ChatGPT & Codex essentials Turn a loose request into a useful result, then use the same habits for coding work. Reviewed 15 November 2026 · Utilising AI Pty Ltd https://www.utilisingai.com/learn/chatgpt-essentials ChatGPT examples are for chat and project work. The developer section applies to Codex with repository access, not an ordinary chat that has not received your code. Independent educational guide. Original examples; documentation-based guidance. Product availability varies. No provider integration or AI request is run by this download. Quick start =========== 1. Choose one outcome: a customer reply, a document comparison or one small bug fix. 2. Attach the relevant source or name the exact file. State your audience, limits and required format. 3. Review the first result against the source. Ask for a specific correction, then keep the reviewed version. Practical tips ============== Define the finish line Describe what you need to receive: for example, a 150-word reply with a clear next step. Include the business facts that change the answer; a long personality prompt is usually less useful than a clear deliverable. https://learn.chatgpt.com/docs/prompting Give files a job Tell ChatGPT which attachment is the source of truth and what to extract from it. For a visual problem, supply a screenshot and point out the part that needs attention. Ask it to identify missing inputs rather than silently fill them in. https://learn.chatgpt.com/docs/use-chatgpt Use a project for recurring work Keep related chats, source files and project instructions together. Start a separate chat for a distinct outcome. Uploaded project sources are different from granting access to a local folder: make sure the needed material is actually attached or connected. https://learn.chatgpt.com/docs/projects Request fresh evidence when facts can change Ask for web research and direct supporting links for current product features or availability. Open the cited page yourself before relying on the answer. Search availability can depend on your workspace; if it cannot search, supply the source instead. https://learn.chatgpt.com/docs/web-search Developer notes =============== Give Codex a concise project brief In a repository, AGENTS.md can carry setup commands, verification steps and project-specific conventions. More local instructions can override broader ones. Ask which instructions loaded if behavior is surprising. Keep credentials out of these files. https://learn.chatgpt.com/docs/agent-configuration/agents-md Send a reproducible bug, not just “fix it” Include the exact error, expected behavior, reproduction steps and relevant file paths. Our working pattern is one bounded change followed by a test that would catch the failure. Request a diff and actual check results before merging. https://learn.chatgpt.com/docs/agent-configuration/agents-md Separate explanation from execution If you want diagnosis only, say so. For a larger change, first request the proposed approach and affected files. A normal ChatGPT answer with code is a suggestion; it has not changed your website or run tests unless connected tools actually performed those actions. https://learn.chatgpt.com/docs/use-chatgpt Original examples to adapt ========================== Business brief → usable draft Using the attached service sheet as the only source for our offer, draft a 150-word reply to a customer asking what we do. Audience: first-time small-business owner. Include one next step. Mark any missing price or availability detail as [confirm]. Keep it as a draft for review. Replace the attachment and audience with your own. This is an original Utilising AI example. https://learn.chatgpt.com/docs/prompting Explain before editing Explain how [feature] works in this repository. Identify the entry point, key files, data flow and existing tests. Point to the relevant file paths. Do not edit anything in this step. Use in Codex after opening the correct project folder. https://learn.chatgpt.com/docs/agent-configuration/agents-md Small, verifiable bug fix Reproduce this bug: [steps and exact error]. Expected behavior: [result]. Make the smallest suitable fix using the existing project patterns. Run the relevant checks and report their actual results, the files changed and anything you could not verify. Replace every bracketed field. Work on a branch or backed-up copy. https://learn.chatgpt.com/docs/prompting Check the result ================ - Can you trace every business claim to the supplied source? - For code, did the intended check actually run and pass? - Does the result solve the original task without unrelated changes? Common mistakes =============== - Assuming a chat can see files that were never supplied. - Treating a plausible code snippet as a tested deployment. - Uploading real credentials or customer records when a redacted example would work. Sources ======= OpenAI · Prompting · Official documentation https://learn.chatgpt.com/docs/prompting OpenAI · Projects and chats · Official documentation https://learn.chatgpt.com/docs/projects OpenAI · AGENTS.md instructions · Official documentation https://learn.chatgpt.com/docs/agent-configuration/agents-md OpenAI · Web search · Official documentation https://learn.chatgpt.com/docs/web-search OpenAI · Use ChatGPT · Official documentation https://learn.chatgpt.com/docs/use-chatgpt