Knowledge base › Daily practice

A lot of people already use an AI to talk through their day — and the most common verdict is that it's pleasant but goes nowhere. The problem isn't the model, it's the question: ask for advice and you get advice; ask for reassurance and you get reassurance. Here are 30 prompts that do something else — they ask back — plus the three things a chat will never know about you, however good your prompt is.

The four rules of a good self-reflection prompt

1. Give it a role and a prohibition. "Don't reassure me and don't give advice. Reply with a single question." The prohibition matters more than the role — without it the model comforts you by default.
2. Give it material. Don't ask in the abstract. Paste your own sentences from the past week and have it question those. An AI is only as precise as the real material it gets.
3. Ask for one question. Ask for a list and you'll get a list you never act on. One good question you will actually work with.
4. Ask for the counter-argument. "What suggests I'm wrong?" That single sentence improves the conversation more than any elaborate framework.

If you can't tell whether a prompt is good, ask yourself: will it be uncomfortable to answer? If not, it probably won't move anything.

1–5. Morning start

1. "This one sentence describes how I woke up: [—]. Don't reassure me. Ask one question that helps me understand what's underneath it."
2. "I have three things to do today: [—]. Which am I likely to put off, and what would you ask me about it?"
3. "Put into a question what it would mean to do today well — but not as a productivity checklist."
4. "Yesterday I wrote this about myself: [—]. What is the one thing I could decide differently today?"
5. "Ask me one question about what I'm avoiding today."

6–10. Evening close

1. "This is what happened today: [—]. Don't summarise it. Ask about the detail I skipped over fastest."
2. "What did I say yes to today that I wanted to say no to? Help me name what it cost."
3. "When did I feel most like myself today? Ask what this had in common with a similar moment last week: [—]."
4. "In one sentence, my most uncomfortable minute today: [—]. What question should I be asking myself about it?"
5. "If I carried only one thing from today into tomorrow, what should it be — and why that one?"

11–15. When you're stuck

1. "I've been writing the same thing for weeks: [—]. Don't empathise. Show me where I contradict myself in this text."
2. "Here's my problem: [—]. Reframe it three ways, none of them self-blaming."
3. "What assumption in this sentence of mine have I never actually checked: [—]?"
4. "If my closest friend told me exactly this, what would I ask them first?"
5. "What is the smallest step that would make this 5% better tomorrow — and what does it cost?"

16–20. Before a decision

1. "I'm considering this decision: [—]. Don't argue for or against. Ask about what I don't know but am acting as if I did."
2. "What do I expect from this decision, specifically and with a date? Help me turn that into a measurable sentence."
3. "What state am I in right now, and how is it distorting this decision?"
4. "If this decision looks wrong in six months, what will the most likely reason be? One sentence."
5. "What would I decide if nobody's opinion mattered — and what's holding me back from that?"

21–25. Relationships

1. "This conversation unsettled me: [—]. Ask me which exact sentence hurt, and why that one."
2. "Who did I think about most today? Help me phrase, as a question, what I actually want from them."
3. "Where do I struggle to set a boundary — and what is the sentence I should say but usually don't?"
4. "In this conflict [—], what might the other person be right about? Don't defend me."
5. "What am I asking for from someone who can't give it?"

26–30. Weekly review

1. "Here are my seven entries from this week: [—]. What single word or theme comes up more than once? Just name it, no explanation."
2. "Based on my week, where did most of my energy go — and does that match what I say is important?"
3. "Which of my sentences sounds least like me and most like an expectation?"
4. "What happened this week that also happened last week? Ask what triggers it."
5. "What is the one thing I'll do differently next week — and how will I know it worked?"

The three things a chat can't do

These prompts genuinely work, and you can start with them tomorrow. But it's worth being honest about where the limit is — not the model's limit, but the limit of chat as a format:

1. It doesn't remember your months. An assistant remembering a few facts about you still doesn't know you wrote this same sentence in March and again in June. And the pattern lives exactly there: between the weeks.
2. It holds you to nothing. At the end of the conversation there's no goal, no daily step, and nobody asking on Sunday what came of it. You can talk through a hard evening; you can't talk your way into a habit.
3. It can't see your other signals. Your sleep, your habit streaks, your goal progress, your spending patterns. The explanation often sits at their intersection, not in the text.

Privacy: before you type anything

A self-reflection prompt is only worth something if you're honest in it — which makes this the most sensitive text you'll ever type. Check four things with your provider: are models trained on your conversations, how long are they stored, who can access them, and can you delete them. In general-purpose assistants, history and training are often on by default but can be turned off. We walk through it here: 7 questions for any app.

How to use this list

Not in order. Pick one from the group that fits today and give it five minutes. If a prompt moves nothing, move on — but if your first reaction is "I don't feel like answering that right now", that's the one to work with.

And if you notice you've been using the same prompt for weeks with the same answer, that isn't the list's fault: it's the signal that what's missing isn't the question, it's what comes after the answer.

Frequently asked questions

What makes a good self-reflection prompt?

One that forbids comfort and advice, gets material from your own words, asks for a single question back, and asks for the counter-argument.

Why does the AI agree with everything?

Models lean towards agreement. You have to request resistance and contradiction-spotting explicitly.

Is a chatbot enough for self-knowledge?

For a conversation yes, for a process no: no memory of your months, no accountability, no view of your other signals.

Is it safe to type personal things in?

Provider-dependent. Check training, retention, access and deletion. We have a checklist.

How much time a day?

Five to ten minutes. One good question beats twenty worked through.

How is this different from a system?

A prompt gives the question; a system gives the question, the memory and the consequence — a goal, a daily step, a weekly review.