula, not just the answer
In my spreadsheet, column A has [what], column B has [what], and column C has [what]. I want to [goal, for example: total column C for each value in column A, only for dates in the last 30 days]. Write the formula for [Excel or Google Sheets], explain each part in one sentence, and give me one small example with made-up numbers I can use to test it.
Why it works: a formula you can see and test is safer than a number the tool calculates for you. AI tools can slip on arithmetic, and a formula in your own sheet does the sums the same way every time.
Which AI prompts help with customer replies?
Customer reply prompts work when you give the tool your real policy and ask it to stay inside it. Without that, it may promise a refund or a deadline that your business never offered.
Prompt 8: draft a reply to a customer message
Act as a calm, helpful customer support specialist at [business name].
Context: A customer wrote the message below. Our policy: [paste the relevant policy in plain words, such as refund window, delivery times, what you can and cannot do]. Our tone: [for example, warm, brief, no jargon].
Task: Draft a reply that answers their question using only the policy above. Do not promise anything that is not in the policy. If the policy does not cover their situation, say so in a note to me instead of making something up.
Format: Under 150 words. Start with their name. End with the next step.
Customer message: [paste message]
Why it works: “using only the policy above” keeps the reply inside what your business will honour. The note to you is where the tool raises cases that need a person.
Prompt 9: turn an angry message into a response plan
Below is a complaint from a customer. Do not write the reply yet. First, tell me (1) what the customer actually wants, (2) which facts I need to check internally before replying, and (3) the two most reasonable ways to resolve this. Then wait for my choice.
Complaint: [paste]
Why it works: splitting thinking from writing stops the tool from rushing to an apology that commits you to something. A person picks the resolution, and then the tool writes it.
For replies that go to real customers, keep a person in the loop until you have reviewed enough of them to trust the pattern. The same advice applies when you move from prompts to automation, covered in AI automation for small business.
Which AI prompts help with meeting prep?
Meeting prep prompts work when you tell the tool who you are meeting and what you want to leave with. They save the most time before sales calls, supplier talks and performance reviews.
Prompt 10: prepare for a meeting
Act as a sharp chief of staff.
Context: I have a [length]-minute meeting with [name and role] from [company] about [topic]. Here is what I know about them and the history: [paste notes or past emails]. The outcome I want: [outcome].
Task: Prepare me. Give me (1) a three-line summary of where things stand, (2) five questions to ask, ranked by importance, (3) the two objections they are most likely to raise and a calm answer to each, and (4) a suggested closing line that asks for the next step.
Only use facts from my notes. Where you are guessing, label it "guess".
Why it works: the labelled-guess rule is the important part. It separates what came from your notes from what the tool filled in.
Prompt 11: turn meeting notes into actions
Below are notes from a meeting. Produce a list of action items. For each, give: the task, the owner, and the due date. If the owner or date was not said, write "not stated" and add it to a separate list of questions to ask the group. Then write a three-sentence summary I can send to everyone who attended.
Notes: [paste notes or transcript]
Why it works: “not stated” stops the tool from assigning people jobs they never agreed to. A person should read the list before it is sent.
How do weak prompts compare with strong ones?
The difference is almost always missing information. The table shows the same request written two ways.
| Job | Weak prompt | Stronger prompt |
|---|---|---|
| “Write a follow-up email.” | “Act as a direct sales professional. We sell scheduling software to dental practices. Write a follow-up under 120 words to a practice manager who asked about pricing and wants to decide by Friday. End with one question.” | |
| Summary | “Summarise this report.” | “Summarise this report in 8 bullet points for an owner deciding whether to renew. List costs, deadlines and anything one-sided, and quote the sentence for each.” |
| Spreadsheet | “Analyse this data.” | “Columns are date, product, units and price. Find the three biggest patterns in the last 6 months, show the numbers behind each, and tell me first if any rows look wrong.” |
| Customer reply | “Reply to this angry customer.” | “Using only our refund policy below, draft a calm reply under 150 words. If the policy does not cover this case, tell me instead of guessing.” |
| Meeting prep | “Help me prepare for my meeting.” | “Prepare me for a 30-minute supplier call. Here are my notes. Give me five ranked questions, two likely objections with answers, and a closing line. Label guesses.” |
Each stronger version names the audience, supplies facts, sets limits on length and format, and tells the tool what to do when it does not know.
How do you improve a prompt that did not work?
Change one thing at a time and keep the version that gives better results. Most failed prompts fall into one of four problems.
- The answer is generic. Add specifics (Anthropic’s guidance also says explaining why you want something helps the tool aim better): who the audience is, what has already happened, a real number or a real constraint.
- The answer is the wrong shape. Say the format outright: length, tone, headings, bullet points or none.
- The answer is too confident. Add the check line: “list your assumptions” or “label anything you are guessing.”
- The answer varies too much. Give one example of what good looks like. Both Anthropic’s and OpenAI’s guidance recommend examples, and Anthropic suggests three to five for the best results.
You can also ask the tool for help: “Here is my prompt and the answer I got. What was missing from my prompt that led to this answer? Rewrite the prompt.” This often works, but you still decide whether the rewrite fits your business.
What should you check before using an AI answer?
Check every fact, number, name, date and promise, and keep customer or confidential data out of tools your business has not approved. Outputs vary, so a prompt that worked on Monday may give a different answer on Friday.
- Facts and figures. Confirm them against your own records or a primary source.
- Promises. A reply that offers a refund, a date or a discount needs a person’s approval.
- Private data. Leave out names, contact details, health or financial information, and passwords until you have read the tool’s data terms and set a company rule.
- Tone. Read it aloud. If it does not sound like your business, edit it.
- Responsibility. The person who sends the output owns it.
How do you turn good prompts into something your team can reuse?
Save the prompts that work in one shared place, with the date, the tool you used and a note on what to change each time. Teams that do this stop starting from a blank box.
- Create a shared document with one section per job: writing, summaries, spreadsheets, customer replies and meeting prep.
- Paste in the prompts above, fill in your business details once, and mark the bracketed parts that change each time.
- After each use, note what you edited in the output, then fix the prompt so you do not have to make that edit again.
- Once a prompt is stable, consider packaging it. The Claude skills pack and the Claude agents resource show how repeatable instructions can be set up so they load when needed.
My rule: stop writing one-off prompts. Anything you already worked through in a chat, tell Claude to turn it into a skill. A skill is just a folder with one text file that teaches Claude one job step by step, and it keeps working after a long chat gets compacted and the original instructions fade.
Video · 15 min
If You're Not Using Skills, You're Using Claude Wrong
I show how to turn a prompt that works into a reusable skill, so Claude does the job the same way every time.
Watch the video →
Prompts are the first step. Training people to use them well matters as much as the prompts themselves; see AI training for employees and AI adoption. My AI workshops teach teams to write prompts like these hands-on.
Where should you go next?
If the prompts above save your team time, the next step is choosing which jobs to hand over more fully and where a person must stay in the loop. A short working session on your own tasks is a faster route than trial and error. See my live AI workshops for teams, or start with the consulting page to talk through your use case.
Related guides
- AI for small business: where to start
- AI tools for small business
- AI automation for small business
- AI use cases for business
- AI training for employees
- AI adoption
Updated November 11, 2026. Sources opened on that date: Anthropic’s “Prompting best practices,” OpenAI’s “Prompt engineering” guide, and the U.S. Census Bureau story “Large Firms With at Least 20 Employees Biggest AI Users” (May 26, 2026).


