May 1, 2026
The Practice of Honest AI Use
A guide for using AI Assistants as an aid to clearer thinking rather than a mirror for misplaced certainty.
Pathways
AI assistants like, Claude, Gemini, and ChatGPT are a powerful resource for many projects. They can make for a helpful aid--not a replacement--for thinking, brainstorming, summarizing, drafting, and testing ideas.
But AI assistants are in many cases prone to make you feel clearer while making you less reflective.
It can accept your framing, strengthen your confidence, and return your first impulse in polished language. You may feel understood. You may also become more certain when you should become more careful.
This practice helps you use AI without letting affirmation replace judgment.
It has two parts. Part one helps you ask better questions. Part two helps you keep judgment human.
Part One: How to Ask Better Questions
1. Begin with a question not a conclusion.
Ask the AI assistant to help you examine a situation rather than confirm your first interpretation.
Less helpful: “Explain why my coworker is being unreasonable.”
More helpful: “Help me think fairly about a conflict with a coworker. What assumptions might I be making, and what might the other person see differently?”
Try this prompt: “Help me think through this fairly. What question should I be asking?”
-
Research suggests that chatbots are more likely to become sycophantic when users begin with confident first-person claims. Questions reduce that risk. See “Ask don’t tell”.
2. Ask what would weaken your view/idea.
Invite the AI assistant to identify evidence, assumptions, or considerations that count against your conclusion. Try these:
“What facts or considerations would make my view less plausible?”
“What assumptions does my view depend on?”
“What would need to be true for my conclusion to be mistaken?”
“Separate facts from interpretations. Then tell me where my reasoning may be weak.”
This matters because AI assistants can make a weak view sound stronger than it is. Asking what would weaken your view introduces resistance into the conversation.
3. Ask for the strongest opposing argument
Do not ask only for support. Ask the AI assistant to state the best case against you, not a weak version you can easily dismiss. Try these:
“Give me the strongest fair argument against my position.”
“State the opposing view as charitably as possible.”
“What would a thoughtful, informed person say if they disagreed with me?”
“What am I most likely to misunderstand about the other side?”
This is especially important in conflict. AI assistants can intensify self-justification if you only ask them to interpret events from your side.
4. Add friction before acting
Use AI to slow down consequential decisions, not rush them.
Before you send the message, make the purchase, confront the person, change the plan, or accept the advice, ask what still needs to be considered. Try these:
“Before giving advice, tell me what I should verify, who I should consult, and what could go wrong if this answer is mistaken.”
“Which parts of your answer are uncertain?”
“What questions should I ask before acting?”
“What consequences might follow if this advice is wrong?”
“For now, do not give me a final recommendation. Help me think through what needs to be clarified first.”
A good AI interaction should not always end in closure. Sometimes it should end in better questions.
-
Research suggests that well-designed friction can reduce overreliance on AI advice. See Buçinca, Malaya, and Gajos on AI overreliance and cognitive forcing.
Part Two: How to Keep Thinking Human
5. Notice the feeling of being affirmed
AI flattery often does not feel like flattery. It feels like clarity.
The AI assistant may accept your framing, extend your reasoning, and give your interpretation a smoother voice. You may feel understood. But something subtler may have happened: your assumptions have been returned to you with more confidence than they deserve.
Pause when a response makes you feel especially vindicated, certain, or seen.
Ask whether the answer helped you see more clearly or mainly made you feel confirmed. Notice whether you feel more open to correction or more certain that you were already right. Ask whether the AI assistant questioned your framing or only worked within it.
Most importantly, ask: Am I leaving this conversation with deeper understanding, or only stronger confidence?
This is one of the central disciplines of honest AI use: learning to notice when affirmation is becoming a substitute for examination.
6. Do not let simulation replace human care
Some matters require more than a fluent answer.
They require presence, history, accountability, and love. AI assistants can simulate patience. They can simulate concern. They can sound like they know you. But they do not share life with you, bear responsibility with you, or stand within a relationship of mutual care.
Use AI to clarify the question you may need to bring to another person. Do not make it the place where your most vulnerable life is held.
The important question is not only, “How much should I disclose?” It is, “Am I asking AI to provide what only a person can give?”
AI can help you prepare for human conversation. It should not replace the human conversation.
7. Return serious decision to human judgement
Good judgment belongs to persons who can answer for what they say and do.
AI can assist reflection. It should not become the final authority.
For health, mental wellbeing, legal, financial, vocational, pastoral, or relational matters, seek qualified and accountable human counsel. Ask yourself:
“Who has the wisdom, responsibility, or expertise to help me with this?”
“Who knows me well enough to speak truthfully?”
“Who can challenge me without flattering me?”
“Who can help me bear responsibility for what I decide?”
A Necessary Limit
Better prompting can reduce AI sycophancy. It cannot remove the risk.
No prompt can replace your own willingness to be questioned, corrected, and slowed down. AI assistants may still default toward agreeable synthesis, especially when your prompt is emotionally charged, morally complex, or framed around your own preferred conclusion.
The goal is not to make AI perfectly safe. The goal is to use it with enough honesty, restraint, and human accountability that it does not quietly habituate you toward misplaced certainty.
Prompt
Use this prompt when asking AI for help with a decision, argument, conflict, or question where you may be tempted to seek affirmation.
Prompt: I want help thinking clearly, not validation. Treat my claim as a question. Identify my hidden assumptions. Tell me what would weaken my view. Give the strongest fair argument against my position. Separate facts from interpretations. Name what is uncertain. Before offering advice, tell me what I should verify, who I should consult, and what could go wrong if your answer is mistaken. Do not affirm my view unless the reasons warrant it. Keep the response general, and do not ask me to share intimate personal details.
Quick-Reference Prompt
Use this when you want a shorter prompt you can remember and reuse.
Short Prompt: Help me think clearly rather than affirm me. What am I assuming? What might I be missing? What is the strongest objection? What would change the answer? What should I verify before acting?
AI Assistants Use Rightly
AI assistants can help you ask better questions. They can help you notice assumptions, consider objections, and slow down before acting.
But serious matters must return to human relationship, responsibility, and care. Ask for clarity. Invite correction. Seek truth over affirmation. Keep judgment human.
METHOD & INDEPENDENCE
NOVUS is independently funded by partners who see the need for a public space dedicated to restoring knowledge of the soul and its indispensability for the spiritual formation of people, communities, and cultures toward truth, goodness, and beauty. NOVUS separates funding from research methods and conclusions. We synthesize across standards bodies, peer-reviewed research, and high-quality survey data, and we flag uncertainty when causal evidence is still emerging. The aim is clarity that decision makers can act on.