ChatGPT innovative uses now extend well beyond asking for a paragraph of text. For a business owner or manager, the more useful question is where artificial intelligence (AI) can help turn existing information into something you can check and use: a comparison, a clearer explanation, a draft procedure or a better prepared conversation.

Our earlier article on innovative uses of ChatGPT explored the possibilities of conversational AI. This new guide takes a practical look at the subject in September 2026, with examples built around everyday business work.

The scenarios below are illustrations you can adapt, rather than claims about named customers or measured results. We link to current official guidance where it supports a capability. Features and connected services differ between products and accounts, so check what is available in your organisation before planning a trial.

Early discussions often focused on whether a chatbot could write convincingly. Today's business question is broader: can it help you work through the information behind a decision and produce a useful first version of the result?

OpenAI's current collection of work examples includes research, document preparation, data analysis and connected work. It also covers separate tools, including Codex, so do not assume every example is available in an ordinary ChatGPT conversation.

Keep three things distinct. ChatGPT is a product people use directly. Other applications can use OpenAI models through a software connection. An AI assistant connected to business tools may be able to carry out actions as well as draft text. These are different arrangements, with different access and review needs.

That distinction matters when someone says a company “uses ChatGPT”. A useful example explains what people actually do, which information they use and who checks the result. A familiar company name alone tells you very little.

Imagine a small distributor comparing three ways to handle customer collections. The manager has supplier brochures, staff observations and a list of essential requirements. Asking “which system is best?” leaves too much room for assumptions.

A better trial is to provide the approved material and ask for a comparison against your requirements. OpenAI documents a research-to-decision workflow that brings evidence, alternatives and unanswered questions into a reviewable brief.

Try this prompt: “Compare these options against the requirements I have supplied. Link each factual point to its source. Separate confirmed information, your interpretation and questions we still need to ask. Do not fill gaps with guesses.”

The result should help prepare a supplier conversation. Someone still needs to open the sources, confirm that they are current and check whether the comparison treats each option fairly. Judge the trial by whether it makes the next decision clearer, including reasons to postpone it.

A service business might have separate exports for enquiries, bookings and completed jobs. Before producing an attractive chart, it needs to know whether those files describe the same periods and use matching identifiers.

OpenAI's spreadsheet consolidation example describes combining exports into a workbook with checks for mismatched records. The practical starting point is a small, permitted sample and an explanation of what each column means.

Try: “Inspect these exports before combining them. Identify missing identifiers, duplicate rows, different date ranges and records that do not match. Explain your proposed matching rule and wait for me to confirm it.”

Ask for the exceptions as well as the summary. Check totals against the originals and examine several matched and unmatched records yourself. A confident explanation does not establish that a calculation is correct. This trial is useful when it exposes questions that would otherwise be hidden inside a monthly report.

An organisation may receive feedback through surveys, support messages and conversations with staff. Reading each source separately can make it difficult to see whether people are describing the same problem in different words.

OpenAI describes grouping feedback across sources in ChatGPT Work. A simple illustrative trial is to supply a permitted, de-identified sample of comments and request themes with references back to the original entries.

Try: “Group these comments by the problem the customer experienced. Show supporting comment references, keep unusual but serious issues visible, and separate customer statements from suggested explanations.”

Review both the common themes and the awkward exceptions. Frequency does not automatically establish importance, and a short comment may be ambiguous. The useful output is a set of questions for your service team, with enough evidence to revisit the interpretation. It should not become an automatic judgement about individual customers or employees.

Consider a customer asking why a delivery date changed. The challenge is not simply producing polite wording. The reply must reflect what staff know, what they are authorised to offer and which detail still needs checking.

For this illustrative trial, provide a fictional enquiry, an approved response policy and the confirmed facts. Ask for a draft that respects those boundaries. This lets the team evaluate the approach without starting with a real customer's personal information.

Try: “Draft a friendly reply using only these facts and this policy. Do not promise a refund, delivery date or exception unless it is explicitly authorised. List anything the sender must confirm before using the reply.”

The staff member remains responsible for the message. Compare the draft with the policy, check its tone and look for implied commitments as well as explicit promises. If the checking takes longer than writing the reply, narrow the task or choose another use case.

Many useful processes live partly in a document and partly in somebody's experience. A handover can reveal that nobody has written down what happens when information is missing, a request is rejected or the usual colleague is away.

A practical experiment is to give ChatGPT a fictional set of working notes and ask for a draft procedure. Specify the audience, the start and finish of the task, and which decisions require a manager.

Try: “Turn these notes into a step-by-step procedure for a new colleague. Keep the existing rules. Identify missing decisions and exceptions as questions rather than inventing a policy.”

Have someone familiar with the work review it, then ask another colleague to walk through an example using the draft. Notice where they hesitate. A polished document can still describe the wrong process; the value comes from making gaps visible and helping the team agree a clearer version.

You do not need to become a software specialist to ask better questions about a proposed system. You do need to understand enough of the proposal to recognise what is clear, what is assumed and what needs explaining.

OpenAI's learning example describes using source material to produce a structured explanation. For a business reader, a useful variation is a short glossary and a list of questions based on an approved supplier document.

Try: “Explain this proposal for an operations manager. Define the unfamiliar terms, give a simple example for each major idea and identify statements that the document does not explain sufficiently.”

Check the explanation against the original and take unresolved questions to the supplier. Treat it as preparation for a conversation, rather than professional assurance that a design is suitable. The aim is to help you participate confidently while keeping the source document in view.

A useful discussion about software often begins with a rough description: “We receive a request, check it, pass it to someone else and update a spreadsheet.” That description can hide several different routes.

Use an illustrative process and ask ChatGPT to organise it into people, inputs, decisions and outputs. Then ask it to challenge the gaps. What happens if a request is incomplete? Who can change a decision? How does the next person know the work is ready?

Try: “Map this process in plain English. Separate what I have told you from assumptions. Ask about missing steps, ownership and exceptions before suggesting improvements.”

This can give colleagues something concrete to discuss. It does not prove that the process can be automated or that new software is necessary. Our software development approach starts with understanding the work; a checked process description can help that conversation begin.

A standalone draft may be enough for an occasional task. Repeated work raises another question: how does the right information reach the assistant, and how does an approved result reach the right business record?

Suppose staff regularly prepare follow-up tasks from reviewed notes. Copying between several systems may become the main problem. A connected application could prepare proposed tasks beside their source, let a person correct them and record what was approved.

This is an illustrative design, not a claim that your ChatGPT account already has those connections. Any implementation needs clear permissions, a defined review point and a way to deal with incomplete or incorrect output. Start by describing the action a person authorises, rather than giving the assistant an open-ended instruction to “handle everything”.

Axisware can add practical AI assistance where it supports a real workflow and appropriate human review remains in place. The useful starting point is the business task and its boundaries.

Choose a task you already understand. Keep the first trial small enough that you can inspect every result, and agree beforehand what would make it worthwhile.

• Choose one output, such as a comparison or draft procedure.
• Use fictional, public or otherwise approved information.
• Record the source material and the instructions you supplied.
• Ask the assistant to identify uncertainty and missing information.
• Have a knowledgeable person check the result against the sources.
• Compare total effort, including corrections, with your usual method.
• Record whether you would repeat, change or stop the trial.

Check your organisation's rules before sharing business information or connecting another service. Confirm relevant account features, permissions and costs before relying on them. Keep consequential decisions with the people responsible for them.

The most useful ChatGPT experiment is often an ordinary task made easier to understand and review. If you would like to explore one such task, contact the Axisware team to discuss the information, checks and connections a practical first version would need.