Back to AI learning

Course reading draft · For review. This material is not a complete or graded course.

The GPS method

Review draft for the AI learning plan. The human-directed method is decided; the lesson title, final wording, and brand treatment are provisional.

What you will learn

Use a small team of AI tools to examine an important task from different angles, then make your own decision. Finish with one checked recommendation, a short list of disagreements, and a clear record of what still needs evidence.

When Taylor says “use the GPS method,” ChatGPT prepares three prompts to copy into Gemini, Perplexity, and Spark. Taylor sends them and brings the answers back. ChatGPT then consolidates and verifies the findings. No tool automatically sends private material to another service.

Grok is an optional extra reviewer when requested or useful for a specific unresolved question. It does not automatically change the three-prompt definition of GPS. Use the planning and personalization checklist to keep handoffs clean and avoid passing unnecessary private context between tools.

The workflow

Start with Understanding AI usage. Its phone-data and original energy-meter examples explain why task size, model choice, tools, context, and reset windows matter. Check the relevant provider's current allowance before a substantial run. A new chat or different mode does not automatically create a fresh allowance.

Shared brief → Independent responses → Compare disagreements → Verify original sources → Human decision → Stop

  1. Define one decision and what a good result must contain. Set a time limit, a usage limit, and a stopping rule
  2. Prepare a short common brief: goal, audience, approved facts, known constraints, public sources, and the exact output format. Remove names, credentials, confidential files, and unnecessary personal details
  3. Assign different jobs by the capability and access available today. A tool may draft, challenge, research, or prioritize. The examples below are role assignments, not claims that one provider uniquely owns a capability
  4. Send the prompts yourself. For independent reviews, do not show one tool the others' conclusions first. If a tool cannot open a source, it must say so rather than imply it inspected it
  5. Bring the responses back to ChatGPT. Separate agreement, disagreement, assumptions, unsupported claims, and unanswered questions. Three models agreeing does not establish truth
  6. Open the original sources for consequential facts. Verify that a source is current, says what the response claims, and applies to this case. Check calculations and actual product behavior separately
  7. Decide what to use, what to test, and what to discard. You own the decision and any action taken from it

Use ChatGPT Pro, Gemini's careful reasoning mode, Perplexity Pro, and Spark when available and useful. A simple task may need only one tool. A GPS request generates all three prompts, but you can choose which to run. Do not buy subscriptions or burn through deep-research runs merely to complete a ritual.

Example shared brief

We are reviewing a fictional public learning website before a wider launch. Our decision is which three improvements deserve attention first. Review only the supplied public pages and non-sensitive test evidence. The audience includes beginners and experienced learners. Human control, clear explanations, accessibility, privacy, and honest claims matter. Do not sign in, submit forms, contact anyone, change the site, or infer private business facts. Return observed facts separately from inferences and proposed tests. A page you cannot inspect is a limitation, not evidence of a defect.

Prompt for Gemini

Mode: use the most careful reasoning or extended-thinking mode available in your account. This is a thoughtful product review; no particular feature name or access level is assumed.

Paste the shared brief and approved public URLs here. Your role is independent product challenger. Examine the learner journey, confusing choices, accessibility risks visible in the supplied evidence, and ways the product could encourage over-reliance. Return at most five findings. For each, give the exact evidence, why it matters, a small proposed improvement, confidence, and a test that could prove you wrong. Distinguish observed behavior from speculation. End with the three highest-priority changes and what you could not verify.

Prompt for Perplexity

Recommended mode: Deep Research for this multi-source website and standards review. Reason: the task needs source comparison and current supporting references. For a single finding or a brief follow-up, use ordinary chat instead. State the mode recommendation each time a new prompt is prepared; verify what is available in the current account.

Paste the shared brief and approved public URLs here. Your role is evidence reviewer. Inspect the supplied public information and locate primary sources relevant to the site's public claims, accessibility guidance, and any stated content-use permissions. Provide exact source links and relevant dates. Separate what a source establishes from your interpretation. Do not claim legal compliance, certification, ownership, or permission from an absence of evidence. Return the three most important evidence gaps and the smallest next check for each. State any sources or pages you could not inspect.

Prompt for Spark

Mode: use the normal available analysis/chat mode for a bounded prioritization pass. If Spark offers different modes, choose the least resource-intensive one that can follow this brief and explain any limitation.

Paste the shared brief and approved public URLs or findings here. Your role is implementation planner. Work independently from the other reviewers' conclusions when possible. Suggest small, reversible improvements and group each as preparation, local test, human decision, or publication requiring approval. Choose no more than three active priorities. For each, give impact, effort, dependency, verification, and a stopping condition. Do not claim you inspected a page or ran a test unless you actually did.

Bring the answers back

Ask ChatGPT: “Consolidate these GPS responses. Identify the strongest evidence, disagreements, unsupported claims, and missing tests. Verify consequential facts against the original sources. Recommend no more than three next actions, separate my decisions from work you can prepare, and do not publish or share anything.”

Human Check and stopping rule

Stop when the decision criteria are met, the important evidence is checked, material disagreements are resolved or clearly recorded, and the next action has an owner. Stop earlier if the next round is repeating the same points. Escalate to a qualified person where the stakes require it. Extra model agreement is not a reason to keep spending time or usage.

Record the tools and modes used, run counts, elapsed time, direct cost if known, useful new findings, corrections caught, and rework needed. Compare against a similar single-tool task before claiming time or usage savings. Savings are an objective to measure, not a promised outcome.

For each prepared prompt, add: “Mode and reason: [choice]. Usage check: [relevant allowance and reset, or unknown]. Expected benefit: [what this review adds]. Stop when: [clear completion test].” A percentage or credit in one service is not an exchangeable unit in another. Some features share a pool. Do not add provider balances together or infer exact costs from message counts. Keep any private account screenshots or balances out of a public lesson.