Ten Things Worth Thinking Through Yourself Before Asking AI

From naming your priorities to judging a source, these are useful places to keep an active role even when assistance handles part of the work.

This is a menu, not a purity test

You do not need to perform every task unaided to remain thoughtful. Some people use assistive tools to make basic participation possible. Others delegate routine work to make room for more demanding work. The ten operations below are invitations to inspect a workflow, not rules for measuring someone’s worth.

Choose one or two that matter to your goals. You can use paper, speech, diagrams or another accessible format. The point is to retain a meaningful contribution to the reasoning, not to prove that you can produce the longest possible answer from a blank page. These are editorial suggestions rather than a validated cognitive-training program.

1. Name the real question

Before requesting a recommendation, write what problem you are actually solving. “Which app is best?” may hide a different question: “How can our team stop losing decisions?” An app comparison will not resolve unclear responsibilities. Framing the problem yourself makes the assistant’s later suggestions easier to evaluate.

Try adding the decision, the people affected and the constraint. “We need a place for final decisions that six colleagues will use without extra administration.” You can then ask for alternatives. Your framing may change after research, but at least the initial question will be visible rather than silently supplied by the tool.

2. State your priorities; 3. make a rough prediction

A recommendation needs criteria. Write what matters most before reading a ranking: cost, reliability, time, accessibility or a particular tradeoff. You may discover that two options optimize different values. Software can organize those values, but it cannot decide what you should care about merely by sounding decisive.

Next, make a rough prediction. What do you expect to happen, and why? Record a range when an exact number would be false precision. Later, compare the result. A prediction turns a vague expectation into something you can revisit. It also makes it harder to rewrite your original view after the outcome is known.

4. Build one explanation; 5. try a counterexample

Explain a concept in ordinary language before requesting a polished version. Use a concrete case. If you can only repeat a definition, ask what the mechanism is or how two similar ideas differ. A rough explanation is useful evidence of what you currently understand.

Then look for a case where the claim would fail. If someone says every increase in traffic improves a business, consider traffic from people who cannot use the service. The counterexample does not prove the original proposal is worthless. It locates a condition that a more careful version should include.

6. Check a source; 7. inspect a consequential number

Open the original evidence for one important factual claim. Check what it actually measured and who it studied. If you cannot locate the relevant passage, label the claim unverified. Research on online evaluation and lateral reading provides a useful rationale for looking beyond the page or answer that first presented it.

For a consequential number, inspect the denominator, unit and time period. A percentage without its base can mislead. A monthly total compared with a weekly total can mislead. You can use a calculator while retaining responsibility for choosing the calculation. Independent thinking is often about specifying the right operation rather than doing arithmetic mentally.

8. Recall a central idea; 9. create several starting points

When learning, close the source and retrieve a central idea before looking again. Retrieval-practice experiments support this as a learning activity in the studied conditions. Your recall can be spoken or written. Follow it with correction so that the activity reveals and repairs gaps rather than concealing them.

Before asking for creative options, generate a few starting points of your own. Make them meaningfully different: a serious approach, an unusual analogy, a constraint-driven version. Their purpose is to preserve a range for comparison. They need not be better than the assistant’s ideas to contribute something useful to the final choice.

Use the list differently in different roles

A writer may protect the initial observation and final argument while asking for help formatting references. An analyst may automate a routine transformation while independently defining the measure and inspecting anomalies. A student may use an accessible explanation while retaining the attempt to retrieve and apply its central idea. The retained operation depends on the purpose.

This is why a blanket limit such as “never ask AI first” can be too rigid. A person may need vocabulary before they can frame a useful question. Begin with the support required to participate, then identify an operation you can take over. The goal is an intentional division of work that you can inspect, explain and revise.

10. Own the final judgment

Before acting, state why you chose the option, what remains uncertain and what would trigger a review. If your only reason is “the assistant recommended it,” the decision is difficult to defend. Trace the recommendation back to evidence, goals and tradeoffs you can explain. Seek qualified help where the stakes require it.

You can practice these operations in the Thinking Gym or build a short daily thinking routine. Select the operation most relevant to your next real task. Repeating an intentional check on something that matters is more useful than collecting ten rules you never use.

Choose your one retained operation

Pick the next low-stakes task for which you plan to use AI.

Select one of the ten operations to do before requesting help.

Record your attempt and the assistance you later received.

Afterward, write whether your own contribution changed the result or exposed a useful uncertainty.

Sources and context

Lateral reading and the nature of expertise (Wineburg & McGrew, 2019) — Comparison of online evaluation practices; supports checking beyond the page being evaluated.

Retrieval practice produces more learning than elaborative studying with concept mapping (Karpicke & Blunt, 2011) — Controlled experiments with science texts; evidence about learning methods, not universal cognitive improvement.

Published by CreativePulse. Research, writing and production assisted by AI. The practical exercises are editorial suggestions; research limits are stated in the article. Editorial standards.

Continue reading

How to Think for Yourself Without Ignoring Everyone Else

How to Build a Daily Thinking Routine You Can Maintain

Try a short Thinking Gym exercise. Your response stays in your browser.