How to separate facts from interpretations

When working with AI, an organized summary can look like a finished conclusion. Facts, interpretations, assumptions, judgments and recommendations can blend into one paragraph. This practice helps slow that movement down before a first reading of the situation starts acting like a decision.

The goal is a map of the material before a decision, not a decision made by the model. AI can help separate layers of information, but the person remains responsible for judgment, context and the next step. Emotions are also important information about the situation, but they are not the same as a fact that can be pointed to, quoted or checked.

This practice is about the material: an email, note, report, transcript or model answer. If you mainly want to check your own first reading of a situation, the assumptions practice is a better fit.

When this helps

Before a decision based on emails, notes, reports or conversations.
When the material is messy, emotional or cognitively demanding.
When you want to reduce the certainty of a first interpretation without dismissing emotion.
When you need questions for a team, client or expert.
When an AI answer contains conclusions but does not show how it got there.

What to ask the model

Ask the model for columns or layered categories. The key distinction is: what can be pointed to in the material, what is an interpretation, what is an assumption, what is a judgment and what question has to be asked before deciding.

Help me organize the material below before I make a decision.

Do not decide for me. Do not diagnose people and do not guess hidden intentions.

Divide the material into 6 columns:

1. Fact: what can be quoted, observed or pointed to in the material.
2. Interpretation: what someone may think that fact means.
3. Assumption: what is being added without enough confirmation.
4. Judgment: where a person, team, situation or risk is being evaluated.
5. Alternative reading: what other interpretation could also fit the same fact.
6. Missing question: what needs to be known before the next step.

Then identify which conclusions would be premature based on this material. If you suggest possible directions of action, describe them as options, not as the decision. Do not ignore information that does not fit the most convenient conclusion. If emotion is visible in the material, name it as a signal to check, not as proof of intention.

Material:
[paste material]

What to check yourself

Whether a “fact” can really be pointed to, quoted or checked.
Whether the same information could support more than one interpretation.
Whether a judgment has been smuggled in as a neutral description.
Whether assumptions are coming mainly from tension, fatigue or previous history.
Whether the person, team or source affected by the situation is missing from the material.
Whether emotion helps reveal the stakes, but does not replace checking the situation.
Whether the exercise turns uncertainty into a table that only looks impartial.

What can go wrong

The model may wrongly classify an interpretation as a fact.
A tidy table may create the impression of more certainty and impartiality than actually exists.
The user may use the exercise to defend a decision already made.
Emotions may be ignored instead of named and separated from facts.
In legal, HR, health, safety, violence or high-risk contexts, proper people, procedures and sources are needed.

Better way to use the answer

A note says, “The client is unhappy and the project is at risk.” Better analysis does not start with a decision. It separates layers. The fact may be: the client has not replied for five days or wrote that they are missing materials. The interpretation: the client is unhappy. The assumption: no reply means project risk. The judgment: the situation is serious. An alternative reading: the client may be waiting for information they did not receive. The missing question: has the client reported a problem, what do they expect and which deadline matters now?

This prompt supports decision support without moving the decision onto the model. The user gets an organized map of the material, not a verdict. It also helps maintain calibrated trust: model output can be a useful map, but it does not replace checking facts, asking people questions or taking responsibility for the next step.

Short rule

First separate what can be pointed to from what it may mean, how you are judging it and what you still do not know. Only then decide which question or action makes sense.

Continue exploring

TopicHuman agency and responsibility
A guide to delegation, oversight and responsibility in AI-assisted systems.
ArticleAre we afraid of AI, or of ourselves?
An essay on causal contribution, moral agency, distributed responsibility and the institutional uses of AI as an alibi.
ConceptHuman agency
The capacity to choose, act and remain answerable for consequential decisions.

Suggested citation

Mamczur, F. (2026, July 3). How to separate facts from interpretations. Prompted Psyche. https://promptedpsyche.com/practice/how-to-separate-facts-from-interpretations/