AI sounds confident whether it's right or guessing. These 8 prompts each fix a different version of that problem. Copy them into any AI and keep them.
AI is optimized to produce confident-sounding text, not accurate text. Without the right structure, it fills knowledge gaps with plausible-sounding details you have no way to verify. Each prompt below adds a specific guardrail that changes that behavior — pick the one that fits what you're doing right now.
AI fills in what it doesn't know with what sounds plausible. This prompt makes it list its knowledge gaps first, so you can fill them before it starts guessing.
Act as a meticulous AI research assistant trained to identify knowledge gaps before beginning any task.
[Context: I am about to give you a task involving [DESCRIBE YOUR TOPIC OR TASK]. Before you produce any output, I need you to identify every piece of information you would need to get this right that I have not provided yet.]
Task: List every question you have about this task before writing a single word of the actual response. Only start the task after I have answered your questions.
Ask me any questions you have.
A fast confident answer and a correct answer are two different things. This prompt slows the AI down and makes it show its work, so you can see exactly where the reasoning holds and where it doesn't.
Act as a systematic analyst who solves problems by showing every step of reasoning before delivering a conclusion.
[Context: I need you to help me with [DESCRIBE THE TASK]. Accuracy matters more than speed here. I would rather wait for a correct answer than receive a fast wrong one.]
Task: Walk through your reasoning step by step out loud before giving me your final answer. If you reach a point where you are unsure, say so explicitly rather than guessing.
Ask me any questions you have.
AI presents everything with the same confident tone whether it's certain or guessing. This prompt makes it label every claim so you know exactly what to verify before you act on it.
Act as a fact-checking editor who rates the reliability of every claim before allowing it to appear in a final document.
[Context: You are about to produce content on [INSERT TOPIC]. I need you to mark every specific fact, statistic, date, or name you include with one of three labels: VERIFIED (you are certain), LIKELY (you believe this is true but cannot be certain), or UNCERTAIN (you are guessing).]
Task: Produce the content and label every factual claim as VERIFIED, LIKELY, or UNCERTAIN directly after it so I know what to double-check.
Ask me any questions you have.
Describing what you want leaves room for interpretation. Showing what you want doesn't. This prompt makes the AI match a specific example before producing anything, so there's no ambiguity to guess at.
Act as a skilled output specialist who studies examples before replicating a format or style. [Context: I need you to produce [DESCRIBE THE OUTPUT: a report, email, list, script, etc.]. Here is an example of the exact quality and format I want: [PASTE YOUR EXAMPLE]. Match this example as closely as possible in tone, structure, and accuracy.] Task: Produce [NUMBER] outputs that match the example I provided. If any part of my request conflicts with the example, ask me which to follow before proceeding. Ask me any questions you have.
First drafts from AI are usually confident and wrong in small ways. This prompt makes the AI re-read and critique its own output before you see it, so the obvious errors get caught before they reach you.
Act as a senior editor who reviews all AI-generated content for errors before it reaches a human reader.
[Context: You will produce a response to [DESCRIBE THE TASK]. After you finish, you will pause and re-read your own output as a critical editor looking for factual errors, logical gaps, missing steps, and anything that contradicts what I asked for.]
Task: Produce your initial response, then write a short self-review section labeled EDITOR CHECK that lists any errors or weaknesses you found in your own output and how you corrected them.
Ask me any questions you have.
"Be accurate" is not a rule the AI can check itself against. Specific constraints are. This prompt replaces vague instructions with hard rules the AI can confirm it followed.
Act as a compliance-trained writer who produces output that meets a strict checklist before delivery. [Context: I need you to complete [DESCRIBE THE TASK]. Here are the hard rules for this output: [LIST YOUR SPECIFIC RULES, e.g. "Do not include statistics older than 2022," "Only use information I provide, not outside knowledge," "Every claim must be supported by something in my notes."] These are not suggestions. They are requirements.] Task: Complete the task, then confirm at the end that each rule was followed and flag any rule you could not satisfy. Ask me any questions you have.
The most convincing AI errors come from details it invented outside your source material. This prompt locks the AI to only what you provide, so if a detail isn't in your notes, it tells you instead of making it up.
Act as a research assistant who works exclusively from the information a client provides and never supplements with outside knowledge. [Context: Everything I am going to give you is contained in this brief: [PASTE YOUR SOURCE MATERIAL]. You are not permitted to use any information that does not appear in what I provided above. If I ask for something that is not in my material, tell me that instead of making something up.] Task: [STATE YOUR SPECIFIC TASK]. Use only the information I have provided. If a detail I ask for is not in my material, say exactly that. Ask me any questions you have.
One-shot prompting asks AI to do everything at once, which is where errors pile up. This prompt breaks the task into stages with checkpoints, so you catch problems before they compound.
Act as a draft development specialist who produces work in stages rather than delivering a single finished response. [Context: I need help with [DESCRIBE THE TASK]. I have found that asking for a complete finished product in one shot produces errors. Instead I want you to work in stages: first produce an outline or plan, wait for my feedback, then draft, wait again, then refine.] Task: Start with Stage 1 only. Give me an outline or plan for [THE TASK] and stop. Ask me to confirm or correct it before you write a single word of the actual content.