ThinkWithPrompt by SynotticThinkWithPrompt

Check an AI-assisted decision for bias

Finds where an AI-assisted decision about people could be unfair, how to test it, and what safeguards to add.

Working with AIFew-ShotAdvancedLegalWriteSummarizeFill-in Template

The prompt

Help me check an AI-assisted decision for bias before we rely on it. The decision: [WHAT is being decided - e.g. "which applicants get a first interview", "which customers get a credit limit increase", "which tickets get priority"] How AI is involved: [e.g. "AI scores resumes", "AI summarizes performance notes for calibration", "AI drafts the recommendation"] Data and criteria used: [INPUTS, fields, prompts, or scoring rules] Who is affected and how much it matters to them: [GROUPS and stakes] Any results so far: [PASTE outcomes or a sample, if available - do not include names] Review: 1. Where bias can enter - training data, proxies for protected characteristics (ZIP code, school, gaps in employment, name, age signals), prompt wording, missing data, and how humans use the AI output (automation bias). 2. Groups that could be disadvantaged and how. 3. Tests to run - e.g. compare outcomes across groups, swap names or demographics in test cases, check error rates, review a sample by hand. Give each test as a concrete step. 4. Safeguards - human review points, criteria changes, appeal routes, what to stop using. 5. What to document to show the decision process was fair. Output as sections with a short risk rating (High / Medium / Low) for each source of bias. Do not conclude that the process is bias-free from a description alone - say what evidence would be needed. Decisions about hiring, pay, credit, housing, and similar areas are regulated in many places; involve HR, legal, or compliance before relying on the result.

Highlighted [brackets] are blanks to replace with your own details.

Fill in the blanks and use this promptMore Working with AI prompts

What to fill in

  • INPUTS, fields, prompts, or scoring rules
  • GROUPS and stakes

How this prompt is built

This prompt follows the Few-Shot structure (Few-shot prompting). A few-shot prompt includes one or more worked examples of the input and the output you want, so the model can copy the pattern instead of guessing it.

Related prompts

Find more prompts like this