Course reading draft · For review. This material is not a complete or graded course.
Plan clearly and hand work over
Companion to Understanding AI usage and the GPS method. These teaching topics are DECIDED; wording and presentation are a review draft. Nothing here changes an account setting.
Prevent avoidable retries
A vague request such as “make my website better” leaves the AI guessing. Before a substantial task, define the page, audience, desired change, source files, constraints, and how you will check success. Good prompts can reduce rework, but do not guarantee lower usage.
Example: “Review this public learning page for first-time visitors. Return the three clearest improvements with evidence. Keep the current public name provisional. Do not edit or publish. Say which parts you could not inspect.”
Use two conversations when it helps you stay organized:
Planning chat: clarify the goal, compare options, check assumptions, and prepare a small handoff.
Work chat or work bot: receive that approved handoff, carry out the bounded task with the available tools, and return the result plus checks and unresolved issues.
Keep one work owner for each editable file. Bring the result back to the planning chat before giving a conflicting instruction elsewhere. Separate chats can keep discussion focused, but they are not automatically isolated from account memory and do not reset usage limits. Copy the necessary approved context yourself; do not assume the next tool knows it.
Copyable handoff
Goal: [one result]
Approved inputs: [exact files or links, version, facts]
Boundaries: [privacy, what must stay unchanged, actions allowed, actions needing approval]
Output: [format and destination]
Check: [how to test success and what cannot be verified]
Stop: [completion condition, time or usage ceiling if known]
Return: [finished artifact, evidence, changes, blockers, and decisions for me]
For a revision, send the specific correction and what must remain unchanged. Do not restart the entire job unless the foundation is wrong.
Choose a helper for the job
- ChatGPT planning chat: explain, structure a brief, compare returned answers, and identify what needs verification
- dot: ask for help progressing an ongoing goal, preparing work, and reporting results. State the project, approved sources, and action boundaries. Ask which files, apps, or computer it can currently use; access must be checked
- Gemini, Perplexity, and Spark: use the GPS prompts for selected extra perspectives. Taylor sends the prompts and returns the answers. Recommend Perplexity's mode and explain the choice every time
- Grok: an optional additional reviewer when requested or clearly useful. Give it the same privacy-minimized brief and ask for evidence or a counterargument. It does not automatically replace or expand the three defined GPS prompts
These are suggested assignments, not exclusive provider abilities. Tools may use related models or shared allowances. Different names and matching answers do not prove independence or truth. Choose only the reviews that add value.
Personalize with a purpose
Use personalization for stable, non-sensitive preferences that genuinely help: a preferred reading level, concise explanations, accessible formatting, or Taylor-only public attribution. Keep the current project brief as the source for specific decisions. Memory is useful context, not a complete or authoritative record. Memory guidance
Custom instructions are guidance you deliberately provide; memory can draw on relevant information across available sources. Review what is remembered and correct outdated details. Personalization can save repetition, but irrelevant history can also distract a task. Do not place sensitive legal, health, identity, or client information into routine personalization. Custom instructions
For a clean-context task, current ChatGPT guidance offers Unpersonalized when starting a Temporary Chat. This excludes memory, custom instructions, and plugins for that chat. A Personalized temporary chat may use them. Neither option creates or updates memories while temporary. A temporary chat is not used for model improvement while it remains temporary, but a copy may be retained for safety. Saving it converts it into a regular chat subject to the account's settings. Temporary therefore does not mean zero retention or permission to share confidential material. Check the current controls before sending. Temporary Chat guidance
Memory, chat history, and model-training controls are different choices. Turning off model improvement does not automatically remove chat history; privacy settings do not replace permission to share someone else's information. Data controls
Human Check: can you explain what context this helper needs, what it may do, and what it should leave out? If yes, send the short handoff. If not, resolve that first.
Official ChatGPT guidance checked 1 October 2026. Controls and availability can vary by account and change over time. This is a teaching example, not a report of Taylor's current settings.