How to use ChatGPT for your business: 5 jobs to try this week

You can use ChatGPT to match customer enquiries to your services, turn meeting transcripts into delivery plans, improve spreadsheets, answer questions from product manuals and make a week of content from a case study. The useful bit is giving it the information your staff would need to do the job well.

A customer asks for a training room that works for people on site and colleagues joining remotely. Your business sells displays, cameras, audio equipment and installation services. Finding a suitable combination means checking their requirements against your catalogue, identifying gaps and writing a clear reply. That’s a worthwhile task to try with AI.

This guide shows how to approach that job and 4 others. Each has a visual, a reusable prompt and a check before you use the result. Download all the prompts if you’d like to keep them beside your work.

The scenarios explain how the workflows work. They are illustrative, not client case studies or measured savings. The prompts are written for your own files and business information.

Give ChatGPT the context your business already has

Start with a small business brief: what you offer, who you serve, your approved policies and how you write to customers. Attach the relevant documents for each job. A catalogue is more useful than asking ChatGPT to guess what your business sells.

For recurring work, ChatGPT Projects can keep instructions, chats and source files together. Update the source material when your services or policies change. Spreadsheet add-in chats are separate, so give them the context they need too.

Help me with work for my business. Use this context alongside the files I supply.

Business and customers: [what we do, who buys from us, and where we operate]
Products or services: [attach our current catalogue, service guide or website pages]
Approved policies: [pricing rules, lead times, support limits and things we cannot promise]
Writing style: [attach our voice guide or paste a message that sounds like us]

Use supplied sources for facts about our business. Keep recommendations separate from confirmed facts. If a source is missing, outdated or contradictory, tell me. Ask for information that would materially change the answer. Never invent prices, availability, capabilities or customer results.

Use your workplace’s approved account and sharing rules for customer records. Remove details the task does not need, and check OpenAI’s data controls for your account. A saved brief is useful context, but it isn’t a guarantee that every answer will be correct.

1. Match a customer enquiry to the right solution

Useful when: a customer has a specific request and your business offers a broad range of products or services.

Think of an audiovisual supplier receiving this enquiry:

We need a training room for 20 people, with colleagues joining through Microsoft Teams. We’d like wireless presenting and to keep our existing display if possible. Can you recommend a setup and tell us what you need to check first?

Give ChatGPT the enquiry, your service catalogue, product specifications and approved sales guidance. Ask it to map the requirements to options you actually offer, then write a reply. It should surface questions about the existing display, room layout and compatibility where your documents don’t settle them.

Bring in the email and your business documents

You can paste the enquiry from any email provider. If you use Gmail or Outlook, ChatGPT’s connected apps can bring email into the conversation where available. The Outlook Email app can search and retrieve messages. Ask for a particular customer’s enquiry or thread, then check that ChatGPT has found the right one.

For your catalogue and policies, connect Google Drive or use SharePoint for supported OneDrive work or school files. Personal OneDrive isn’t supported by that SharePoint app. With another cloud storage service, use an available approved app or upload the relevant files. Check your plan, workplace permissions and which actions the connection supports. Reading an enquiry is separate from sending the reply.

The result should give your colleague something to work with: requirement → suitable option → source → detail to verify, followed by the customer reply. For example, if display compatibility isn’t confirmed, the reply should ask for the display model rather than claim your solution will work with it.

Match this customer enquiry to our products or services using the attached catalogue, business brief and approved policies.

Customer enquiry: [paste the full message]

1. List the customer's requirements, constraints and unanswered questions.
2. Suggest up to 3 suitable options from our catalogue. For each, show which requirements it meets and cite the source page or section. Mark anything not confirmed by our documents as "needs checking". If no option fits, say so.
3. Identify compatibility, stock, pricing or delivery details a colleague must verify.
4. Draft a helpful reply in our voice: explain the best-supported option, ask the questions needed to progress, and propose a sensible next step.

Keep internal notes separate from the customer reply. Do not promise a price, booking, integration or specification that the sources do not confirm.

Watch for hallucinations: ChatGPT can invent a convincing capability or recommend the wrong combination of products. Open the cited specification yourself. Treat a missing source as an unanswered question. For a technical recommendation, have the person responsible for the solution check it before it reaches the customer.

2. Turn AI meeting notes into a delivery brief

Useful when: a client call leaves you with decisions, requests, dependencies and several pieces of work to distribute.

Fireflies and Fathom can capture meeting notes. Fireflies supports downloading transcripts and summaries; Fathom lets you copy its formatted summary. Their own features can cover routine summaries. ChatGPT becomes useful when you want to compare the meeting with a project brief or turn it into several different working documents.

For example, a software implementation call covers a delayed data migration, a request for extra reporting and a revised training date. Supply the transcript and current scope. Ask ChatGPT to separate what was agreed from what was merely requested, identify dependencies and prepare the follow-up work.

The useful distinction is “client requested extra reports” versus “team agreed to deliver extra reports”. A summary that blurs those statements can create work nobody approved. Include timestamps or source passages so a colleague can check them.

Use the attached meeting transcript, current project brief and previous action list to prepare the work after this meeting.

Meeting date: [date, so relative deadlines can be interpreted]
Audience: [our delivery team / client / other]

Produce:
1. Confirmed decisions, each with a timestamp or source passage.
2. An action table with task, owner, deadline, dependency and source. Write "not agreed" for missing owners or dates.
3. Changes requested to scope, budget or timing. Separate requests from approved changes and compare them with the project brief.
4. Unresolved questions and blockers, including what we need from the client.
5. A client follow-up email covering what was agreed and what still needs confirmation.
6. A short CRM note with the customer's priorities, objections and next step.

Do not turn a suggestion into a commitment. Flag unclear names or transcription errors. Keep internal commercial notes out of the client email.

If you only have a short summary, ask for the same outputs but expect more gaps. Copying or uploading meeting notes is enough to try this flow; an automatic connection between the notetaker, ChatGPT and your CRM needs a separate setup.

3. Use AI to improve and query a spreadsheet

Useful when: a workbook has become hard to maintain, formulas are breaking or useful information is buried across tabs.

Some people hate spreadsheets. I use AI with them all the time. A useful sheet keeps work visible; AI can help you understand the formulas, tidy the structure and ask questions without knowing every spreadsheet function.

Picture a sales tracker with duplicate companies, inconsistent dates, missing websites and a summary tab that no longer agrees with the detail. The goal is a usable tracker with consistent records and totals you can check.

Work in Excel or Google Sheets

OpenAI now offers ChatGPT for Excel and ChatGPT for Google Sheets. These add-ins work inside the spreadsheet and can build, update and explain it. They can also use web search to find information for a workbook. Install the official add-in through the Microsoft Marketplace or Google Workspace Marketplace. OpenAI’s setup guide explains the installation steps and workplace access settings.

You can also upload an Excel workbook or CSV to ChatGPT for analysis. A CSV carries values, but not the workbook’s formulas, formatting or multiple tabs. Use the workbook when those matter. Connecting a file source is another way to supply context; it doesn’t automatically mean that ChatGPT can edit the live sheet.

Connect the files behind your spreadsheet

The Google Drive app can work with supported Google Sheets actions. For Excel files stored in work OneDrive or SharePoint, the SharePoint app can find and reference supported content. Access and editing depend on the connection’s enabled actions and permissions.

The Excel and Google Sheets add-ins also support connected apps: open +, then browse the apps available to your account. This can bring relevant project documents into a workbook task. Use the add-in when you want to work inside the sheet, and verify changed cells before relying on it. OpenAI’s current guide covers these routes and their limits.

Clean up the sheet and make it useful

Work on a duplicate. Ask for a plan before broad edits, then keep a cleaned tab and a checks tab. The output should make it easier to filter the work, see missing information and understand where totals come from. Don’t let an uncertain duplicate match quietly delete a customer record.

Help me make this workbook clearer and more useful for [the business task it supports].

First inspect the tabs, headers and formulas. Explain what the workbook appears to do, then propose a cleanup plan before making changes. Ask me about ambiguous fields.

After I approve the plan, work in a copy or new "Cleaned" tab:
- Standardise headings, date formats and inconsistent category labels.
- Identify duplicates using suitable business identifiers. Flag uncertain matches for review rather than merging them.
- Preserve source rows, IDs and working formulas. Explain every proposed formula repair with the affected cell and its dependencies.
- Add useful columns for this task, explaining how each is calculated or populated. Leave unsupported values blank.
- Apply readable formatting, filters, frozen headings and appropriate validation.
- Add a "Checks" tab with missing fields, suspected duplicates, formula issues, row counts and control totals compared with the original.

Finish with a change log and explain 3 useful questions I can now ask this workbook. If you cannot edit the live sheet, prepare a revised downloadable workbook instead and tell me what needs importing.

Fill missing website URLs with evidence

A company name isn’t enough to identify a business reliably. Give ChatGPT the location or other identifiers in your sheet, ask it to search official company sites and store the evidence beside each match. Review a small sample first. Leave uncertain matches blank.

Use web search to fill the missing company website URLs in this spreadsheet.

Match each company using its name plus the location or other identifiers already in the row. Prefer the company's own website. Do not assume a matching name is the same business.

Add Website URL, Evidence URL, Checked date and Match notes columns. Only fill Website URL when the match is supported. Leave uncertain rows blank and put the reason in Match notes. Do not invent contact names, email addresses or missing business facts.

Show me 5 proposed matches for review before processing the rest. Preserve all original columns and values. If web search is unavailable, tell me rather than guessing.

Find out why a formula is breaking

Point to the actual cell. Ask ChatGPT to trace its dependencies, explain the cause and propose a correction you can test. Hiding an error with a fixed number won’t repair the workbook.

Investigate the error in [sheet name and cell reference] in this workbook.

Trace the formula and its referenced cells. Explain the cause in plain English, including any broken reference, data-type mismatch or missing input.

Propose the smallest correction and explain how it affects dependent cells. Before applying it, show the old formula, proposed formula and a way to test the result against source values. Flag any business assumption that needs my decision. Do not replace a formula with a fixed number just to hide the error.

After cleanup, ask business questions such as “Which enquiries have no next action?”, “Which totals don’t reconcile?” or “Show the records missing information needed for a quote.” Ask for the rows and calculations behind the answer, so you can inspect them.

The add-ins have usage limits and spreadsheet chats don’t share your main ChatGPT history or memory. Complex formulas and macros need particular care. Check OpenAI’s current limitations before relying on a workbook change.

4. Use manuals to answer technical support questions

Useful when: staff need to answer questions about software or hardware that require detailed product knowledge.

A customer reports that a label printer stopped connecting after a software update. A newer member of staff knows the symptoms, but not where the relevant settings and troubleshooting procedures sit in several manuals. Give ChatGPT the model, version, ticket, current manual and approved support notes. Ask it to find the relevant guidance and prepare a response for review.

This can also help with a software configuration question or onboarding issue. The important input is the documentation for the actual product and version, rather than general knowledge about a similar product.

Check that ChatGPT can read the part of the manual the answer depends on. Embedded diagrams are not read in every file and plan combination; OpenAI’s file guide explains the limits. If a wiring diagram or screenshot matters, supply it separately as an image and check that its labels are legible.

The request might arrive by email or through a helpdesk such as Zendesk or Freshdesk. Start by supplying the relevant ticket conversation and attachments. Keep the ticket ID with the draft so your colleague can check the history before replying. Freshworks also offers Freshdesk MCP, a connection that lets compatible AI tools access tickets, conversations and customer information. Check what your account and chosen AI tool support before building an automatic workflow.

A useful answer has a documented check, its source and what to do if it doesn’t resolve the issue. For example, it should identify the applicable connectivity section and any required version check. It should not invent a reset command because similar devices have one.

Help our staff respond to this support ticket using only the attached product manual, approved support guidance and release notes.

Ticket: [paste the customer's problem and any error message]
Product / model: [name]
Software or firmware version: [version, or "unknown"]

1. Check that the documents apply to this model and version. Ask for missing details that change the advice.
2. Summarise the symptoms and distinguish documented causes from unconfirmed possibilities.
3. Give a step-by-step troubleshooting checklist using documented procedures. Cite the document and page or section for each step. Label any step that requires a qualified technician or could change settings, lose data or affect service.
4. Draft a clear customer response with suitable checks and questions.
5. Prepare an escalation brief: symptoms, evidence, checks completed, unresolved questions and the source sections for a specialist to review.

If the documentation does not support an answer, say so. Do not invent a feature, command, repair procedure or guarantee. Keep internal reasoning separate from the customer response.

Watch for hallucinations: a plausible technical instruction can still be wrong. ChatGPT can cite an irrelevant section or combine guidance from different versions. Verify the references and keep any destructive action or specialist procedure with the right member of staff. If the documents don’t answer the question, an escalation brief is a useful result.

5. Turn a customer case study into a week of content

Useful when: your business has already explained good work in a blog or case study, but that material hasn’t reached your social channels.

Suppose a software company publishes a case study about replacing a client’s disconnected booking and customer-management systems. The article explains the original problem, the implementation and the results the client approved. That gives you several useful angles: the warning signs, a process walkthrough, a lesson from delivery and the customer’s outcome.

Give ChatGPT the source article and your brand guidance. Ask it to adapt those ideas to the way each platform works, while keeping every claim tied to the source.

A sensible mix could include a LinkedIn story about the implementation, an X thread explaining the problem, an Instagram carousel with practical lessons, a TikTok demonstration of the flow and a Pinterest guide people can save. The same lesson can travel across channels, but it needs a different treatment in each format.

Take the attached customer case study or blog and turn it into a week of useful content ideas.

Audience: [who we want to help]
Platforms: LinkedIn, X, Instagram, TikTok and Pinterest [edit to our actual channels]
Brand voice: [attach our guide or paste an example]
Call to action: [the next step we want the reader to take]

First extract the customer's problem, the work completed, the useful lessons and the results actually stated in the source. Cite the source passage for each factual claim. Do not invent metrics, customer quotes or outcomes.

Build a 7-day content plan. For each day, give the angle, target platform, opening hook, useful takeaway, suggested format, visual or video brief, call to action and supporting source passage. Reuse an idea across platforms only when the format and wording suit each one.

Include a LinkedIn story, an X thread, an Instagram carousel, a TikTok demonstration script and a Pinterest guide idea. Mark anything that needs a new screenshot, demonstration or customer permission before use.

Then draft the LinkedIn post, carousel slide copy and TikTok script. Make each useful on its own and invite interested readers to the full article. Keep client identities and confidential details out unless approved for public use.

The content plan should say what the audience gets from each post and what needs making: a diagram, screenshot, carousel or video. A claim about faster processing needs evidence in the source. If the case study doesn’t give a percentage, the social post shouldn’t invent one.

Put AI to work across your business

Try the job closest to the work on your desk. Check whether the output is useful, where it goes wrong and which sources it needs. Save the instructions that work, and keep the business information current.

Once the result is reliable enough for your team to use, you can connect the tools around it: meeting notes into a task system, approved answers into customer support, or case-study drafts into a content calendar. That’s where the individual tasks become a working process.

The AI skills guide explains how to save repeatable instructions. The small-business AI cost guide helps with choosing a sensible starting setup.

Astraeus Business Solutions helps businesses choose useful AI tools, connect them to their existing systems and build processes their teams can use. That might mean linking meeting notes to a delivery plan, making a spreadsheet easier to maintain or helping staff find answers in technical documentation. We can also walk your team through how to use the tools and check their results.

Frequently asked questions

How can I use ChatGPT for my small business?

Give ChatGPT your business context and the material behind a task: a service catalogue, meeting transcript, spreadsheet, product manual or case study. Ask for a specific result, such as a matched solution and reply, a delivery brief, a repaired workbook, a sourced support answer or a week of content. Check the output before using it.

Can ChatGPT work with Excel and Google Sheets?

Yes. OpenAI offers ChatGPT add-ins for Excel and Google Sheets that can build, update and explain spreadsheets, including formulas, and search the web for information. You can also upload a workbook to ChatGPT. Availability and usage limits depend on your plan and workplace settings; connected file access and live editing are different capabilities.

How do I reduce ChatGPT hallucinations at work?

Supply current source documents, specify the product or version, ask for page or cell references and tell ChatGPT to flag missing information rather than guess. Separate documented facts from suggestions. Verify citations, formulas and important claims against the originals. Test with questions whose answers you already know and have a colleague review customer-facing advice. These checks reduce risk; they do not eliminate mistakes.

Do I need a paid ChatGPT plan to try these prompts?

You can start on the Free plan, including limited access to file tools and the spreadsheet add-ins. Usage limits, apps and workplace controls vary. Check the current availability for your account before choosing a paid plan.

Can ChatGPT connect to Gmail, Outlook and Google Drive?

Yes, where the relevant apps are available to your account. Connect them through the Apps or Plugins settings in ChatGPT and review the permissions. Google Drive supports connected files and permitted Google Sheets actions. Supported work or school OneDrive content uses the SharePoint app. Availability depends on your plan, location and workplace settings.

Will these prompts send emails or update my business software?

The enquiry, meeting, support and content prompts ask for drafts to review. The spreadsheet add-ins can make workbook changes, so use a copy and review the proposed edits. Sending messages or updating other business systems requires the relevant connection and permissions. A prompt alone does not set up those connections.

Provider logos identify the tools discussed. They belong to their respective owners; this guide is published independently by Astraeus Business Solutions.