Prompt engineering — the practice of designing inputs to AI systems that produce high-quality, reliable outputs — is fast becoming a core journalism skill. The difference between a well-designed and a poorly-designed prompt is not marginal: it can be the difference between accurate, balanced reporting and confident, biased misinformation.
Five High-Impact Prompt Patterns
1. The Source-First Pattern: Begin every research prompt with "Cite your sources for every factual claim you make." This forces the model to retrieve before generating. 2. The Steelman Pattern: "Present the strongest argument for and against [position] with equal rigour." This combats the model's tendency to reflect majority-view bias. 3. The Devil's Advocate Pattern: After receiving a draft, prompt: "What are the three most likely factual errors in this draft?" AI models are better at evaluating their own errors when asked to than when left unchallenged. 4. The Persona Pattern: "You are a senior editor at AP who applies AP Style and requires every statistic to be sourced. Review this draft and flag all unsourced claims." 5. The Constraint Pattern: "Write a 300-word summary of [topic]. Include exactly three statistics, each with a source. Use no superlatives or emotional language."
What Not to Do
Avoid open-ended prompts on contested topics without constraints ("Write an article about AI regulation"). Avoid prompts that presuppose conclusions ("Write an article showing that AI is dangerous"). Both patterns produce confident, biased output that reads well but requires extensive fact-checking to be publishable.