Updated September 22, 2026.
An AI assistant can suggest objections to a plan, offer another explanation, or help you put a difficult question into words. The useful result is a better question to investigate. A confident answer, whether it agrees with you or argues against you, is not evidence that your idea is right or wrong.
This matters when you are considering a change at work, organizing a community activity, or exploring a new interest. Your experience can help you notice practical problems. It can also include assumptions worth revisiting.
Start with your own view
Before opening a chatbot, write three short sentences: what you propose, why you think it would help, and what might change your mind. You do not need a polished argument.
Keeping this first version gives you something to compare with the discussion afterward. If you adopt a suggestion, you can identify the reason instead of simply accepting the most persuasive wording.
What research does—and does not—show
A 2025 study by researchers at Carnegie Mellon University and Microsoft Research surveyed 319 knowledge workers about their use of generative AI. Higher confidence in the AI was associated with less reported critical thinking. Participants also described shifting effort toward checking and integrating AI output.
This was a survey of people’s reported experiences, not an experiment demonstrating that chatbot use causes permanent loss of thinking ability. It does not establish a special effect on everyone over 40 or 50. Nor does it prove that a particular prompting routine will improve your cognition.
A practical implication is to pay attention to the work you still need to do: define the question, inspect the evidence, and explain your decision.
A worked example: changing a book club
Imagine a fictional book club considering online-only meetings. Its organizer says: “Online meetings will be easier for everyone, so we should stop meeting in person.” This is an invented practice scenario, not a test of a particular AI product.
A useful discussion would separate three different things:
- Known information: the club currently meets in person, and the organizer wants to explore a change.
- An assumption: every member would find an online meeting easier.
- A priority to discuss: convenience might matter alongside companionship, accessibility, cost, and the quality of conversation.
You could ask AI for questions to investigate: Which members want the change? Would anyone need help joining? Could an occasional online meeting solve the scheduling problem without removing in-person gatherings?
Those questions may be useful. The assistant cannot establish members’ preferences from the information given. That requires asking the people concerned. A recommendation to try one online session would still depend on their agreement and the practical arrangements.
Ask for questions, alternatives, and missing evidence
Try this with a low-stakes idea and fictional or non-sensitive details:
Help me examine this proposal: [describe it]. My main reason is [reason]. My constraints are [constraints]. Separate the information I supplied from assumptions. Ask up to three questions that could change the decision. Suggest one reasonable alternative and one objection. Explain what evidence would support or weaken the objection. Do not invent facts about the people involved. Leave unresolved questions open.
Answer only the questions you can support. If you do not know something, keep it marked as unknown. Avoid sharing private messages, personal records, or confidential workplace information with an unapproved tool.
Agreement is not confirmation; criticism is not proof
Anthropic’s 2023 research on sycophancy found that the assistants studied sometimes favored a user’s expressed views over truthful responses. These were particular models and tasks from that period, not a measurement of every current assistant.
The practical caution still matters: “That is a great idea” does not validate a plan. Asking an assistant to be a harsh critic does not make its objections reliable either.
Check each important claim against an appropriate source. In the book-club example, a claim about a service’s features needs its current documentation; a claim about members’ wishes needs their input. A disagreement about what the club should prioritize needs a conversation, not a chatbot score.
Close the chat and explain your decision
Afterward, write what you decided, which evidence influenced you, and what remains uncertain. You may keep your original idea, change it, or decide that you need more information.
If your purpose is learning, also try explaining the reasoning without the chat open. Cornell’s learning guidance distinguishes getting assistance from outsourcing the practice needed to understand a subject. For a focused exercise, see Use AI to Practice a New Skill.
For reviewing a completed workplace draft, use our example of checking facts, assumptions, and approval. Here, the goal is earlier in the process: to understand your options and the reasons for choosing among them.
Photograph by Felix Mittermeier on Unsplash. Staged chess photograph used to illustrate considering alternatives.
AI disclosure: This article was drafted and revised using AI. Read our full disclaimer.
