Prompt Kits | CogBias

Home/Prompt Kits

Prompts

Prompt kits for catching bias without outsourcing your judgment

These are designed to make an AI model slow the structure of the reasoning process. The goal is not to let the model decide for you. The goal is to expose what your current process may be skipping.

Best use

Use these after you have written the live decision, forecast, or case as concretely as possible. Vague prompts produce flattering but low-value outputs.

Important limit

A model can help surface missing comparisons, hidden defaults, or softened alternatives, but it can also confidently mirror your framing. The prompt should therefore widen the lens, not merely request a verdict.

Decision Debias Brief

Use this when a live decision feels urgent and you want the model to slow the structure of the judgment rather than merely justify a preferred option.

Use when: Paste the current decision, the options under consideration, and any real constraints or deadlines.

Anchoring effect Status quo bias Sunk cost effect Overconfidence effect Gambler's fallacy

Open prompt

Analyze the decision below as a debiasing brief rather than as a recommendation memo.

Your tasks:
1. Restate the decision in one sentence.
2. Identify which options are being compared, including the default or do-nothing option.
3. Flag any likely cognitive biases that may be distorting the choice.
4. For each flagged bias, explain exactly how it may be shaping the current judgment.
5. Provide a cleaner forward-looking evaluation of the options that ignores sunk costs and separates evidence from comfort.
6. End with three concrete questions the decision-maker should answer before committing.

Output format:
◉ Decision restatement
◉ Likely biases at work
◉ Clean forward-looking comparison
◉ Questions before commitment

Decision to analyze:
[PASTE DECISION HERE]

Forecast And Postmortem Pair

Use this when you want one prompt that first produces a calibrated forecast and then sets up a cleaner later postmortem.

Use when: Paste the upcoming event, decision, project, or forecast target and include any historical comparison cases if available.

Overconfidence effect Planning fallacy Hindsight bias Outcome bias Clustering illusion

Open prompt

Build a two-part forecasting and postmortem aid for the situation below.

Part A: Forecast now
- Give a probability range or outcome range rather than a single-point answer.
- Name the reference class or outside view.
- List the main assumptions carrying the forecast.
- Run a short premortem: imagine the forecast failed and list why.

Part B: Preserve the record for later learning
- Rewrite the forecast as a timestamped note that could be reviewed after the outcome.
- List three things that should be judged as process quality later rather than judged by the final result alone.
- End with a short reminder against hindsight bias and outcome bias.

Situation:
[PASTE PROJECT, EVENT, OR FORECAST TARGET HERE]

People Judgment Scan

Use this when a person is being evaluated and you want the model to separate behavior, context, impression, and trait inference more carefully.

Use when: Paste the behavior, the context you know, and the judgment you are tempted to make about the person.

Egocentric bias Fundamental attribution error Halo effect Implicit bias Confirmation bias

Open prompt

Analyze the situation below as a people-judgment scan.

Do the following:
1. Describe the observable behavior without interpretation.
2. List plausible situational explanations before any trait explanation.
3. Identify which cognitive biases could be distorting the current evaluation.
4. Separate what is actually evidenced from what is merely inferred.
5. Suggest a fairer next step for gathering information before making a high-confidence judgment.

Use this output structure:
◉ Observable behavior
◉ Situational explanations
◉ Bias risks
◉ What is known vs inferred
◉ Fairer next step

Situation:
[PASTE THE PERSON-JUDGMENT CASE HERE]

Media Narrative Bias Scan

Use this when you want help separating vividness, repetition, and narrative fit from actual representativeness in an article, thread, or speech.

Use when: Paste the relevant excerpt or provide a link and enough context for the model to quote the key passage accurately.

Availability heuristic Negativity bias Survivorship bias Base-rate neglect Apophenia Clustering illusion Frequency illusion

Open prompt

Analyze the passage below for cognitive-bias pressure rather than for partisan agreement or disagreement.

Your tasks:
1. Quote the most vivid or emotionally loaded claims.
2. Identify which cognitive biases the passage may be exploiting or triggering in readers.
3. Explain whether the passage substitutes anecdote, salience, repetition, or winner-story logic for representative evidence.
4. List what denominator, base rate, missing sample, or comparison would be needed to evaluate the claim more responsibly.
5. End with three discussion questions for a reader who wants to stay calibrated.

Passage to analyze:
[PASTE PASSAGE OR LINK CONTEXT HERE]

Groupthink Meeting Scan

Use this when a team discussion seems to be converging fast and you want help separating real agreement from status pressure, premature harmony, or hidden dissent.

Use when: Paste the decision under discussion, who is in the room, the current lean, and any objections that have surfaced so far.

Groupthink Authority bias Social desirability bias False consensus effect

Open prompt

Analyze the situation below as a groupthink and meeting-pressure scan.

Your tasks:
1. Restate the current decision and the room's apparent consensus.
2. Identify which voices, roles, or incentives may be shaping what people are willing to say.
3. Flag any signs of groupthink, authority bias, social desirability pressure, or false-consensus assumptions.
4. List the objections or alternatives that most need airtime before the discussion closes.
5. End with a cleaner meeting protocol for the next 15 minutes of discussion.

Output format:
◉ Apparent consensus
◉ Social pressure risks
◉ Missing objections
◉ Better next-step meeting structure

Situation:
[PASTE THE MEETING CASE HERE]

Framing And Default Audit

Use this when a decision may be getting steered by wording, preselection, loss framing, or the moral comfort of doing nothing.

Use when: Paste the choice, the current wording or interface, and the default option if one exists.

Framing effect Default effect Loss aversion Omission bias

Open prompt

Analyze the decision below as a framing and default audit.

Do the following:
1. Restate the decision in neutral language.
2. Identify the current frame, emotional emphasis, and default path.
3. Explain how framing effect, default effect, loss aversion, or omission bias may be steering judgment.
4. Rewrite the choice in at least two alternative frames, including one that removes the default's privilege.
5. End with a cleaner comparison of the options on their actual merits.

Decision or interface to analyze:
[PASTE THE DECISION, FORM, OR MESSAGE HERE]

Explanation Depth Probe

Use this when someone seems highly confident or fluent and you want to test whether the underlying understanding is genuinely robust.

Use when: Paste the claim, explanation, lesson, or proposed plan that needs to be pressure-tested.

Illusion of explanatory depth Dunning-Kruger effect Overconfidence effect Curse of knowledge

Open prompt

Analyze the material below as an explanation-depth probe rather than as a style critique.

Your tasks:
1. Restate the core explanation or claim in plain language.
2. Identify where the explanation relies on labels, vague transitions, or hidden steps.
3. Flag signs of illusion of explanatory depth, overconfidence, or curse-of-knowledge communication.
4. Generate three challenge questions that would test whether the understanding is robust.
5. Rewrite the explanation so a thoughtful novice could follow the mechanism.

Material to probe:
[PASTE THE EXPLANATION, CLAIM, OR PLAN HERE]

Conflict Interpretation Reset

Use this when a disagreement is starting to harden into a story about bad motives, threat, or tribal hostility.

Use when: Paste the disputed behavior, the surrounding context, and the hostile or character-level interpretation currently on the table.

Hostile attribution bias Naïve realism Ingroup bias Fundamental attribution error

Open prompt

Analyze the situation below as a conflict-interpretation reset.

Do the following:
1. Separate the observable behavior from the current hostile interpretation.
2. List plausible non-hostile or less-loaded explanations that still fit the facts.
3. Identify which biases may be intensifying threat, tribal reading, or motive certainty.
4. Show what additional evidence would be needed before endorsing the harsher interpretation.
5. End with a lower-heat way to continue the conversation or investigation.

Situation:
[PASTE THE CONFLICT CASE HERE]

Misinformation Residue Scan

Use this when a claim has already been corrected or contested, but you suspect the first version is still quietly shaping explanation and uptake.

Use when: Paste the original claim, the correction if one exists, and the current explanation or conversation that still seems influenced by the earlier version.

Continued influence effect Availability cascade Bandwagon effect Confirmation bias False memory Frequency illusion

Open prompt

Analyze the material below as a misinformation-residue scan.

Your tasks:
1. Restate the original false or disputed frame and the later correction separately.
2. Explain what explanatory work the original frame was doing for readers or listeners.
3. Identify signs that the earlier frame is still influencing memory or reasoning after correction.
4. Suggest a replacement explanation that could occupy the space the misinformation used to fill.
5. End with three questions that would help a team or reader stop reasoning with the old frame.

Material to scan:
[PASTE THE CLAIM, CORRECTION, AND CURRENT DISCUSSION HERE]

Choice Architecture Audit Plus

Use this when a decision may be getting steered by the menu, the frame, the decoy, the proxy question, or the push to act visibly.

Use when: Paste the choice, the options, any default or recommended option, and how the options are currently being presented.

Decoy effect Attribute substitution Contrast effect Action bias Automation bias

Open prompt

Analyze the situation below as a deeper choice-architecture audit.

Do the following:
1. Restate the actual decision and the target attribute that should decide it.
2. Identify which easier proxy, comparison trick, system recommendation, or urgency cue may be steering the choice.
3. Explain how the current setup could be manufacturing preference rather than revealing it.
4. Rebuild the decision in a cleaner comparison format that removes the most distorting cue.
5. End with a better decision protocol for the next pass.

Decision setup:
[PASTE THE OPTIONS, PRESENTATION, AND CURRENT LEAN HERE]

Self-Justification Conflict Probe

Use this when a person, team, or institution seems to be editing memory, standards, or narrative to make a contradiction easier to live with.

Use when: Paste the contradictory commitments, the decision or behavior in question, and the justification currently being used to smooth the conflict.

Cognitive dissonance Choice-supportive bias Bias blind spot Self-serving bias Consistency bias Egocentric bias

Open prompt

Analyze the material below as a self-justification and contradiction probe.

Your tasks:
1. Name the commitments, values, or memories that are currently in tension.
2. Show how the current narrative reduces discomfort without necessarily resolving the contradiction honestly.
3. Identify which biases may be editing memory, standards, or self-description.
4. Separate the psychologically comforting story from the most evidence-responsible story.
5. End with three questions that would keep the contradiction visible long enough for a cleaner revision.

Material to probe:
[PASTE THE DECISION, JUSTIFICATION, OR CONFLICT HERE]