## 1 What Is AI Hallucination? AI hallucination refers to confidently incorrect, fabricated, or illogical responses generated by the model. ## 2 Strategies to Prevent Hallucination Each method below reduces the chance of misleading outputs. ### 2.1 Use Clear, Specific Prompts Why: Prevents AI from guessing user intent. Example: “List three cited statistics from UK retail trends published in 2023.” ### 2.2 Apply the Verifier Pattern Why: Ensures the output is internally consistent and fact-based. Example: “Check your previous answer and flag any unsupported claims.” ### 2.3 Provide Firm Instructions Why: Inhibits the AI from inventing facts. Example: “Do not guess—say ‘unknown’ if data is missing.” ### 2.4 Request and Validate Citations Why: Identifies fabricated or unverifiable sources. Example: “Provide real citations and then verify each one.” ### 2.5 Use Retrieval-Augmented Generation (RAG) Why: Anchors the AI to known documents. Example: “Based only on this policy PDF, list three operational risks.” (Works best in ChatGPT with file upload or Gemini with source documents.) ### 2.6 Use Tools or Plug-ins Why: Enables external checks or calculations. Example: “Search for the latest inflation figures from the ONS and summarise.” ### 2.7 Ask for Confidence Levels Why: Forces the AI to self-assess uncertainty. Example: “Indicate your confidence in each claim using a 1–5 scale.” ### 2.8 Keep to Known Domains Why: Hallucinations are more common in poorly documented or niche areas. Example: Use prompts tied to well-understood domains like project management. ### 2.9 Add a Fact-Checking Step Why: Identifies and isolates any falsehoods. Example: “Highlight and fact-check all claims in the previous answer.” ### 2.10 Require Uncertainty Flagging Why: Makes grey areas visible to the user. Example: “Label speculative or unverified information