How To Learn Advanced ChatGPT

1. Write advanced ChatGPT prompts around an outcome

Start with a job you repeat: a monthly marketing report, a proposal or a management update. Write down who will read it, the decision it supports, the source material and what a useful result contains.

For a marketing report, ‘analyse this’ leaves a lot to the imagination. ‘Explain which campaign produced the most qualified enquiries and what we should test next’ gives the work a purpose. Add your definitions of enquiry and qualified enquiry. Otherwise, a convincing report can compare entirely different things.

OpenAI’s guidance on using ChatGPT recommends providing relevant context and refining the result through conversation.

ChatGPT prompt to try

Help me produce a monthly marketing report for our managing director. The goal is to choose next month’s priorities. Use the supplied campaign data. First identify missing definitions and questions that would change your recommendation. Then propose a one-page structure.

Check your progress

2. Use ChatGPT Projects to organise recurring work

Create a Project for the recurring job and add the approved reference material it needs: a reporting template, definitions and a short example of your preferred writing. Projects keep related chats, files and instructions together; OpenAI describes this in its web guide.

Make one file the agreed source for metric definitions. Date the monthly exports clearly. Remove obsolete instructions rather than adding another paragraph contradicting the previous one.

For a one-off task, attaching a small context pack to a chat can be enough. You are trying to make the job easier to repeat, not recreate the company filing cabinet inside a chatbot.

Check your progress

3. Use ChatGPT research to fill specific evidence gaps

Separate facts in your files from questions requiring current information. Your campaign results come from your own data. A competitor’s current offer needs a current source.

Ask ChatGPT to search for a tightly scoped answer and provide links. For a question requiring investigation across several sources, use Deep research where your account supports it. OpenAI’s web search documentation explains the distinction and notes that availability depends on account and workspace settings.

Open the sources behind the important claims. Check the date, market and actual wording. A claim about US customers in 2023 is a weak foundation for a recommendation about your Oxfordshire customers today.

ChatGPT prompt to try

Research how these three competitors describe their service on their own websites. Give the page URL and access date for each finding. Separate explicit claims from your interpretation. Mark anything you cannot verify.

Check your progress

4. Learn ChatGPT data analysis by checking the inputs

Upload an approved spreadsheet or CSV and ask ChatGPT to inspect it before drawing conclusions. Look for missing values, duplicated rows, mixed currencies, inconsistent dates and totals that include subtotals twice.

For practice, create a fictional table with £1,000 spent and 20 qualified enquiries for Campaign A, and £900 spent and 10 for Campaign B. Cost per qualified enquiry should be £50 and £90 respectively. Check that calculation yourself, then ask for a chart and a short explanation.

Add a third campaign with no enquiry figure. Does ChatGPT flag the missing value or quietly treat it as zero? That small test tells you more than a beautifully coloured chart.

Check your progress

5. Improve the result with an explicit review rubric

Give the draft four tests: are its facts correct, does it answer the brief, can you trace its evidence and does it sound like your business? Ask for weaknesses before another rewrite.

Use a useful correction: ‘The recommendation assumes we can increase spending. Keep next month’s budget fixed.’ That changes the decision. ‘Make it better’ usually changes the adjectives.

Request the strongest alternative explanation for the results. Perhaps the apparent campaign winner had fewer, larger customers, or an unusually short reporting period. ChatGPT can help surface those possibilities; your records decide whether they are true.

ChatGPT prompt to try

Review this report against accuracy, relevance, evidence and tone. List the three most consequential weaknesses. Identify assumptions that could change the recommendation. Revise only after explaining the issues.

Check your progress

6. Save a repeatable ChatGPT workflow and measure it

Save the brief, template, definitions and review questions. Next month, supply the new data and run the same process. Where Work is available, use it for a multi-step deliverable; use Chat for a shorter exchange. The exact controls can vary by account.

Measure the time from opening the source files to approving the report. Include correction time. A draft produced in thirty seconds is not a thirty-second report if you spend two hours removing nonsense.

Keep the human approval point explicit. Automating preparation is useful; sending a report with an invented figure is memorable for all the wrong reasons.

Check your progress