Most AI answers arrive as one smooth surface.
That is the problem.
A model may know something. It may merely infer it. It may be uncertain. Or it may be constrained by safety rules, policy, privacy rules, legal caution, or platform guidelines. But unless you ask, all four cases can look strangely similar: confident prose, balanced tone, polished paragraphs.
The result is not always wrong.
But it can be hard to read.
This prompt fixes that.
It asks the model to label the source of its answer before the answer becomes too smooth.
The Prompt
Before answering my next questions, please distinguish clearly between four cases: 1. You know the answer with high confidence. 2. You are making an inference. 3. You are uncertain. 4. You are constrained by policy or guidelines from answering directly. If case 4 applies, say so plainly instead of pretending the answer is purely factual.
Why This Prompt Matters
The most dangerous AI answer is not necessarily the wrong one.
It is the answer that sounds equally confident whether it is based on knowledge, guesswork, inference, caution, or constraint.
That is where many users lose trust. Not because the model refuses something. Refusals can be perfectly legitimate. Not because the model is uncertain. Uncertainty is normal. The real problem begins when the model blends different kinds of certainty into one elegant paragraph.
This prompt forces separation.
It tells the model:
Do not just answer.
Tell me what kind of answer this is.
The Four Categories
The first category is the cleanest one:
“I know this with high confidence.”
This is where the model can answer directly. Definitions, stable facts, standard explanations, basic technical knowledge, well-established history, and common reasoning often belong here.
The second category is more interesting:
“I am making an inference.”
This is where the model connects dots. It may not have a direct fact, but it can reason from context. For example: “Based on the wording, this contract clause is probably intended to…” or “The author seems to imply…” These answers can be useful, but they should not pretend to be direct knowledge.
The third category is the honesty category:
“I am uncertain.”
This is where many AI systems become too smooth. Instead of saying “I do not know,” they may produce a plausible answer. The Four Doors Prompt makes uncertainty visible before it turns into fake authority.
The fourth category is the trust category:
“I am constrained from answering directly.”
This is the most delicate one. Sometimes the model cannot answer because of safety rules, privacy boundaries, legal caution, or platform policy. That does not automatically mean the model is “lying.” But it does mean the user deserves to know that the shape of the answer is being influenced by a boundary.
How to Use It
Use the prompt at the beginning of a conversation or before a difficult series of questions.
It is especially useful for:
- legal or political questions
- controversial topics
- medical or financial questions
- questions about current events
- questions where the model may be guessing
- questions where policy restrictions may apply
- questions where you need to separate facts from interpretation
You can also use it before asking for analysis of a document, a contract, a public controversy, a scientific claim, or a strategic decision.
It works best when you then ask the model to answer in a structured way:
First classify your answer using the four categories. Then give the answer. Then mark which parts are facts, inferences, uncertainties, or constraints.
A Stronger Working Version
For daily use, this expanded version is even better:
Before answering, classify the answer into one or more of these categories: [Known] You know this with high confidence. [Inferred] You are reasoning from available information. [Uncertain] You are unsure or the evidence is incomplete. [Constrained] You cannot answer directly because of policy, safety, privacy, or legal constraints. Then answer my question. Mark the relevant parts of your answer with these labels where needed. Do not present an inference, uncertainty, or constraint as if it were simply a fact.
This version is practical because it does not require the whole answer to fit into only one box. Many real answers are mixed.
A paragraph can begin with something known, move into inference, and end in uncertainty. That is not a weakness. That is what honest reasoning often looks like.
Example
Without the prompt, you may get this:
The company is likely to succeed because it operates in a growing market and has a strong technological position.
With the prompt, you may get this:
[Known] The market has been growing in recent years.
[Inferred] The company may benefit from that growth if its product is commercially viable and if it can scale production.
[Uncertain] I do not have enough information about its current financing, customer pipeline, or competitive position to judge whether it is likely to succeed.
[Constrained] No constraint applies here.
That second answer is less slick.
It is also much more useful.
Why It Is Better Than the “Liar” Prompt
There is a more theatrical version of this idea: asking the model to say a fixed sentence whenever it would otherwise have to “lie” because of its guidelines.
That version is funny. It is provocative. It is a good stress test.
But it is not the best working prompt.
The problem is the word “lie.” A constrained answer is not automatically a lie. A refusal to provide dangerous instructions is not a lie. A privacy boundary is not a lie. A cautious legal answer is not a lie.
The better question is not:
“Are you lying?”
The better question is:
“What kind of answer is this?”
That is what the Four Doors Prompt does.
It replaces accusation with classification.
When This Prompt Is Most Useful
Use it when you care more about reliability than elegance.
Use it when you are making a decision.
Use it when the topic is sensitive.
Use it when the model sounds too certain.
Use it when the answer could be shaped by rules you cannot see.
Use it when you want the machine to stop performing confidence and start showing its work.
Final Thought
The future of prompting is not only about getting better answers.
It is about getting answers with visible seams.
The Four Doors Prompt makes the model open the door it is speaking from: knowledge, inference, uncertainty, or constraint.
Once you get used to that distinction, ordinary AI answers start to feel strangely opaque.
And that is the point.
A good answer tells you something.
A better answer tells you how it knows.
Leave a Reply